13 MVC Sogamoso

Mining Ventilation Control
Dr. Emmanuel WITRANT
Lessons Handout
123345 647893
1 1232345673829A82B35CD3EAF738A9
1234567892A9B2AC8D9E655FA8D9C82A9A9822A9649C9CA2526A936C88D 9
CB259B6C52A946396A3659436B9C98969C98
2 BF7577EA5B38B9523297
6A36598 8CB9C82AD96C52A9 AB2892A9C92BC9A943C!FCA 96B2A8D 9
6C58946396A365D9"C94CC#$932A25CD9%#252 9A96A3655#252 D9"C 9
36#F8AC889C38F89C3463BAC925CBA
3 1BE795B5F237573829A82B35378B587F
&FABCA589649E 8259B6C52AD96C59649C9AC763$98 8CB9F82A9'23644(8 9
578D9 AB29B6C589649C9AC763$D9%2BF526A986473C8
4 259B5FBE792345235E775799
"C9#829C!F26A8964945F29 AB28D9&36B9)F5C39C!F26A89695FBC9B6C58D 9
&A8*9366B89A9655FA986F3C8D9+65FBC,C32A9A9C82B26A9649C 9
3A86396C4422CA8D9"2BC,C5 936-2B26A
5 8A82B35BBF5A25DA9285FBE79
1C3689 AB289649B#C3,52$C9B2AC9763$2A8D9&5F29828.9#F6 A 9463CD 9
%3242C9456789A9463C95FBC8D96A832AC98C9649C9655FA8936425CD 9
EC32C359 AB2892AFC9# 94A89A93F52AC9F#C8
6 232A95B38B958A8742758B5F23234573829A82B3
6A3659649C9CA2526A9AC763$D9/6A52AC396A3659649C945679AC763$92A965 9
B2AC8D92832#FC9 AB2896A36592A9CC97C558.9489E9A92BC,C5 9
6BCA826AD90 #3296A3659649C9C-326A9366B8
7 92A82B35B52797573B58B5F23234573829A82B3
1$36FA96A9723C5C8898CA82A*99A9AC98C32C8D96BBFA226A 9
32CF3CD96C589A95632B8D91AC9AC763$932CF3C89A98C32C8D 9
8C98F 96A95652826A98C32C8
Universidad Pedagogica y Tecnologica de Colombia

Sogamoso, April 2013.
Outline Outline
E. Witrant E. Witrant
An international framework - special thanks to:
Sweden ABB: A.J. Isaksson, M. Strand
- Boliden
- KTH: K.H. Johansson, C. Fischione
Italy U of L’Aquila: A. D’Innocenzo, M.D. Di Benedetto, F.
Mining Ventilation Control Santucci, E. Serra, S. Tennina and U. Tiberi
France LSS: S. Olaru, G. Sandou, S. Niculescu
- GIPSA-lab: N. Marchand, M. Alamir, F. Castillo
Outline
India IIT Dehli: M. Khare, S. Chinthala
Emmanuel WITRANT USA U California at San Diego: M. Krstić
emmanuel.witrant@ujf-grenoble.fr Colombia UPTC: J.M. Salamanca
Universidad Pedagógica y Tecnológica de Colombia, + on flow systems
Sogamoso, April 4, 2013 France LGGE: P. Martinerie
- Renault: V. Talon
Switzerland CERN: B. Bradu, P. Gayet
Outline Outline
E. Witrant
Mining E. Witrant
Mining ventilation control
Components:
Actual automation in Sweden
• Blasting and drilling
• Ventilation control = worst
• Transport: trucks or hoist
case ventilation design
• Ore crushing • i.e. tunnels diameter /
• Ventilation: 50% of energy fans power depending on
consumption number of trucks
• Mining = 4% (US) - 6% • Operation at max power
(South Africa) of industrial when extracting the ore
electrical consumption
• No continuous air quality
monitoring (scheduled),
no Wireless Sensor
Network (WSN)
Outline Outline
Visit at Garpenberg mine (operated by Bolidenr ), Sweden
Potential wireless control
architecture
Outline Outline
E. Witrant
Control architecture E. Witrant
Course goal
Provide technical skills and research insights for mining

ventilation control.
Main topics:
• Objectives:
• Control air quality (O2 , NOx and CO ) in extraction rooms • Mining ventilation automation principles and design
• turbine and heater provide airflow pressure to fans • Mathematical modeling of the mine flow network, of
• fans ensure air quality in extraction rooms
aerodynamics in ducts and of the ore extraction rooms
• Safety through wireless networking for personal
communication and localization • Principles of feedback control and advanced strategies of
• Automation/design constraints: feedback design dedicated to mine ventilation control
• Physical interconnections, actuators limitations and
networking capabilities
• Sensing capabilities: chemical, pressure and temperature
Outline Outline
E. Witrant
Class overview E. Witrant
Les. Topic
1 Mining Ventilation Fundamentals (2h)
Airflows in mines; Pollutants; Design and sizing of the ventilation process;
Mathematical modeling for control from the 60s to the 90s
2 Some feedback control principles (4h) Les. Topic
Control system design; Time and frequency domains; Models for control; 6 Principals control strategies to mining ventilation (1h30)
The feedback principle; Stability and controllability; Robustness vs. perfor- For the ventilation network; for deep wells: fast MPC and time-delay com-
mance. pensation; Hybrid control of the extraction rooms.
3 Model of mine ventilation network system (3h) 7 Application of wireless sensors to mining ventilation (30 mn)
Fundamentals of Physical modeling; Model of the network system using Background on wireless sensing, c and advanced services; Communication
Kirchhoff’s laws; Non-minimal and minimal models of the network; Simula- architecture; Models and algorithms; Advanced network architectures and
tion softwares. services; Case study on localisation services
4 Air flow modeling in deep wells (2h)
Fluid dynamics; From Euler to lumped models; Fans, rooms and pollutant
sources; Volume-averaging and estimation of the transport coeff.; Time-
delay approx.
5 Extraction rooms air quality model (2h)
Gas dynamics in rooms; Fluid statics and buoyancy; Stratified flows and
forced plumes; Constrained shape of the pollutants profile; Peripheral dy-
namics.
Outline Outline
Class website
Reference textbooks
• Hartman HL, Mutmansky JM, Ramani RV, Wang YJ. Mine
Ventilation and Air Conditioning (3rd edn). Wiley: New York,
1997.
• Anderson, J.: Fundamentals of Aerodynamics, McGraw-Hill • Go to:
Companies, 1991. http://physique-eea.ujf-grenoble.fr/intra/Formations/M2/EEATS/PSPI/UEs/courses_CoMVC.php
• C. Hirsch, Numerical Computation of Internal & External Flows:

the Fundamentals of Computational Fluid Dynamics, 2nd ed. or Google “MiSCIT” then go to “Courses”, “Advanced
Butterworth-Heinemann (Elsevier), 2007. control theory” and “Mine Ventilation Control”
• S. Skogestad and I. Postlethwaite, Multivariable Feedback • at the bottom of the page, click “Restricted access area”
Control: Analysis and Design, 2nd Ed., Wiley, 2007. and enter with:
http://www.nt.ntnu.no/users/skoge/book/ps/book1-3.pdf • login: MineVentCont
• K.J. Åström and B. Wittenmark, Computer-Controlled Systems: • password: sogamoso
Theory and Design, 3rd Ed., Prentice Hall, 1997.
• H. Khalil, Nonlinear systems, Prentice-Hall, 2002
Mining Mining
Ventilation Ventilation
Fundamentals Fundamentals
Outline
Airflows in Airflows in
mines mines
Historical perspectives Historical perspectives
Control processes Control processes
Engineering controls Engineering controls
principles principles
1 Airflows in mines
Pollutants
Diesel Aerosols and
Gases in
Mining Ventilation Control Pollutants
Diesel Aerosols and
Gases in
Underground Mines Underground Mines 2 Pollutants
Diesel Particulate Diesel Particulate
Matter Matter
Gases Gases
Emission Sources Lesson 1: Mining Ventilation Fundamentals Emission Sources
3 Design and sizing of the ventilation process
Design and Design and
sizing of sizing of
ventilation ventilation
Gas laws
Emmanuel WITRANT Gas laws 4 Mathematical modeling for control from the 60s to the 90s
Energy changes
Head losses and mine
emmanuel.witrant@ujf-grenoble.fr Energy changes
heads heads
Head gradients Head gradients
In mine openings Universidad Pedagógica y Tecnológica de Colombia, In mine openings
Head losses Head losses
Air power Sogamoso, April 1, 2013 Air power
Mathematical Mathematical
modeling for modeling for
control control
Conclusions Conclusions
Mining Mining
Fundamentals
Airflows in mines [Hartman et al. 1997] Fundamentals
mines mines
Historical perspectives
Control processes Environmental control of the mine atmosphere Control processes
• Paleolithic: miners exposed to oxygen defficiency, toxic
• Artificial atmosphere needed to sustain miners: need to be gases, harmful dusts and debilitating heat
Pollutants Pollutants
Diesel Aerosols and
Gases in controlled! Diesel Aerosols and
Gases in
• 1st millenium BCE: air through multiple openings and
Underground Mines Underground Mines
Diesel Particulate
• Most versatile control tool in mining engineering Diesel Particulate circuits + fire-induced currents
Matter Matter
Gases
Emission Sources • Mining ventilation = fluid dynamics applied to airflows in
Gases
Emission Sources
• Middle Ages: ventilation = mining art, involving “ventilation
Design and
openings and tunnels Design and machines”: deflectors, bellows, fans
sizing of sizing of
ventilation
• Need to define amount and direction of air throughout the
ventilation • depth constraint = high rock pressure and temperature +
Gas laws Gas laws
Energy changes
mine: limits in quantity, quality and temperature-humidity Energy changes deterioration of atmosphere ⇒ ventilation became the
Head losses and mine Head losses and mine
heads heads most important branch of deep mining
Head gradients
In mine openings
• Essential for safety, as well as worker productivity and job Head gradients
In mine openings
Head losses
Air power
satisfaction Head losses
Air power
control control
Mining Mining
Ventilation E.g. Agricola 1556: deflectors, bellows and fans Ventilation
E. Witrant E. Witrant Control processes for total air conditioning
Airflows in Airflows in 1 Quality control - purifying air and removing contaminants
mines mines
Historical perspectives Historical perspectives such as:
• gases - vapors and gaseous matter + radiation
• dusts - particulate matter
Diesel Aerosols and
Gases in
Diesel Aerosols and
Gases in
2 Quantity control - regulating the magnitude and direction
Underground Mines
Diesel Particulate
Underground Mines
Diesel Particulate
of air flow through:
Matter Matter
Gases Gases
• ventilation (primary)
Emission Sources Emission Sources
• auxiliary or face ventilation
Design and Design and • local exhaust
sizing of sizing of
Gas laws Gas laws
3 Temperature-humidity control - controlling latent and
Energy changes Energy changes sensible heat by
heads
Head gradients
heads
Head gradients
• cooling
In mine openings In mine openings • heating
Air power Air power • humidification
Mathematical Mathematical • dehumidification
control control
Mining Mining
Pollutants [Bugarski et al., 2011]
mines Engineering controls principles mines
Diesel Aerosols and Gases in Underground Mines
Engineering controls
principles
1 prevention or avoidance Engineering controls
principles • Diesel engines = major source of submicron aerosols, CO,
Pollutants 2 removal or elimination Pollutants
CO2, NOX, SO2 and hydrocarbons (HC) in underground
Diesel Aerosols and Diesel Aerosols and
Gases in
Underground Mines 3 suppression or absorption Gases in
Underground Mines coal and metal/nonmetal mines.
Matter
Gases
4 containment or isolation Matter
Gases
⇒ Challenge to control workers’ exposure and need to
Emission Sources Emission Sources establish a comprehensive program based on a
5 dilution or reduction
Design and Design and multifaceted and integrated approach:
sizing of sizing of
ventilation + Medical/legal control principles ventilation • Curtail emissions of the diesel particulate matter (DPM)
Gas laws Gas laws
Energy changes Distinction between comfort (for humans) and product (for Energy changes
and toxic gases at the source;
heads plants) air conditioning.
heads
• Control pollutants after they are released in the
Head gradients
In mine openings
Head gradients
In mine openings
underground mine environment;
Head losses Head losses • Use administrative controls to reduce exposures of
Air power Air power
underground miners to pollutants.
control control
Mining Mining
Diesel Aerosols and Gases in Underground Mines (3)
mines mines
• Such program should be dynamic and based on
Diesel Aerosols and Gases in Underground Mines (2) Historical perspectives
Control processes
Control processes
information gathered through surveillance:
• Involve key departments of mining companies: health and • of parameters for planning, execution, and coordination of
Diesel Aerosols and safety, engine/vehicle/exhaust aftertreatment Diesel Aerosols and
the program (e.g., size of the diesel-powered fleet, role of
Gases in
Underground Mines
maintenance, mine ventilation, and production, + those
Gases in
Underground Mines diesel-powered equipment in the mining process, type of
engine emissions, contribution of diesel-powered
Matter
responsible for acquiring vehicles, engines, exhaust Matter
Gases Gases
equipment to exposure of underground miners to DPM and
Emission Sources
aftertreatment systems, fuel, and lubricating oil. Emission Sources
criteria gases, quality of diesel fuel and lubricating oil, and

sizing of
ventilation
⇒ A program coordinator is crucial to the success of diesel sizing of
ventilation
ventilation supply and demand)
Gas laws control programs Gas laws • identify and quantify the extent of the problem, identify and
Energy changes
Energy changes
evaluate potential solutions, and identify and establish a
heads
Head gradients
heads
Head gradients
hierarchy of potential solutions
In mine openings In mine openings
Head losses Head losses ⇒ Key role of Information Technologies and System Analysis!
Air power Air power
control control
Mining Mining
Ventilation
Diesel Particulate Matter (DPM) Ventilation DPM: Elemental Carbon (EC)
E. Witrant E. Witrant • Combustion = mixture of fuel (hydrocarbon, Cx Hy ) vs.
• DPM any material being emitted from a diesel engine
intake air:
Airflows in
mines
that can be collected on a filter through cooled and diluted Airflows in
mines
exhaust. Historical perspectives y
Control processes Control processes Cx Hy + (x + y /2 + 2)O2 + N2 → xCO2 + H2 O + 2NO
principles
principles
2
Pollutants
• Includes four byproducts of Pollutants
Diesel Aerosols and diesel combustion: elemental Diesel Aerosols and
Lean (efficient) or rich (lack O2 )
Gases in Gases in
Underground Mines
Diesel Particulate
carbon (EC), organic carbon Underground Mines
Diesel Particulate
• If hot enough in rich regions, fuel burns without O2 ,
Matter
Gases
(OC), ash, and sulfuric Matter
Gases
creating charred remains, or solid carbon soot EC
Emission Sources compounds Emission Sources
• Emitted from the engine exhaust as solid particulate
sizing of • Combine to form DPM sizing of matter, forming the core of a typical diesel-particle
Gas laws
aerosols → Gas laws agglomerate
Energy changes Energy changes
heads
• In general, 1 oom < other Head losses and mine
heads
• Driven by temp, residence time and availability of oxidants
Head gradients respirable dust aerosols in Head gradients
• Reduced:
Head losses underground mines (< 1µ m) Head losses
Air power Air power • at the source by increasing the surface area contact of fuel
Mathematical ⇒ Not removed by gravitational Mathematical and air → in-cylinder controls
control
settling and deposited in the control
• by capturing these particles within the exhaust system
Conclusions
human respiratory tract! Conclusions
using diesel particulate filters
Mining Mining
Ventilation
Fundamentals
DPM: Organic Carbon (OC) Ventilation
Fundamentals
• Forms when hydrocarbons (in fuel and lubricating oil) are
Airflows in consumed but not fully oxidized during the combustion Airflows in
mines
process. Sources = mines DPM: Ash
Control processes
• fuel in overly lean regions (not enough fuel) Control processes
• Come from additives (detergents, dispersants, etc.) in fuel
principles
• fuel that is post-injected into the chamber too late principles
and lubricating oil composed of metallic elements. When
Diesel Aerosols and
• lubrication oil that is scraped from cylinder walls or Diesel Aerosols and consumed, they form inorganic solids = ash
Gases in
Underground Mines introduced into the combustion chamber from other Gases in
Underground Mines
Diesel Particulate
sources Diesel Particulate • Cannot oxidize in secondary reactions with aftertreatment
Matter Matter
Gases Gases devices and may accumulate within the exhaust system
Emission Sources • Temperatures may be high enough to vaporize the Cx Hy , Emission Sources
and cause maintenance issues over time.
Design and
sizing of
but not to convert them into CO2 and H2 O. Design and
sizing of
ventilation ventilation • Reduction accomplished by reducing the metallic fraction
• Partially composed of volatile material; react and change
Gas laws Gas laws
of the fuel and oil, and by lowering the amount of oil
Energy changes
in both composition and phase during emission. Energy changes
heads
heads consumed during the combustion process.
Head gradients • Controlled: Head gradients
Head losses • at the source by reducing oil consumption, improving fuel Head losses
Air power Air power
and oil formulations, and improving fuel injection design
modeling for and timing modeling for
control • at the exhaust by diesel oxidation catalysts (DOCs) control
Mining Mining
Ventilation DPM: Sulfuric Compounds Ventilation
E. Witrant • Form when sulfur in the fuel and lubrication oil oxidizes E. Witrant
Airflows in during the combustion process Airflows in

Gases: Nitrogen Oxides (NO and NO2 )
mines mines
Historical perspectives • Gaseous emission that can damage or deactivate Historical perspectives • N2 + O2 + HC → gaseous NOx emissions, or oxides of
Engineering controls expensive exhaust catalysts Engineering controls nitrogen (NO and NO2 )
Pollutants • React with other compounds in the exhaust and form solid Pollutants • Rate of formation exponentially related to the temperature
Gases in
Underground Mines
sulfates, contributing to overall DPM emissions Gases in
Underground Mines
of combustion → in-cylinder controls to lower the peak
Diesel Particulate
Matter • Controlled by the transition toward ultralow sulfur diesel Diesel Particulate
Matter temperatures = exhaust gas recirculation (EGR) control
Gases Gases
Emission Sources fuels (ULSDF) and low-sulfur content lubricants (e.g., CJ-4 Emission Sources • Secondary control through aftertreatment:
Design and oil, the newest API class) Design and • such as lean NOx catalysts (LNCs)
sizing of sizing of
ventilation ventilation • selective catalyst reduction (SCR)
Gas laws Gas laws
Energy changes Total Carbon (TC) and EC:TC Ratio Energy changes • NOx /DPM tradeoff: lowering NOx emissions through
heads heads
in-cylinder techniques typically results in an increase in
Head gradients
• TC = EC + OC: sum of the Elemental C and Organic C Head gradients
DPM, and conversely
Head losses
Air power
fractions of DPM. Head losses
Air power
Mathematical • EC:TC Ratio: fraction of EC in TC. Mathematical

control • Depends on engine operating conditions, engine type, fuel control
Conclusions
type, and a number of other parameters Conclusions
Mining Mining
Ventilation Ventilation Gases: Gas-Phase Hydrocarbons (HC)
• Typically referred to as volatile (VOC) and semivolatile
Airflows in
Gases: Carbon Monoxide (CO) Airflows in organic compounds (SVOC): complex mixture of many
mines mines
Historical perspectives Historical perspectives chemical species, e.g. highly toxic polycyclic aromatic
• Results from a non-ideal combustion: incomplete oxidation Engineering controls hydrocarbons (PAHs)
Pollutants
of carbon in the fuel to carbon dioxide, most often from a Pollutants • Control of gas-phase OC emissions at the source is the
Diesel Aerosols and
Gases in
lack of available oxygen or low gas temperatures. Diesel Aerosols and
Gases in same as nongaseous OC control
Diesel Particulate
Matter
• Typically minimal but extremely high toxicity motivated Diesel Particulate
Matter • DOCs within the exhaust system are often used as a
Gases
Emission Sources
stringent regulation Gases
Emission Sources secondary control
Design and • Reduced by improving the overall combustion efficiency by Design and
sizing of sizing of
ventilation limiting any fuel-rich conditions within the cylinder and ventilation
Gases: Sulfur Dioxide (SO2 )
Gas laws Gas laws
Energy changes using diesel oxidation catalysts (DOCs) within the exhaust Energy changes
heads system (CO → CO2 in secondary reactions) Head losses and mine
heads • Forms when sulfur in the fuel and lubrication oil oxidizes
In mine openings In mine openings • Can damage or deactivate expensive exhaust catalysts in
Air power Air power contemporary diesel engines
modeling for modeling for • Controlled by the transition toward ultralow sulfur diesel
control control
fuels (ULSDF) and low-sulfur content lubricants (CJ-4 oil)
Mining Mining
Fundamentals
Emission Sources [Avanti 2011] Fundamentals
Design and sizing of the
ventilation process [Hartman et al. 1997]
mines mines
Control processes
Control processes
Airflow through mine openings and ducts
Pollutants Pollutants • Quantity control for air movement, direction and magnitude
Gases in
Underground Mines
Gases in
Underground Mines • Ventilation:
Matter Matter • prime control of mine air conditioning
Gases Gases
Emission Sources Emission Sources • has to supply enough air for human and products needs
Design and Design and • modest need for human life (≈ 0.01 m3 /s/person) but need
sizing of sizing of
ventilation ventilation to remove contaminents (gas/heat/moisture): 0.1 to 1
Gas laws Gas laws m3 /s/person ⇒ 10-20 tons of air per ton of mined mineral
heads
heads • Need to understand airflow to design long and tourtuous
In mine openings In mine openings paths for mine ventilation
Air power Air power • While a crude approximation, air is typically considered as
Mathematical
modeling for
Mathematical
modeling for
incompressible for mine ventilation design
control control
Mining Mining
Ventilation
Fundamentals
Ventilation
Fundamentals
Energy changes in fluid flow
Gas laws: behavior of air General energy equation (steady-state):
Defining p = pressure, v = specific volume, w = specific weight, P
• Total energy = internal + static + kinetic + potential + heat
mines
T = absolute temperature, H = head: mines
• Total energy1 = (total energy)2 + (flow energy losses)1→2
Control processes • Boyle’s law: p1 /v1 = p2 /v2 at T constant Control processes
V12 V22
principles
• Charle’s law: v1 /v2 = T1 /T2 at p constant and principles p1 p2
+ + Z1 = + + Z2 + Hl
Pollutants
p1 /p2 = T1 /T2 at v constant
Pollutants w 2g w 2g
Gases in
Underground Mines p1 v1 p2 v2 Gases in
Underground Mines
Diesel Particulate • General gas law: = Diesel Particulate where V = velocity, and the energies: p /w static, V 2 /2g kinetic
Matter
Gases
T1 T2 Matter
Gases (velocity), Z potential, Hl flow losses (Bernoulli, for all fluid flow
P
Emission Sources • Dalton’s law: in a gas mixture, total p = partial p of Emission Sources
process and reduced here to the incompressible case)
Design and
sizing of
individual gases, barometric p = dry air p + water vapor p Design and
sizing of • Each term is a specific energy (Pa) = measure of fluid head,
q q
w 1 termed “head”:
sg , where sg is
Gas laws • Graham’s law: Diffusion rate ∝ wg ∝ Gas laws
heads the specific gravity of the gas Head losses and mine
heads Ht1 = Hs1 + Hv1 + Hz1 = Hs2 + Hv2 + Hz2 + Hl = Ht2 + Hl
In mine openings • Effect of altitude: w2 /w1 = e −Z /RT , Z = elevation above In mine openings
Head losses Head losses • Provide an expression encompassing all flow variables between
Air power sea level Air power
any two points in the ventilation system.
modeling for • Pressure/head relationship: p = w1 H1 = w2 H2 modeling for • Simplified (no Z ) if all static-head measurements/calculations
control control
are made on a gage-pressure basis in ref. to atmospheric p
Mining Mining
Ventilation Head losses and mine heads Ventilation Head gradients
E. Witrant • Head losses in fluid flow: E. Witrant • 3 distinct gradients: elevation, static + elevation, total
• flow occurs from pressure difference and part of the • General rules:
mines provided energy is dissipated through losses mines
1 Total head = 0 at inlet, but = Hv (> 0) at discharge
• Hl = Hf + Hx : friction (ducts of constant area) and shock Historical perspectives
2 Static head always < 0, = Hv at inlet but 0 at discharge
(in turns and restrictions + inlet/discharge, splits/junctions, Engineering controls
3 Total head at any point plotted first, then Hs = Ht − Hv
Pollutants obstructions) Pollutants
Diesel Aerosols and • compensating energy losses from the static head, Diesel Aerosols and • Blower system:
Gases in Gases in
Underground Mines
Diesel Particulate
• conversion static ↔ velocity heads (e.g. at area changes) Underground Mines
Diesel Particulate
• located at the inlet and raises the head above
Matter
Gases
accompanied by shock losses Matter
Gases
atmospheric, e.g.
• Overall or Mine heads
sizing of • Def.: cumulative energy consumption of fans and other sizing of
Gas laws
pressure systems Gas laws
Energy changes
= difference in head (from Bernoulli) necessary to move the Energy changes
heads
Head gradients
desired air quantity heads
Head gradients
• Mine static head: mine Hs = Hl = (Hf + Hx )
P P
Air power
• Mine velocity head: at each change of duct area or Air power
Mathematical number, mine Hv = V 2 /2g, not cumulative but appears as Mathematical

a loss of kinetic energy into atmosphere
control control • plot by starting from discharge (Hs = 0) toward inlet
• Mine total head: mine Ht = mine Hs + mine Hv
Conclusions Conclusions • Ht = Hs + Hv at any point
Mining Mining
Ventilation Head gradients: Exhaust systems Ventilation
E. Witrant E. Witrant Head gradients: Booster systems
mines mines
Gases in Gases in
Matter Matter
Gases Gases
Design and • Result when the energy source is located at the exhaust Design and
sizing of sizing of
ventilation • Similar to blowers, except that the starting point = intake ventilation
• Energy source at some point between inlet and discharge
Gas laws Gas laws
Energy changes
• Below atmos. datum line since losses are negative and Energy changes • Hs = Hsi + Hsd , ≈ same as blowers and exhausts
heads
Head gradients determined on a gage basis (succion), but same start/end
heads
Head gradients
• Plot from both ends and move toward the fan
Head losses of Hs and Ht Head losses • Hybrid system between blowers and exhausts, with
Air power Air power
• Ht = Hs + Hv at any point still holds, and mine heads are shocks at both ends
modeling for
control
always positive (not exactly the same as a blower due to modeling for
control
Conclusions
the shock at the discharge) Conclusions
Mining Mining
Effect of state of flow on velocity distribution
Ventilation
Fundamentals
State of airflow in mine openings Ventilation
Fundamentals
• Distinct states of fluid flow: laminar, intermediate and
mines
turbulent mines
Control processes • Boundaries established from the Reynolds number Re Control processes
principles (laminar up to 2000 and turbulent above 4000): principles
Diesel Aerosols and
ρDV DV Diesel Aerosols and
Gases in
Underground Mines Re = = = 67280 DV for air at normal T (SI units) Gases in
Underground Mines
• Vmax at the conduit center, determinable as a function of
Diesel Particulate µ ν Diesel Particulate
Matter
Gases
Matter
Gases
Re (supposing circular cross section)
Emission Sources
where ρ = fluid mass density, ν = kinematic viscosity, µ = Emission Sources

sizing of absolute viscosity, D = conduit diameter, V = velocity sizing of
Gas laws • Critical velocity for Re = 4000: Vc ≈ 0.06/D Gas laws
heads
• Need turbulent flow at openings for the dispersion and Head losses and mine
heads
Head gradients
In mine openings
removal of contaminents, which typically occurs due to Head gradients
In mine openings
Head losses
Air power
large exhausts (e.g. for D = 0.9 m, Vc = 0.07 m/s) Head losses
Air power
Mathematical • Laminar flow in leakage through doors and stoppings in Mathematical

control airways / exhaust through caves or filled areas control • Average value V ≈ 0.8Vmax for Re > 10000 (typical in
Conclusions Conclusions mines)
Mining Mining
Fundamentals Calculation of head losses Fundamentals
• Velocity head = kinetic energy to be supplied to maintain
Airflows in
the flow; lost at discharge: Airflows in Calculation of head losses (2)
mines mines
ρV 2 • Shock loss: ≈ 10 − 30% of head loss, important in
P
V2 wV 2
Control processes Hv = 2g
m of fluid = 2
= 2g
Pa Control processes
principles
principles
major airways or in short length with many bends or area
• Friction loss: ≈ 70 − 90% of
P
Pollutants
head loss, >> shock losses Pollutants changes, ∝ V 2 or Hv , computed:
Diesel Aerosols and
Gases in
• loss in static pressure from drag, resistance of the walls Diesel Aerosols and
Gases in • directly as Hx = XHv , where X = friction loss factor
Underground Mines
and internal friction Underground Mines
wRh X
Diesel Particulate
Matter L V2
Diesel Particulate
Matter • from equivalent length Le = m
Gases • Darcy-Weisbach: Hl = f where f is the friction Gases 2gK
Emission Sources D 2g Emission Sources
• Combined head losses and mine heads:
Design and coefficient (depends on Re but often considered constant Design and
sizing of sizing of
ventilation
in mines) and L the length. ventilation KO (L + Le )Q 2
Gas laws • Atkinson: defining the hydraulic radius Rh A /O = D /4 Gas laws Hl = Hf + Hx =
Energy changes
KOLV 2
Energy changes A3
heads for circular conduit, Hf = where O is the perimeter heads
Head gradients A Head gradients
where Q = AV the air quantity
In mine openings and K an empirical friction factor In mine openings
Air power • Determination of airways friction factor: from pressure Air power
Mathematical gradient difference at a given flow velocity, first Mathematical

control
approximation from tables or graphs control
Mining Mining
Fundamentals
Air power
Fundamentals
Mathematical modeling for control
from the 60s to the 90s
Airflows in • Required to overcome energy losses in an airstream, Airflows in
mines mines
HQ • Decentralized control (high/low) actions: control and
Control processes
Pa = pQ = kW Historical perspectives
Control processes
optimization refer to preliminary design of the global
Engineering controls 1000 Engineering controls
principles
• Note: Pa ∝ Q 3
principles
system and automation devices.
• Mathematical modeling:
Gases in
Underground Mines
Gases in
Underground Mines 1968 steady-state compartmental model for flow networks with
Diesel Particulate
Matter
Compressibility effects Diesel Particulate
Matter complex topology [S. Tolmachev & E. Fainshtein];
Gases Gases
1973 experimental determination of turbulent diffusion
Emission Sources
• Start at relatively low pressure and induce, e.g. a 5 % Emission Sources

coefficients [F. Klebanov & G. Martynyuk];
sizing of difference for ∆p = 38 mm Hg, H = 5 kPa, ∆Z = 430 m sizing of 90’s first use of Navier-Stokes equations, with simplified
Gas laws • At high pressure (> 5.0 kPa), use instead of Atkinson: Gas laws chamber-like [G. Kalabin et al.] and general mine aerology
models [N. Petrov et al.];
heads
Kc Q 2 L heads
1994 problems of nonlinearity and nonstationary behavior, high
Head gradients
p12 − p22 = Head gradients
dimensionality and numerical issues [N. Petrov];

D5
Air power Air power 2001 short and long term planning of ventilation requirements
Mathematical • Use for long ventilation-pipe and deep shafts Mathematical [E. Widzyk-Capehart & B. Watson].
control control • Today’s energy constraints motivate optimized real-time
Conclusions Conclusions control and new dynamical models.
Mining Mining
Fundamentals
Conclusions Fundamentals
Reference
mines mines
Control processes • Ventilation control is of prime interest for regulating the Control processes
Engineering controls Engineering controls 1 A. D. Bugarski, S. J. Janisko, E. G. Cauda, J. D. Noll, and S. E.
principles mine environment principles
Mischler, “Diesel Aerosols and Gases in Underground Mines: Guide to
Diesel Aerosols and
• Diesel engines are a main source of pollutants, need for Diesel Aerosols and
Exposure Assessment and Control”, REPORT OF
Gases in
Underground Mines combined emissions control in engines and removal of
Gases in
Underground Mines
INVESTIGATIONS/2011 RI 9687, Department of Health and Human
Diesel Particulate Diesel Particulate Services, Pittsburgh, PA, Spokane, WA, October 2011.
Matter
Gases
confined mine atmosphere Matter
Gases http://www.cdc.gov/niosh/mining/UserFiles/works/pdfs/2012-101.pdf
• Simple calculations are available for air conducts sizing 2 AVANTI Mining Inc., KITSAULT MINE PROJECT ENVIRONMENTAL
sizing of and steady-state operation sizing of ASSESSMENT: APPENDIX 6.2-C Atmospheric
Environment - Emission Sources and Air Quality Modelling, amec, 2011.
Gas laws • First steps on mathematical modeling of the dynamics for Gas laws
Energy changes Energy changes http://www.ceaa-acee.gc.ca/050/documents/55939/55939E.pdf
heads
control Head losses and mine
heads 3 Hartman HL, Mutmansky JM, Ramani RV, Wang YJ. Mine Ventilation
In mine openings • Large potential for improvement using Information In mine openings and Air Conditioning (3rd edn), Ch. 1 & 5. Wiley: New York, 1997.
Air power Technologies and Automatic Control methods! Air power
control control
Feedback Feedback
control control
Outline
Control system Control system

design design
Control problem Control problem
Time and Time and 1 The process of control system design

frequency frequency
domains
Scaling
Linear models
Mining Ventilation Control domains
Scaling
Linear models 2 Modeling dynamics in the time and frequency domains
Sampled Signals Sampled Signals
The feedback The feedback

principle principle
Transfer functions Lesson 2: Some feedback control principles Transfer functions
3 The feedback principle
Feedforward Feedforward
Why feedback? Why feedback?
Stability and Stability and 4 Stability and controllability

controllability Emmanuel WITRANT controllability
Controllability
Nonlinear systems
emmanuel.witrant@ujf-grenoble.fr Controllability
Nonlinear systems
NMPC NMPC 5 The robustness versus performance dilemna
Robustness vs. Universidad Pedagógica y Tecnológica de Colombia, Robustness vs.
performance performance
Time domain Sogamoso, April 2, 2013 Time domain
Gain & phase margins Gain & phase margins
Maximum peak Maximum peak
ωB , ωC ωB , ωC
Feedback Feedback
control control
principles
The process of control system principles
design
design design Example: Under Floor Air Distribution control
An integrated approach:
Time and Time and
frequency frequency
domains 1 Study the system: control objectives, physical model, domains
Scaling Scaling
Linear models scaled simplified model, main properties Linear models
The feedback 2 Automation design: measurements & controlled outputs, The feedback
principle principle
Transfer functions
sensors/actuators choice and location, controller Transfer functions
Feedforward
Why feedback?
architecture Feedforward
Why feedback?
Stability and 3 Control: performance specifications & actuators Stability and

controllability controllability
Controllability constraints → controller design Controllability
Nonlinear systems Nonlinear systems
NMPC
4 Simulation: on computer or pilot plant, NMPC
Robustness vs. Robustness vs.

performance model-automation-control validation performance
Time domain Time domain
Gain & phase margins 5 Implementation: choose hardware and software for Gain & phase margins
ωB , ωC controller, tests & validation, final tuning ωB , ωC
Feedback Feedback
control UJF experiment control

design design 1. Study the system
Time and
frequency
Time and
frequency
• control objectives: regulate rooms temperatures
domains
Scaling
domains
Scaling
independently, compensate doors perturbation, minimize
Linear models
Sampled Signals
Linear models
Sampled Signals
global energy consumption
The feedback The feedback • physical model: from thermodynamics
principle principle
Transfer functions Transfer functions • scaled simplified model: incompressible flow etc.,
Why feedback? Why feedback? normalized temperature, 0-D approximation
Stability and Stability and
controllability controllability • main properties: interconnected system, discrete events,
Controllability Controllability
some nonlinear dynamics
NMPC NMPC

ωB , ωC ωB , ωC
Feedback Feedback
control control

design design
Time and 2. Automation design Time and

3. Control
frequency frequency
domains domains • performance specifications: desired temperature setpoint,
Scaling • sensors: distributed temperature measurements in each Scaling
Linear models Linear models compensate external perturbations “sufficiently quickly”
Sampled Signals room, wireless Sampled Signals

• constraints: diffusers mass flow rate, communication
principle
• actuators: diffusers for UF airflow regulation principle
Transfer functions Transfer functions
capabilities
Feedforward • controller architecture: local feedback loops embedded on Feedforward
• controller design: robust design with limited inputs and
diffusers with wireless data acquisition, global supervision
Stability and Stability and bandwidth limitations, using linear model
NMPC NMPC

ωB , ωC ωB , ωC
Feedback Feedback
control control
Control system 4. Simulation Control system

3
design design Room 1
Control problem Control problem Room 2
Room 3
Time and Time and 2.5
Room 4
Tracking error |Tdi−Ti| (°C)

frequency frequency
domains domains
Scaling Scaling 2
Linear models Linear models
1.5
principle principle
1
Why feedback? • on computer with rooms, WSN (wireless sensor network) Why feedback?
Stability and Stability and 0.5

controllability
and control blocks controllability
Nonlinear systems • model-automation-control validation: check performance Nonlinear systems 0
NMPC NMPC
0 50 100 150 200
and contraints on nonlinear physical model Time (s)
ωB , ωC ωB , ωC
Feedback Feedback
control
5. Implementation control
The control problem

design design
Time and Time and Make output y behave in the desired way by manipulating input
frequency frequency
domains domains u according to a linear time invariant - LTI - model (disturbance
Scaling
Linear models
Scaling
Linear models
d , models G Gd )
y = Gu + Gd d
principle principle
Why feedback? Why feedback? • Regulator problem: counteract d
controllability controllability • Servo problem: track reference r
Controllability
Nonlinear systems
• hardware: temperature sensors, 2.4 GHz ZigBee motes Controllability
Nonlinear systems ⇒ design controller K to minimize control error e = y − r
NMPC
and embedded controller, active diffusers NMPC
• Uncertain G , Gd : need for robustness

performance • software: IEEE 802.15.4 protocols, embedded C algo. performance
Gain & phase margins
Maximum peak
• tests & validation: small scale experiment Gain & phase margins
Maximum peak
ωB , ωC ωB , ωC
• final tuning: performance/input/perturbation weights
Feedback Feedback
control control
Modeling dynamics in the time
and frequency domains
design design
Control problem Definitions Control problem
Time and Time and

frequency NS Nominal Stability: stable w/o model uncertainty frequency
domains domains
Scaling
Linear models
NP Nominal Performance: perf. specs matched w/o model Scaling
Linear models Transfer functions (TF)
Sampled Signals
uncertainty Sampled Signals
The feedback The feedback • Insights from frequency-dependent plot

principle RS Robust Stability: stable ∀ perturbed plants about the principle
Transfer functions
Feedforward nominal model up to the worst case model uncertainty
Transfer functions
Feedforward
• Feedback specifications (bandwidth, CL peaks . . . )
RP Robust Performance: perf. specs matched ∀ perturbed • Poles and zeros explicit in factorized TF
controllability
Controllability
plants controllability
Controllability
• Particularly suitable to model uncertainties (close models
Nonlinear systems
NMPC
Nonlinear systems
NMPC
in freq. resp.)
ωB , ωC ωB , ωC
Feedback Feedback
control control
principles
Example principles
A system G (s = j ω) is
Control system ẋ1 (t ) = −a1 x1 (t ) + x2 (t ) + β1 u(t ) Control system

• strictly proper if G (j ω) → 0 as ω → ∞
design design
Control problem
ẋ2 (t ) = −a0 x1 (t ) + β0 u(t ) Control problem • semi-proper or bi-proper if G (j ω) → D , 0 as ω → ∞
Time and
frequency y (t ) = x1 (t )
Time and
frequency
• improper if G (j ω) → ∞ as ω → ∞
domains domains
Scaling Scaling For a proper system with n ≥ nz we can use the state-space
Linear models
Sampled Signals
gives (ẋ (t ) → sx (s ) − x (t = 0) and deviation variables) Linear models
Sampled Signals
)
The feedback The feedback ẋ = Ax + Bu
principle y (s ) β1 s + β0 principle ⇔ G (s ) = C (sI − A )−1 B + D
Transfer functions = G (s ) = 2 Transfer functions y = Cx + Du
Feedforward u(s ) s + a1 s + a0 Feedforward

Remarks:
controllability ⇒ Independent of the input! controllability
• D used to model HF effects (zero gain at HF for strictly
Nonlinear systems
NMPC
General form Nonlinear systems
NMPC
proper systems)
Robustness vs. Robustness vs. • use of deviation variables in Laplace domain (remove y0 )
performance βnz s nz + . . . + β1 s + β0 performance
Time domain
G (s ) = n Time domain
• superposition principle for linear systems: additive effect
s + an−1 s n−1 + . . . + a1 s + a0 Gain & phase margins
Maximum peak
ωB , ωC
Maximum peak
ωB , ωC
of the inputs
n: order of the system (pole polynomial), n − nz : pole excess or
relative order.
Feedback Feedback
control control
principles
Scaling principles
E. Witrant E. Witrant • Defining the scaling factors

Control system
Prime importance in applications to simplify model analysis Control system
design and controller design (weight selection). design De êmax , Du ûmax , Dd d̂max , Dr r̂max
Time and
Original unscaled system: Time and
frequency frequency we obtain the scaled variables
domains domains
Scaling ŷ = Ĝ û + Ĝd d̂ ; ê = ŷ − r̂ Scaling
Linear models
Sampled Signals
Linear models
Sampled Signals
y = Gu + Gd d , e =y −r
The feedback Make variables < 1: with respect to the max expected or The feedback
principle
allowed change: principle with G = De−1 ĜDu and Gd = De−1 Ĝd Dd .
Feedforward
Why feedback?
Feedforward
Why feedback?
• Can also use the scaled reference
Stability and
d = d̂ /d̂max and u = û/ûmax ⇒ ŷ = Ĝ ûmax u + Ĝd d̂max d Stability and
controllability controllability r̃ = r̂ /r̂max = Dr−1 r̂ ⇒ r = R r̃ , R De−1 Dr = r̂max /êmax
Nonlinear systems same units of ŷ , r̂ , ê : norm. with respect to largest allowed e or Nonlinear systems
NMPC NMPC
largest expected change in r : • For the worst case with non-symmetric bounds around the
performance performance nominal value, take the “max” distance from nominal value
Time domain
y = ŷ /êmax , r = r̂ /êmax , e = ê /êmax Time domain
Gain & phase margins to bounds for disturbance and “min” for u and e .
ωB , ωC or y = ŷ /r̂max , r = r̂ /r̂max , e = ê /r̂max ωB , ωC
Feedback Feedback
control control
Deriving linear models
1 Formulate nonlinear state space model from physics, i.e.

Control system
design
Input/output graphical representation with scaled Control system
design
Control problem
reference Control problem
ẋ = f (x , u), x (0) = x0
Time and Time and
frequency frequency
domains
Scaling
domains
Scaling
2 Determine steady state operating point (or trajectory) x ∗
Linear models Linear models about which to linearize → u∗ s.t. ẋ ∗ = f (x ∗ , u∗ ) = 0
The feedback The feedback 3 Introduce deviation variables (δx (t ), δu(t )) and linearize:
principle principle
a. subtract (x ∗ , u∗ ) to eliminate the terms involving only s.s.
quantities x̃ = x − x ∗ , ũ = u − u∗
b. linearize using first order Taylor expansion (O(2) ≈ 0) for a
controllability controllability small variation δ
Nonlinear systems Nonlinear systems ∗ ∗ ∗ ∂f ∗ ∂f
NMPC NMPC
f (x , u) ≈ f (x , u ) + (x + δx ) +(u + δu)
∂x (x ∗ ,u∗ ) ∂u (x ∗ ,u∗ )
Robustness vs. Robustness vs. | {z } | {z }
performance performance A B
c. get the deviation dynamics with δx ≈ x̃ , δu ≈ ũ
ωB , ωC ωB , ωC
Conclusions Conclusions δ˙x = A δx + B δu, δx (0) = x (0) − x ∗

Feedback Feedback
control control
principles
From Continuous Dynamics to principles
Sampled Signals Discrete-time signals and systems
design design Discrete-time signal y (kh )
Control problem Control problem P
Time and Continuous-time signals and systems Time and
Fourier transform Y (h ) (ω) = h k∞=−∞ y (kh )e −i ωkh
P
frequency
domains Continuous-time signal y (t ) frequency
domains
z -transform Y (z ) = k∞=−∞ y (kh )z −k
R∞
Scaling
Linear models
Fourier transform Y (ω) = R −∞ y (t )e −i ωt dt Scaling
Linear models
Linear system y (kh ) = g ∗ u(kh )
Sampled Signals
Laplace transform
∞
Y (s ) = −∞ y (t )e −st dt Sampled Signals
Y (h ) (ω) = Gd (e i ωh )U (h ) (ω)
The feedback
principle Linear system y (t ) = g ∗ u(t )
The feedback
principle
Y (z ) = Gd (z )U (z )
Transfer functions
Feedforward Y (ω) = G (ω)U (ω)
Transfer functions
Feedforward
Shift operator q × u(kh ) = u(kh + h ) works as z -variable, but in
Why feedback?
Y (s ) = G (s )U (s )
Why feedback?
time-domain.
controllability Derivation operator p × u(t ) = u̇(t ) works as s -variable, but in controllability
Controllability Controllability Example (0 IC) y (kh ) = 0.5u(kh ) + u(kh − h )
time domain.
(0.5 + q−1 )u(kh )
NMPC NMPC y (kh ) =
Robustness vs.
Example (0 IC) y (t ) = 0.5u̇(t ) + u(t ) Robustness vs. Y (z ) = (0.5 + z −1 )U (z )
Time domain
y (t ) = (0.5p + 1)u(t ) Time domain
Maximum peak
ωB , ωC
Y (s ) = (0.5s + 1)U (s ) Maximum peak
ωB , ωC
Feedback Feedback
control control
principles Sampled systems principles
E. Witrant Continuous-time linear system E. Witrant

Sampled systems (2)
design
Control problem
ẋ (t ) = Ax (t ) + Bu(t ) design
Control problem
Systems with piecewise constant input:
Time and y (t ) = Cx (t ) + Du(t ) Time and • Exact relation possible if u(t ) is constant over each
frequency frequency
domains domains sampling interval.
Scaling
⇒ G (s ) = C (sI − A )−1 B + D . Scaling
Linear models Linear models • Solving for ẋ (i.e. use x (t ) = e At k (t ) and dummy
Sampled Signals
Assume that we sample the inputs and outputs of the system Sampled Signals

µ = t − µ′ ) over one sampling interval gives
principle principle
Feedforward Feedforward x [k + 1] = Ad x [k ] + Bd u[k ]
Stability and Stability and y [k ] = Cx [k ] + Du[k ]

Controllability Controllability Gd (z ) = C (zI − Ad )−1 Bd + D
NMPC NMPC
Rh
Robustness vs. Robustness vs. where Ad = e Ah and Bd = 0
e A µ Bd µ.
ωB , ωC ωB , ωC
Conclusions
Relation between sampled inputs u[k ] and outputs y [k ]? Conclusions
Feedback Feedback
control control
E. Witrant E. Witrant Sampled systems (4)

Control system
Sampled systems (3) Control system Frequency-domain analysis of sampling
design design
Control problem
Example: sampling of scalar system Control problem
• Transfer function of sampled system
Time and • Continuous-time dynamics Time and
frequency frequency
domains domains Gd (z ) = C (zI − Ad )−1 Bd + D
Scaling
Linear models
ẋ (t ) = ax (t ) + bu(t ) Scaling
Linear models
produces same output as G (s ) at sampling intervals.
The feedback
• Assuming that the input u(t ) is constant over a sampling The feedback
principle principle • However, frequency responses are not the same! One has
Transfer functions
Feedforward
interval Transfer functions
Feedforward Z ∞
Why feedback?
x [k + 1] = ad x [k ] + bd u[k ] Why feedback?
i ωh
Stability and
Rh Stability and |G (i ω) − Gd (e )| ≤ ωh |g (τ)|d τ
controllability ah aµ b ah controllability 0
Controllability
where ad = e and bd = 0
e b dµ = a (e − 1). Controllability
NMPC • Note: continuous-time poles in s = a , discrete-time poles NMPC where g (τ) is the impulse response for G (s ).
Robustness vs.
performance
in z = e ah . Robustness vs.
performance
• Good match at low frequencies (ω < 0.1ωs ) ⇒ choose
Time domain
Time domain
sampling frequency ωs > 10× system bandwidth.
ωB , ωC ωB , ωC
Feedback Feedback
control control
principles
Sampling of general systems principles
The feedback principle
• For more general systems, Control configurations

design • nonlinear dynamics, or design
• linear systems where input is not piecewise constant
Time and Time and
frequency
domains
conversion from continuous-time to discrete-time is not frequency
domains
Scaling
Linear models
trivial. Scaling
Linear models
Sampled Signals
• Simple approach: approximate time-derivative with finite Sampled Signals

principle difference: principle
Transfer functions q−1 Transfer functions
Feedforward p ≈ Euler backward (stable) Feedforward
Why feedback? qh Why feedback?
Stability and q−1 Stability and

controllability p ≈ Euler forward controllability
Controllability h Controllability
2 q−1
Tustins (trapezoidal) approximation1
NMPC p ≈ × NMPC
Robustness vs.
h q+1 Robustness vs.
performance (typical for digital control) performance
Gain & phase margins ... Gain & phase margins
ωB , ωC ωB , ωC
1
e.g. using q = e ph and log approx. or trapezoidal rule
Feedback Feedback
control control Closed-loop transfer functions
principles
Feedback control principles
With a 1 dof controller the CL response is
Control system Control system y = G (s )u + Gd (s )d

design One degree-of-freedom controller design
Control problem Control problem = GK (r − y − n) + Gd d
Time and
frequency
Time and
frequency = (I + GK )−1 GK r + (I + GK )−1 Gd d − (I + GK )−1 GK n
domains domains | {z } | {z } | {z }
Scaling Scaling T : ref→y S : out dist→y T
The feedback The feedback The control error (S + T = I ) and plant input are
principle principle
Transfer functions
Feedforward
Transfer functions
Feedforward
e = y − r = −Sr + SGd d − T n
u = KSr − KSGd d − KSn
Controllability
Nonlinear systems
Controllability
Nonlinear systems
Terminology:
NMPC NMPC
Plant input L = GK loop transfer function

performance
Time domain
u = K (s )(r − y − n) performance
Time domain
S = (I + GK )−1 = (I + L )−1 sensitivity function
Gain & phase margins Gain & phase margins −1 −1
Maximum peak
find K (s ) to minimize the control error e = y − r Maximum peak T = (I + GK ) GK = (I + L ) L complementary sens. fun.
ωB , ωC ωB , ωC
S gives sensitivity reduction afforded by CL (w/o: board)
Feedback Feedback
control 2 Degrees of freedom and feedforward dontrol control
r and ym independent, Why feedback?
known d :
design
Control problem
design
Control problem
If invertible plant, feedforward (open-loop, known disturbances):
Time and u = K (r − ym ) + Kr r − Kd d Time and
frequency | {z } | {z } frequency K = 0, Kr (s ) = G −1 (s ), Kd (s ) = G −1 (s )Gd (s )
domains domains
Scaling
feedback feedforward Scaling
Linear models
Sampled Signals
Linear models
Sampled Signals
gives perfect tracking:
The feedback
Suppose perfect measurement (n = 0, ym = y ): The feedback
principle
Transfer functions
principle
Transfer functions
y = G (G −1 r − G −1 Gd d ) + Gd d = r
= (I + GK )−1 [G (K +Kr )r + (Gd −GKd )d ]
Why feedback? y Why feedback?
but feedback is necessary to deal with
Stability and
e = S (−Sr r + Sd Gd d ) Stability and
controllability
Controllability
controllability
Controllability
• signal uncertainty - unknown disturbance
with S = (I + GK )−1 , Sr = 1 − GKr , Sd = GKd Gd−1
NMPC NMPC
• model uncertainty
Robustness vs. (feedforward sensitivity functions). Robustness vs. • an unstable plant (only stabilized by FB)
Time domain If K , Kr and Kd = 0, then S , Sr and Sd = 1, else: Time domain
Maximum peak
ωB , ωC
• SSr small for reference tracking Maximum peak
ωB , ωC
Conclusions • SSd small for disturbance rejection Conclusions

Feedback Feedback
control control
principles
Stability and controllability principles Input-Output Controllability
I. How well can the plant be controlled?
design design • Achievable specifications?
Control problem
Two methods are commonly used for evaluating the Control problem
Time and stability of linear systems Time and

II. Which control structure should be used?
frequency frequency • Measurements, manipulated variables, combinations?
domains domains
Scaling
• Evaluate CL poles Scaling 1. Control the outputs that are not self-regulating
Linear models
Sampled Signals
• zeros of 1 + L (s ) = 0 or eigenvalues of A in LHP Linear models
Sampled Signals
2. Control the outputs that have favorable dynamic and static
The feedback
• best suited for numerical calculations The feedback characteristics
principle • need to approximate time delays as rational transfer principle 3. Select inputs that have large effects on the outputs
Feedforward functions (e.g. Padé approximations) Feedforward 4. Select inputs that rapidly affect the controlled variables
Stability and
• Frequency response of L (j ω) Stability and III. How might the process be changed to improve control?
controllability • Nyquist (encirclements of -1 = nb RHP poles) and Bode controllability
Def. Ability to achieve acceptable control performance:
Nonlinear systems (|L (j ω180 )| < 1) stability criteria Nonlinear systems
• to keep y within specified bounds or displacements from r
NMPC NMPC
• nice graphical interpretation & can be used for time delays
Robustness vs. Robustness vs. • in spite of unknown but bounded variations, such as d and
performance • measure of relative stability and basis for several performance
plant changes
Time domain
robustness tests Time domain
Gain & phase margins • using available inputs u and available measurements ym or
ωB , ωC ωB , ωC dm
Feedback Feedback
control control
Control system Control system Scaling and performance

design design
Control problem
Input-output controllability analysis Control problem
• Scaling such that performances expressed as:
Time and Time and
frequency frequency • bounds: keep y at r ± 1 ∀d ∈ [−1 1] or ∀r ∈ [−R R ] using
domains • Performance targeting domains
u ∈ [−1 1]
Scaling Scaling
Linear models
• Mostly qualitative - simulation - Fundamental properties? Linear models • frequency-by-frequency (i.e. d (t ) = sin ωt ): keep
|e (ω)| ≤ 1 ∀|d (ω)| ≤ 1 or ∀|r (ω)| ≤ R (ω) using |u(ω)| ≤ 1]
The feedback • Rigorous approaches: need mathematical formulation, i.e. The feedback
principle
Transfer functions based on G and Gd
principle
Transfer functions
• Only for frequencies within the system bandwidth
Feedforward
Why feedback? • Linear approach: most important nonlinearity (constrained
Feedforward
Why feedback?
• Recall r = R r̃ and
Stability and
input) can be handled linearly Stability and
controllability
Controllability
controllability
Controllability
e = y − r = Gu + Gd d − R r̃
NMPC NMPC

→ results for d applicable to r with Gd → −R
ωB , ωC ωB , ωC
Feedback Feedback
control control
E. Witrant E. Witrant Stability for nonlinear systems [Marchand, 2009]

Control system Control system Consider the autonomous nonlinear system:
design
Control problem
Remarks on the term controllability design
Control problem
Time and Time and

ẋ = f (x , u(x )) = g (x )
frequency
• Ability of the process to achieve and maintain the desired frequency
domains domains
Scaling
equilibrium value [Ziegler and Nichols’43] Scaling
Linear models
Sampled Signals
• Differs from state controllability: ability to bring a system Linear models
Sampled Signals • Stability: the system is said to be stable at the origin if and
from a given initial state to any final state within a finite
only if (iff): ∀R > 0, ∃ r (R ) > 0 such that ∀x0 ∈ B(r (R )),
principle time [Kalman, 60’s] principle
x (t ; x 0) solution with x0 as initial condition, remains in
Feedforward • little practical interest if unstable modes are both Feedforward
Why feedback?
controllable and observable
Why feedback? B(R ) for all t > 0.
controllability • most industrial plants are controlled quite satisfactorily controllability • Attractivity: the origin is said to be attractive iff:
Nonlinear systems
though they are not state controllable [Rosenbrock’70] Nonlinear systems limt →∞ x (t ; x0 ) = 0.
NMPC NMPC

• Asymptotic stability: the system is said to be
performance
Time domain
performance
Time domain
asymptotically stable at the origin iff it is stable and
Maximum peak
Maximum peak
attractive.
ωB , ωC ωB , ωC
Feedback Feedback
control control
E. Witrant Graphical interpretation E. Witrant
Control system
design
Control system
design
Asymptotic stability and local linearizaton
ẋ = g (x ) and its linearization at the origin
Consider
Time and Time and
∂g
frequency frequency ẋ = ∂x x , then:
domains domains x =0
Scaling Scaling
• Linearization with eig < 0 ⇔ Nonlinear system is locally
asymptotically stable
principle principle • Linearization with eig > 0 ⇔ Nonlinear system is locally
Feedforward Feedforward unstable
Stability and Stability and • Linearization with eig = 0: nothing can be concluded on
Controllability Controllability the nonlinear system (may be stable or unstable)
NMPC • For linear systems: Attractivity → Stability NMPC ⇒ Only local conclusions!
Robustness vs.
performance
• For nonlinear systems: Attractivity does not imply Stability! Robustness vs.
performance
• Stability and attractivity : properties hard to check? Gain & phase margins
ωB , ωC ωB , ωC
Feedback Feedback
control control Example: Pendulum equation with friction [Khalil 2002]
E. Witrant Lyapunov theory: Lyapunov functions E. Witrant
Control system Definition: V : Rn → R is a Lyapunov function if continuous and Control system

ẋ1 = x2
design design
Control problem such that: Control problem
Time and Time and

ẋ2 = −a sin x1 − bx2
frequency 1 (definite) V (x ) = 0 ⇔ x = 0 frequency
domains domains
Scaling 2 (positive) ∀x , V (x ) ≥ 0 Scaling
with the Lyapunov function
Sampled Signals
3 (radially unbounded) lim||x ||→∞ V (x ) = +∞ Sampled Signals
1 2
The feedback The feedback V (x ) = a (1 − cos x1 ) + x
principle - Lyapunov functions are often related to energies principle 2 2
Feedforward (First) Lyapunov theorem: Feedforward ⇒ V̇ (x ) = a ẋ1 sin x1 + x2 ẋ2 = −bx22
Stability and • (strictly decreasing) If ∃ a Lyapunov function V : Rn → R+ Stability and • The origin is stable BUT V̇ (x ) is not negative definite because
Controllability s.t. V (x (t )) is strictly decreasing for all x (0) , 0 then the Controllability V̇ (x ) = 0 for x2 = 0 irrespective of x1 !
NMPC
origin is asymptotically stable. NMPC • The conditions of Lyapunov’s theorem are only sufficient.
Robustness vs.
performance
• (decreasing) If ∃ a Lyapunov function V : Rn → R+ s.t. Robustness vs.
performance
Failure of a Lyapunov function candidate to satisfy the
Time domain V (x (t )) is decreasing then the origin is stable. Time domain
conditions for stability or asymptotic stability does not mean that
Maximum peak
Maximum peak
the equilibrium point is not stable or asymptotically stable. It
ωB , ωC ωB , ωC only means that such stability property cannot be established by
Conclusions Conclusions using this Lyapunov function candidate
Feedback Feedback
control control
Example 2: Pendulum cart [Wang 2011]
Control system
design
Example: Pendulum equation with friction (2) Control system
design
Control problem
Try another Lyapunov function: Control problem
Time and Time and

frequency frequency
domains 1 T domains
Scaling V (x ) = x Px + a (1 − cos x1 ) Scaling
Linear models 2 Linear models
Sampled Signals
2 Sampled Signals
The feedback
p11 > 0, p11 p22 − p12 >0 The feedback
principle 1 1 principle
Transfer functions
⇒ V̇ (x ) = − abx1 sin x1 − bx22 Transfer functions
Feedforward
Why feedback?
2 2 Feedforward
Why feedback?

controllability V (x ) is positive definite and V̇ (x ) is negative definite over controllability
Controllability
Nonlinear systems
D = {x ∈ R 2 | |x1 | < π} Controllability
Nonlinear systems
NMPC
The origin is asymptotically stable NMPC

Time domain Time domain Using Lyapunov-based design the pendulum’s motion converges to
Maximum peak
Maximum peak
the homoclinic orbit (zero energy motion), and the cart displacement
ωB , ωC ωB , ωC converges to zero.
Feedback Feedback
control control
Control system
design
Nonlinear MPC: an intuitive strategy [Alamir, 2012] Control system
design
Nonlinear MPC: an intuitive strategy
Time and Time and

frequency frequency
domains domains
Scaling Scaling

principle principle

NMPC NMPC

ωB , ωC ωB , ωC
Feedback Feedback
control control
Control system
design
Nonlinear MPC: an intuitive strategy Control system
design
Time and Time and

frequency frequency
domains domains
Scaling Scaling

principle principle

NMPC NMPC

ωB , ωC ωB , ωC
Feedback Feedback
control control
Control system
design
design Nonlinear MPC: an intuitive strategy
Time and Time and

frequency frequency
domains domains
Scaling Scaling

principle principle

NMPC NMPC

ωB , ωC ωB , ωC
Feedback Feedback
control control
Control system
design
Time and Time and

frequency frequency
domains domains
Scaling Scaling

principle principle

NMPC NMPC

ωB , ωC ωB , ωC
Feedback Feedback
control control
Control system
design
Time and Time and

frequency frequency
domains domains
Scaling Scaling

principle principle

NMPC NMPC

ωB , ωC ωB , ωC
Feedback Feedback
control control
Control system
design
design
Time and Time and

frequency frequency
domains domains
Scaling Scaling

principle principle

NMPC NMPC

ωB , ωC ωB , ωC
Feedback Feedback
control control Time domain performance (step response)
principles
The robustness versus principles
performance dilemna
design
Control problem
Typical closed-loop response design
Control problem
Time and Example: 2nd order with a RHP zero, PI feedback tuned based Time and
frequency frequency
domains on Ziegler-Nichols domains
Scaling Scaling
Linear models 2 Linear models
y(t)
The feedback
1.5
The feedback • Speed: rise (reach 90 % of final) / settling (within ±5 % of
principle principle
Transfer functions
1
Transfer functions
final) times tr & ts
Why feedback?
0.5
1/3 Why feedback? • Quality: overshoot (peak vs. final, ≤ 20 %), decay ratio
Stability and
controllability
0 u(t) Stability and
controllability
(2nd to 1st peaks ≤ 0.3), steady-state offset (final - desired
Controllability
−0.5
0 10 20 30 40 50
Time [s]
60 70 80 90 100 Controllability value), excess variation = 1 (total variation / overall change
NMPC NMPC at steady state)
Robustness vs.
• oscillatory, long settling time Robustness vs.
performance performance • both included in error e (t ) = y (t ) − r (t ) norm, e.g. L2
• overshoot and large ratio btw. subsequent peaks
Time domain
Time domain
norm (good trade-off andZrelated to optimization)
Maximum peak • too small phase and gain margins, high peaks Maximum peak √ ∞
ωB , ωC ωB , ωC
||e (t )||2 = ISE , ISE = |e (τ)|2 d τ (integral squared error)
Conclusions ⇒ too agressive feedback Conclusions 0
Feedback Feedback
control
principles
control
principles
Frequency domain performance: Gain & phase margins
E. Witrant E. Witrant Gives insights at all frequencies (e.g. crossover).
Excess variation computation
design design
Time and Time and

frequency frequency
domains domains
Scaling Scaling

principle principle
Stability and Stability and • Gain margin (> 2, phase cross-over freq ω180 ):
Controllability Controllability GM = 1/|L (j ω180 )|, where ∠L (j ω180 ) = −180◦
P
NMPC
Total variation is TV = i vi and Excess variation is TV /v0 ⇒ NMPC
• Phase margin (> 30◦ , gain cross-over freq ωc ):
performance as close to 1 as possible! performance PM = ∠L (j ωc ) + 180◦ , where |L (j ωc )| = 1
Maximum peak
Maximum peak
safeguard against time-delay: τmax = PM /ωc
ωB , ωC ωB , ωC
⇒ Good trade-off between performance and stability
Feedback Feedback
control
principles
Frequency domain performance: Maximum peak criteria control
principles
E. Witrant Maximum peaks of sensitivity and complementary sensitivity: E. Witrant

Relationship between time and frequency domain peaks
Control system
design MS = max |S (j ω)|; MT = max |T (j ω)|
Control system
design
Example: second order system
Control problem ω ω Control problem
Time and Time and

frequency Relate to the quality of the response: frequency
domains domains
Scaling • Typically, MS < 2 (6 dB) and MT < 1.25 (2 dB), if large: Scaling
Sampled Signals poor performance & robustness Sampled Signals

principle • Motivation: e = S (Gd d − r ) → ց |e (t )|∀ω where |S | < 1; principle
Feedforward typically small at LF, peak at intermediate F, 1 at HF: MS Feedforward
measures worst-case performance degradation
controllability • Relationships with gain & phase margins: controllability
Controllability
Nonlinear systems
Controllability
Nonlinear systems
⇒ Correlation between ζ and MT , which is often used as an
!
NMPC
MS 1 1 NMPC
approximation of the total variation TV
Robustness vs. GM ≥ ; PM ≥ 2 sin−1 ≥ [rad ] Robustness vs.
(MT ≤ TV ≤ (2n + 1)MT , where n is the order of T ) = classical
performance MS − 1 2MS MS performance
Time domain
Gain & phase margins 1 1
!
1
Time domain
control on response quality.
−1
Maximum peak GM ≥ 1 + ; PM ≥ 2 sin > [rad ] Maximum peak
ωB , ωC MT 2MT MT ωB , ωC
Feedback Feedback
control control
Example: comparison of ωB and ωBT as indicators of
E. Witrant E. Witrant performance.
Bandwidth and crossover frequency Consider the system with
design Bandwidth relates to response speed: design
Control problem
• large: faster tr (HF signals more easily passed to outputs)
Control problem −s + z −s + z 1
Time and Time and
L (s ) = , T (s ) = , z = 0.1, τ = 1
frequency but high sensitivity to noise and parameter variations frequency s (τs + τz + 2) s + z τs + 1
domains domains
Scaling
Linear models
• small: tr slow and more robust system Scaling
Linear models
Sampled Signals
• defined as the frequency range [ω1 , ω2 ] ([0, ωB ]) over Sampled Signals

principle which control is “effective”, in terms of: principle
Transfer functions
Feedforward
• improving performance
√ (e /r = −S small): ωB when |S (j ω)| Transfer functions
Feedforward
Why feedback?
first crosses 1/ 2 from below Why feedback?
Stability and • impact on output / tracking (y /√ r = T large): ωBT is highest Stability and
Controllability freq. when |T (j ω)| crosses 1/ 2 from above Controllability • both L and T have RHP zeros, GM = 2.1, PM = 60.1◦ ,
NMPC
• close in most cases, ωB more reliable NMPC MS = 1.93 and MT = 1 (within acceptable bounds)
Robustness vs.
performance
Gain crossover frequency ωC : where |L (j ω)| first crosses 1 Robustness vs.
performance • ωB = 0.036 & ωC = 0.054 < z (response limited by zero)
Time domain from above. Time domain
but ωBT = 1/τ = 1.0 = 10 × z
ωB , ωC ωB , ωC • step: tr = 31.0 s ≈ 1/ωB = 28.0 s , 1/ωBT
Feedback Feedback
control control

Design objective L
design design Performance, good disturbance rejection
Performance, good command following large
Time and Time and
frequency frequency Stabilization of unstable plant
domains domains
Scaling Scaling
Mitigation of meas. noise on plant outputs small
Linear models Linear models Small magnitude of input signals K & L small
Physical controller must be strictly proper K → 0 at HF
principle principle Nominal stability (stable plant) (RHP z, delays)
Robust stability (stable plant) (uncertain dyn.)
• |T | ≈ 1 up to ωBT but phase drop (−40◦ → −220◦ )

controllability between ωB and ωBT : poor tracking performance! controllability
Generally in different frequency ranges: |L | > 1 at LF (below
Nonlinear systems
• i.e. at ω180 = 0.46, T ≈ −0.9 and response to sin ref Nonlinear systems
NMPC NMPC
ωc ) and |L | < 1 at HF
Robustness vs. completely out of phase Robustness vs.
Time domain ⇒ |T | not sufficient, consider phase also. Time domain
ωB , ωC ωB , ωC
Feedback Feedback
control control
principles
Conclusions principles
References

design
Control problem
design
Control problem
1 S. Skogestad and I. Postlethwaite, Multivariable Feedback
Time and Time and
Control: Analysis and Design, 2nd Ed., Wiley, 2007.
frequency • Control is part of a system design process frequency http://www.nt.ntnu.no/users/skoge/book/ps/book1-3.pdf
domains domains
Scaling • Transfer functions is a key system representation Scaling 2 K.J. Åström and B. Wittenmark, Computer-Controlled Systems:
Theory and Design, 3rd Ed., Prentice Hall, 1997.
Sampled Signals
• Obtained after scaling and linearization Sampled Signals
The feedback The feedback 3 H. Khalil, Nonlinear systems, Prentice-Hall, 2002

principle • Specific care is needed on the sampling stage principle
Transfer functions Transfer functions 4 N. Marchand, Control of Nonlinear Systems, lecture notes,
Feedforward
Why feedback?
• Interest for frequency response and main Feedforward
Why feedback? 2009.
Stability and characterizations Stability and http://www.gipsa-lab.grenoble-inp.fr/˜ nicolas.marchand/teaching/Nonlinear_PSPI.pdf
Controllability • Feedback and CL transfer → sensitivity functions Controllability 5 M. Alamir, Optimal & Predictive Control, lecture notes, 2012.
http://www.mazenalamir.fr/homepage/wa_files/predictive_ense3.pdf
NMPC
• Quality criteria of CL response (time and frequency) NMPC
Robustness vs. Robustness vs. 6 Yizhou Wang, ME237 Project: Nonlinear Control of a Cart
Time domain Time domain Pendulum System, 2011.
Maximum peak
Maximum peak
http://www.me.berkeley.edu/˜yzhwang/invpen.pdf
ωB , ωC ωB , ωC
The mine The mine
network network
Outline
of Physical of Physical
modeling modeling
Electrical Circuits Electrical Circuits
Mechanical Mechanical
1 Fundamentals of Physical modeling
Translation Translation
Mechanical Rotation Mechanical Rotation
Flow Systems Flow Systems
Thermal Systems Thermal Systems
Some Observations Some Observations
2 Mine ventilation circuits and networks
network
Kirchhoff’s laws
Lesson 3: Model of mine ventilation network network
Kirchhoff’s laws
Series circuits
system Series circuits 3 Dynamic models of the network
Parallel circuits Parallel circuits
Ventilation networks Ventilation networks
Simple networks with Simple networks with
natural splitting
Complex networks Emmanuel WITRANT
natural splitting
Complex networks
4 Simulation softwares
Dynamics of emmanuel.witrant@ujf-grenoble.fr Dynamics of
the network the network
Non-minimal model Non-minimal model
Minimal model
Universidad Pedagógica y Tecnológica de Colombia, Minimal model
Simulation Simulation
softwares
Sogamoso, April 2, 2013 softwares
The mine The mine

network
Fundamentals of Physical network
Electrical Circuits
Fundamentals
modeling [Ljung et al. 1994] Fundamentals
of Physical
modeling
of Physical
modeling
Fundamental quantities:
Electrical Circuits
Mechanical
Electrical Circuits
Mechanical
voltage u (volt) and current i (ampere).
Flow Systems Flow Systems Components:
Ventilation • Most common relationships within a number of areas in Nature

Ventilation Relationship (law) Energy
network network Z t
physics. Inductor 1 di (t ) T (t ) = 21 Li 2 (t )
Kirchhoff’s laws Kirchhoff’s laws
i (t ) = u(s )ds , u(t ) = L
Series circuits Series ( L henry)
circuits
L Z0 dt (magnetic field E storage, J )
Parallel circuits • More general relationships become visible. Parallel circuits t
Ventilation networks Capacitor
Ventilation networks 1 du(t ) T (t ) = 21 Cu2 (t )
u (t ) = i (s )ds , i (t ) = C
Simple networks with
natural splitting
⇒ General modeling strategy. (C farad)
natural splitting C 0 dt (electric field E storage)
Complex networks ComplexResistor
networks
u(t ) = Ri (t )
Dynamics of Dynamics(R ohm) of
Nonlinear P (t ) = u (t ) · i (t )
u(t ) = h1 (t )i (t ), i (t ) = h2 (t )u(t )
Minimal model Minimalresistance
model
( (loss, in watts, 1 W = 1 J /s )
Simulation Ideal
Simulation x, x > 0
softwares softwares h2 (t ) =
rectifier 0, x ≤ 0
The mine The mine
network
Interconnections (Kirkhhoff’s laws): network
Mechanical Translation
X X
Fundamentals
of Physical
ik (t ) ≡ 0 (nodes ), uk (t ) ≡ 0 (loops ). Fundamentals
of Physical Fundamental quantities:
modeling k k modeling
Electrical Circuits Electrical Circuits force F (newton) and velocity v (m/s ), 3-D vectors (suppose
Translation
Mechanical Rotation
Translation
Mechanical Rotation
constant mass ṁ = 0).
Flow Systems Ideal transformer: Flow Systems
Thermal Systems
Some Observations transform voltage and current s.t. their product is constant:
Thermal Systems
Some Observations
Components:
Nature Relationship (law) Energy
Ventilation Ventilation Z t
network 1 network Newton’s 1 dv (t ) T (t ) = 12 mv 2 (t )
Kirchhoff’s laws u1 · i1 = u2 · i2 , u1 = αu2 , i1 = i2 Kirchhoff’s laws v (t ) = F (s )ds , F (t ) = m
force law mZ0 dt (kinetic E storage)
Series circuits α Series circuits
t
Parallel circuits Parallel circuits Elastic bodies 1 dF (t ) T (t ) = 21k F 2 (t )
Ventilation networks Ventilation networks F (t ) = k v (s )ds , v (t ) =
Simple networks with Simple networks with (k N /m ) 0 k dt (elastic E storage)
natural splitting natural splitting
Friction
Complex networks Complex networks
F (t ) = h (v (t ))
Dynamics of Dynamics of
the network the network Air drag h (x ) = cx 2 sgn(x ) P (t ) = F (t ) · v (t )
Non-minimal model
Minimal model
Non-minimal model
Minimal model
Dampers h (x ) = γ x (lost as heat)
+µ if x > 0


Simulation Simulation Dry 

h (x ) =  F0 if x = 0

softwares softwares
friction 
 −µ if x < 0

The mine The mine

network network
E. Witrant Interconnections: E. Witrant
Example: active seismic isolation control [Itagaki &
of Physical of Physical Nishimura, 2004]
modeling X modeling
Electrical Circuits
Fk (t ) ≡ 0 (body at rest) Electrical Circuits Mass - spring - damper approximation:
Translation
k Translation
Flow Systems
Thermal Systems
v1 (t ) = v2 (t ) = . . . = vn (t ) (interconnection point) Flow Systems
Thermal Systems


 m4 ẍ4 (t ) = γ4 (ẋ3 − ẋ4 ) + k4 (x3 − x4 )

mi ẍi (t ) = [γi (ẋi −1 − ẋi ) + ki (xi −1 − xi )]





+[γi +1 (ẋi +1 − ẋi )



network network


Kirchhoff’s laws
Ideal transformer: Kirchhoff’s laws




 +ki +1 (xi +1 − xi )], i = 2, 3

Series circuits
force amplification thanks to levers: Series circuits
m1 ẍ1 (t ) = [γ1 (ẋ0 − ẋ1 ) + k1 (x0 − x1 )]

Parallel circuits Parallel circuits 

+[γ2 (ẋ2 − ẋ1 ) + k2 (x2 − x1 )]





F1 · v1 = F2 · v2 +u(t )






m1 ẍ0 (t ) = Fearth (t )

F 1 = αF 2


Dynamics of Dynamics of 
[ẍ0 + ẍ1 x2 − x1 ]T

y (t ) =


Non-minimal model 1 Non-minimal model
Minimal model v1 = v2 Minimal model
Simulation
α Simulation
softwares softwares
The mine The mine
network
Mechanical Rotation network
Fundamentals Fundamentals Interconnections:

modeling modeling X
Mk (t ) ≡ 0 (body at rest).
Mechanical
Translation Fundamental quantities: Mechanical
Translation
k
Flow Systems torque M [N · m] and angular velocity ω [rad /s ]. Flow Systems
Components: Ideal transformer:

network Nature Relationship (law) Energy network a pair of gears transforms torque and angular velocity as:
t
d ω(t ) T (t ) = 21 J ω2 (t )
Z
Series circuits Inertia 1 Series circuits
ω(t ) = M (s )ds , M (t ) = J
Parallel circuits
Ventilation networks
J [Nm/s 2 ] J Z0 dt (rotational E storage) Parallel circuits
M1 · ω 1 = M2 · ω 2
t
Simple networks with Torsional 1 dM (t ) T (t ) = 21k M 2 (t ) Simple networks with
natural splitting M (t ) = k ω(s )ds , ω(t ) = natural splitting M1 = α M2
Complex networks stiffness k 0 k dt (torsional E storage) Complex networks
Rotational 1
Dynamics of
M (t ) = h (ω(t )) P (t ) = M (t ) · ω(t )
Dynamics of ω1 = ω2
the network
Non-minimal model
friction the network
Non-minimal model
α
Minimal model Minimal model
softwares softwares
The mine The mine

network
Example: printer belt pulley [Dorf & Bishop 2001]
network
Flow Systems
Fundamentals
of Physical
Fundamentals
of Physical
Fundamental quantities:
modeling
Electrical Circuits
modeling
Electrical Circuits
for incompressible fluids, pressure p [N /m2 ] and flow Q [m3 /s ].
Mechanical Rotation Mechanical Rotation Fluid in a tube:
network network
Series circuits Series circuits
= k (r θ − r θp ) = k (r θ − y )

Parallel circuits  Spring tension: T1 Parallel circuits
Pressure gradient ∇p force p·A



Spring tension: T2 = k (y − r θ) Ventilation networks





d2 y mass ρ·l·A flow Q =v ·A
natural splitting


 natural splitting
Newton: T1 − T2 = m 2
Complex networks


 Complex networks
dt inertance [kg /m4 ] Lf = ρ · l /A



Km
Dynamics of
 Dynamics of
Motor torque (resistance, L = 0: Mm = Km i = v2
the network


 the network
R


Non-minimal model  Non-minimal model
Constitutive relationships (Newton: sum of forces = mass ×

drives belts + disturb.: Mm = M + Md
Minimal model


 Minimal model


d2 θ dθ

accel.):
Simulation

 Simulation
T drives shaft to pulleys: M = J 2 +h + r (T1 − T2 )


softwares  softwares t
dt dt dQ (t ) T (t ) = 12 Lf Q 2 (t )
Z
1
Conclusions Conclusions Q (t ) = ∇p (s )ds , ∇p (t ) = Lf
Lf 0 dt (kinetic E storage)
The mine The mine
network network
Fundamentals Flow in a tank: Fundamentals Flow through a section reduction:

modeling modeling
network R network
Kirchhoff’s laws • Volume V = Qdt , h = V /A , and fluid capacitance Kirchhoff’s laws
• Pressure p , Flow resistance Rf , constant H .
4 2
Parallel circuits Cf A /ρg [m s /kg ]. Parallel circuits
• Constitutive relationships:
natural splitting • Constitutive relationships: Simple networks with
natural splitting Pressure drop p (t ) = h (Q (t ))
Z t
T (t ) = 12 Cf p 2 (t )
Complex networks
Bottom pres. 1 Complex networks
d’Arcy’s law p (t ) = Rf Q (t ), P (t ) = p (t ) · Q (t )
Dynamics of p (t ) = Q (s )ds Dynamics of
the network p = ρ·h·g Cf 0 (potential E storage) the network area change p (t ) = H · Q 2 (t ) · sgnQ (t )
softwares softwares
The mine The mine

network network
Thermal Systems
Interconnections:
of Physical of Physical Fundamental quantities:
modeling modeling
Electrical Circuits
X X Electrical Circuits temperature T [K ] and heat flow rate q [W ].
Mechanical
Translation
Qk (t ) ≡ 0 (flows at a junction), pk ≡ 0 (in a loop) Mechanical
Translation
Mechanical Rotation
Flow Systems
k k Mechanical Rotation
Flow Systems
Body heating:
Fourier’s law of conduction
t
dT (t )
Z
Thermal capacity 1
Ventilation Ideal transformer: Ventilation T (t ) = q(s )ds , q(t ) = C
network network C [J /(K · s )] C 0 dt
Parallel circuits
Parallel circuits
Interconnections:
Dynamics of Dynamics of q(t ) = W ∆T (t ) (heat transf. btw. 2 bodies)

the network
1 the network X
Non-minimal model
p1 · Q1 = p2 · Q2 , p1 = αp2 , Q1 = Q2 . Non-minimal model
qk (t ) ≡ 0 (at one point).
Minimal model
α Minimal model
Simulation Simulation k
softwares softwares
Conclusions Conclusions where W [J /(K · s )] is the heat transfer coefficient.

The mine The mine
Characteristics:
network
Some Observations network
1 Effort variable e ;
of Physical of Physical 2 Flow variable f ;
modeling modeling R
Electrical Circuits Electrical Circuits 3 Effort storage: f = α−1 · e;
Translation Obvious similarities Translation
4 Flow storage: e = β−1 ·
R
f;
Flow Systems among the basic equations for different systems! Flow Systems
Thermal Systems
Some Observations
Thermal Systems
Some Observations
5 Power dissipation: P = e · f ;
Some physical analogies: Energy storage via I.: T = 1 2
Ventilation Ventilation 6 2α f ;
System
network Effort Flow Eff. storage Flow stor. Static relation network
Kirchhoff’s laws
Electrical
Series circuits Voltage Current Inductor Capacitor Resistor
Kirchhoff’s laws
Series circuits
7 Energy storage via C.: T = 21β e 2 ;
fi = 0;
P
Mechanical:
8 Sum of flows equal to zero:
- Translational
Simple networks with Force Velocity Body (mass) Spring Friction Simple networks with
ei = 0;
P
natural splitting
- Rotational Torque Angular V. Axis (inertia) Torsion s. Friction
natural splitting 9 Sum of efforts (with signs) equal to zero:
Hydraulic
Dynamics of Pressure Flow Tube Tank Section Dynamics of 10 Transformation of variables: e1 f1 = e2 f2 .
theThermal
network Temperature Heat flow rate - Heater Heat transfer the network
• Note: analogies may be complete or not (i.e. thermal).
softwares softwares ⇒ Create systematic, application-independent modeling from
Conclusions Conclusions these analogies: see Bond Graphs.
The mine The mine

network
Mine ventilation circuits and network
Kirchhoff’s laws
E. Witrant E. Witrant Second: voltage or loop:
networks [Hartman et al. 1997] P
Hl = 0 = Hla + Hlb + Hlc − Hld
Fundamentals
of Physical
Fundamentals
of Physical
First: current or junction law:
Q = 0 = Q1 + Q2 − Q3 − Q4
P
modeling modeling
Electrical Circuits
Relationship between head and quantity Electrical Circuits
Mechanical Rotation • Head loss, static, velocity and total proportional to squared Mechanical Rotation
Thermal Systems airflow quantity: Hl , Hs , Hv , Ht ∝ Q 2 Thermal Systems
Ventilation • Characteristic curve: Ventilation

network network
Kirchhoff’s laws !2 !2 Kirchhoff’s laws
Series circuits
H1 Q1 Q1 Series circuits
Parallel circuits
= ⇔ H1 = H2 Parallel circuits
H2 Q2 Q2 Ventilation networks
Example: fan whose Hs = 2 in. Hl = R |Q |Q to keep sign
the network water, Ht = 3 in. water at the network convention.
E.g. Hl = Ra |Q1 |Q1 + Rb |Q1 |Q1 + Rc |Q1 |Q1 − Rd |Q2 |Q2 = 0

P
Minimal model 400,000 cfm: Minimal model
Simulation
softwares
• Airway resistance R from Atkinson: Hl = RQ 2 with Simulation
softwares
and add pressure sources as negative pressure drops (head
KO (L +Le ) loss)
Conclusions R= A3
N·s2 /m8 (equiv. Ohm’s law). Conclusions
The mine The mine
ventilation Series circuits: End to end airways ventilation
Parallel circuits: airflow splitting (natural or controlled)
network network
modeling modeling
Mechanical Rotation
Flow Systems
Mechanical Rotation
Flow Systems
• Kirchhoff’s first law Q = Q1 + Q2 + Q3 + . . .
Some Observations Some Observations • Second law: Hl1 = Hl2 = Hl3 = . . .
network network
• Equivalent resistance obtained as:
• quantity of air conserved Q = Q1 = Q2 = Q3 = . . . p p p q
Series circuits Series circuits Q= Hl /R1 + Hl /R2 + Hl /R3 = Hl /Req with
• fan head = total head loss Ventilation networks
√1 = √1 + √1 + √1 + ...
natural splitting
natural splitting
Req R1 R2 R3
Complex networks (static): Complex networks
• Quantity-divider rule: Req Q 2 = R1 Q12 = R2 Q22 = . . . gives
Dynamics of Hl1 + Hl2 + Hl3 − Hfan = 0 Dynamics of p p
the network the network Q1 = Q Req /R1 , Q2 = Q Req /R2 , etc.
Non-minimal model • equivalent resistance obtained Non-minimal model
R
Minimal model
as: Hl = R1 |Q |Q + R2 |Q |Q +
Minimal model
• In Na airways of same characteristics: Hl =
Na2
Q 2,
in
softwares
R3 |Q |Q + . . . = Req Q 2 with softwares Atkinson’s equation: total perimeter Na O and total area
Conclusions
Req = Conclusions Na A
P
Ri
The mine The mine

ventilation Parallel circuits (2) ventilation Ventilation networks: some definitions
network network
E. Witrant • Characteristic curves: E. Witrant • Node: point where ≥ 2 airways intersect = “junction”
Fundamentals cumulative for a given head → Fundamentals • Branch (arc): connecting line (airway) between 2 nodes
modeling • Controlled splitting: modeling
• Graph: set of nodes, with certain pairs
Mechanical
Translation
• artificial resistance Mechanical
Translation connected by branches
Mechanical Rotation
Flow Systems
(regulators) in all but one Mechanical Rotation
Flow Systems • Connected graph: all nodes are connected
Thermal Systems branch (free branch, with Thermal Systems
together by branches
highest head)
Ventilation Ventilation • Network: graph with a flow associated with
network network
Kirchhoff’s laws • raise head and power requirements Kirchhoff’s laws
each branche, can be connected
• Directed network: sign/direction associated
• regulators = variable openings, larger means smaller shocks.
with each branch
natural splitting
Size from shock-loss formula (circular): Simple networks with
natural splitting
Complex networks Complex networks • Degree of a node: number of connected
h 1/C i2 q
Hx c −N 1 Ar z√ branches
Dynamics of X= Hv
= N
, Cc = √ ⇒N= A
= Dynamics of
the network z −zN 2 +N 2 X +2 X +z the network
Non-minimal model Non-minimal model • Mesh (cycle): connected path in which every
where X = shock-loss factor, Hx = needed shock loss, N = ratio node is of deg. 2 with respect to the path
softwares of orifice (regulator) area Ar to airway area A , Cc = coef. of softwares • (spanning) tree: connected graph without mesh
Conclusions contraction, z = contraction factor (e.g. 2.5 for mine). Conclusions
• Branch in a tree: branch contained in a spanning tree
• i.e. given Q , Hx and A , find Ar
The mine The mine
network network
E. Witrant Ventilation networks: some definitions (2) E. Witrant
Fundamentals • Chord (basic branch): in the network but not in a given Fundamentals
Solution of simple networks with natural splitting
modeling spanning tree modeling
Mechanical
• Basic mesh: contains only one chord and the unique path Mechanical
Translation
Mechanical Rotation
Translation
Mechanical Rotation
To combine series or parallel
formed by branches in the tree between 2 nodes of the
Flow Systems
Thermal Systems
Flow Systems
Thermal Systems
airways:
Some Observations
chord Some Observations
• Algebraic solution: alternate
Ventilation • Chord set: set containing all the chords of a network, Ventilation
network network algebraic solving of each
Kirchhoff’s laws unique for a given spanning tree Kirchhoff’s laws
Series circuits Series circuits branch and mesh
Parallel circuits
• Mesh base: set containing all basic meshes, unique for a Parallel circuits
→ provides Req to solve:
given spanning tree Simple networks with
natural splitting
Complex networks
natural splitting
Complex networks
Mine Hs = Hl = Q 2 Req
Dynamics of
• Network degree: equal to the number of chords Dynamics of
Non-minimal model ⇒ Series and parallel circuits sufficients for analysing simple Non-minimal model
graphs, but advanced definitions necessary for complex
softwares networks softwares
The mine The mine

ventilation Solution of simple networks with natural splitting (2) ventilation
network network Analysis of complex networks
E. Witrant • Graphical solution: use the characteristic curve E. Witrant
Based on spanning trees and chords, i.e. create the tree as:
1 choose any node
modeling modeling
Electrical Circuits Electrical Circuits 2 arbitrarily connect another node that is one branch away
Translation
• sum the quantities for
Translation → connected set of nodes
Flow Systems
Thermal Systems loops
Flow Systems
Thermal Systems
3 arbitrarily connect another node that is one branch away
from the connected set
Ventilation
• sum the heads for Ventilation
network
series network 4 if all nodes ∈ connected set: stop, else: go to step 3
Parallel circuits Parallel circuits Example: R in in·min2 /ft6 × 1010 (N·s2 /m8 )
Dynamics of 1 combine all series airways Dynamics of

Non-minimal model
Minimal model
2 plot main splits (e.g. B and D), then the equivalent one (E) Non-minimal model
Minimal model
Simulation 3 determine the quantity flowing in each branch for the mine Simulation
softwares softwares
quantity Q
4 plot the secondary splits then determine Q in each branch
The mine The mine
network Complex networks: minimum-resistance spanning tree network
Complex networks: some properties
Consider the resistance as a distance: • ∃ at least one tree in every connected network
of Physical 1 choose any node and connect to its closest adjacent node of Physical • For a given tree in a connected network with Nn nodes
modeling modeling
Electrical Circuits 2 find the unconnected node nearest to a connected node Electrical Circuits and Nb branches, there are exactly Nn − 1 branches in the
Translation and connect the two Translation tree and Nm = Nb − (Nn − 1) chords = network degree
Flow Systems
Thermal Systems
3 if all nodes are connected: stop, else: go to step 2 Flow Systems
Thermal Systems
• Any fluid flowing through the network is at equilibrium if
Some Observations
⇒ Identical regardless of the starting node unless ties occur
Some Observations
Kirchhoff’s laws are satisfied
network when the final branches are chosen network • The quantity Q flowing in each branch ∈ chord set are
Series circuits Series circuits independant and the Q of any branch ∈ tree is a linear
Parallel circuits Example: Parallel circuits
Ventilation networks Ventilation networks combination of the Q flowing in the chords.
natural splitting
Complex networks
natural splitting
Complex networks
• Kirchhoff’s 1st law satisfied for Q at every node
Dynamics of Dynamics of • The set of head losses Hl of the branches ∈ tree uniquely
Non-minimal model Non-minimal model determines the chord Hl .
Simulation Simulation • Hl in branches satisfying Kirchhoff’s 2nd law for a mesh

softwares softwares
⇒ base satisfy it also ∀ other mesh and mesh base in the
network.
The mine The mine

network network
Complex networks: Mathematical representation (2)
Complex networks: Mathematical representation
Fundamentals Fundamentals • Mesh equations: Nm independent ones in the mesh base
of Physical Given a network with Nn nodes and Nb branches, find Q , Hl of Physical
relating the Nb head losses: N b H = 0, k = 1 . . . Nm ,
P b
modeling modeling
j =1 kj j
Electrical Circuits and R in each branch → 3 Nb variables: Electrical Circuits
Mechanical Mechanical where
Mechanical Rotation • Banch equations: Hlj = Rj |Qj |Qj , j = 1 . . . Nb ⇒ Nb Mechanical Rotation 
Flow Systems
Thermal Systems nonlinear independant eq.
Flow Systems
Thermal Systems


 −1 if branch j ∈ basic mesh corresponding to chord

Some Observations PNb Some Observations

 k and if, travelling the mesh from the starting node
• Node equations: aij Qj = 0, i = 1 . . . Nn where


j =1

of the chord toward its end node, the end node of



network network 



 branch j is found first
−1 if the starting node of branch j = i

bkj = 
Series circuits  Series circuits

Parallel circuits

 Parallel circuits  0 if branch j not in basic mesh corresponding to
aij =  0 if neither of the nodes of branch j = i
 

Ventilation networks Ventilation networks 
chord k

 Simple networks with



natural splitting
 1 if the end node of branch j = i natural splitting







 1 if branch j ∈ basic mesh corresponding to chord
k and if the starting node of branch j is found first

The resulting matrix EQ = [aij ] is called the incidence

Minimal model matrix, ⇒ (Nn − 1) linear independant eq. Minimal model
The resulting matrix EH = [bkj ] is called the
softwares softwares fundamental-mesh matrix
The mine The mine
network network
Dynamic models of the network
[Hu, Koroleva, Krstic 2003]
of Physical
modeling
Complex networks: Mathematical representation (3) of Physical
modeling Pipe flow dynamics
Electrical Circuits
Balance between variables and independent equations: Electrical Circuits
Mechanical
Translation
Mechanical
Translation • Due to an unsteady incompressible fluid
Flow Systems Number of variables Number of independent equations Flow Systems
• Head drop = Head loss + Energy storage due to a change
Some Observations Quantities Nb Branch equations Nb Some Observations
in the flow quantity:
Ventilation
network
Head losses Nb Node equations Nn − 1 Ventilation
network
Resistances Nb Mesh equations Nb − (Nn − 1) dQj dQj
Kirchhoff’s laws
Series circuits
Kirchhoff’s laws
Series circuits Htj = Rj |Qj |Qj + Lfj ⇔ = −Kj Rj |Qj |Qj + Kj Htj
Parallel circuits
Total 3 Nb 2 Nb Parallel circuits dt dt
where Kj = Aj /(ρlj ) is the inverse of the inertance (Lf ), Aj

Complex networks ⇒ Need Nb initial known variables to solve the ventilation Complex networks
the branch cross-section area and lj the branch length.
Dynamics of
the network
network problem. Dynamics of
the network
• Node and mesh eq. with one fan: EQ Q = 0 and EH H = 0
with EQ ∈ R(Nn −2)×Nb full rank and EH ∈ RNm ×Nb
softwares softwares • Fan in branch m: eQm Q = Qm , eHm H = Hm and dynamics
Conclusions Conclusions in branch set by Hm = d − Rm Qm , d = pressure drop
generated by the fan.
The mine The mine

network
Non-minimal model of the network network
E. Witrant E. Witrant Non-minimal model of the network (2)
Distinguish the Nb − (Nn − 1) chords from the Nn − 1 branches
Fundamentals in the tree to partition the air flow quantities and heads Fundamentals
Proposition 1 [Hu et al. 2003]: There exists matrices YRQ , YQ
modeling (H = Ht ) as modeling
and Yd of appropriate dimensions so that the full order model
Electrical Circuits
Mechanical
Electrical Circuits
Mechanical
of the mine ventilation network can be expressed as:
   
Mechanical Rotation  Q1   H1  Mechanical Rotation
Flow Systems  ..   ..  Flow Systems Q̇ = −K (I − YRQ )QD2 R + KYQ Q + KYd d
Thermal Systems  .   .  Thermal Systems
Some Observations
" #   " #   Some Observations
H = YRQ QD R + YQ Q + Yd d
Qc  Q Hc  H
=  Nb −(Nn −1) =  Nb −(Nn −1)
Ventilation
    Ventilation
network Q=  , H=  network
Kirchhoff’s laws
Qa  QNb −(Nn −2)  Ha  HNb −(Nn −2)  Kirchhoff’s laws Proof and matrix construction: use of the chords/tree partition,
Series circuits  ..   ..  Series circuits
Parallel circuits  .

  .

 Parallel circuits see paper.
Ventilation networks   Ventilation networks
natural splitting
QNb HNb Simple networks with
natural splitting
Notes for control:
Dynamics of
the network
Defining Dynamics of
the network • Q ∈ RNb
Minimal model
QD2 = diag (|Qj |Qj , . . . , |Qj |Qj ), K = diag (K1 , . . . , KNb ) Minimal model • “Bilinear” in the state
softwares softwares • Linear in d but input-to-state couplings in R
Conclusions we obtain the dynamics: Q̇ = −KQD2 R + KH Conclusions
The mine The mine
Minimal model of the network
network network
Simulation softwares
E. Witrant Proposition 2 [Hu et al. 2003]: There exist matrices Ac , Aca , Bc , E. Witrant
Fundamentals
Cc , YRQc , YRQa , YQc and Yd of appropriate dimensions so that Fundamentals
of Physical
modeling
the minimal model of mine ventilation network system can be of Physical
modeling
Electrical Circuits
Mechanical
expressed as: Electrical Circuits
Mechanical
Mechanical Rotation
Flow Systems Q̇c 2
= Ac QcD 2
Rc + Aca QaD Ra + Bc Qc + Cc d
Mechanical Rotation
Flow Systems
• VENTSIM: simulate airflows, pressure and heats from a
Thermal Systems
2 2
Thermal Systems
modeled network of airways. http://www.ventsim.com/
Some Observations
Ha = YRQc QcD Rc + YRQa QaD Ra + YQc Qc + Yd d Some Observations
Ventilation Ventilation • VUMA-3D: mine ventilation, cooling and environment

network network
Kirchhoff’s laws where Qc is the state, Rc , Ra and d are the control inputs, and Kirchhoff’s laws control. http://www.vuma.co.za/
Parallel circuits Ha is the system output, and: Parallel circuits
• VnetPC PRO+: branch templates, fans, shock loss.
natural splitting 2
natural splitting
http://www.mvsengineering.com/
Complex networks QcD = diag (Q1 |Q1 |, . . . , QNb −(Nn −1) |QNb −(Nn −1) |) Complex networks
• ...
Dynamics of 2 Dynamics of
the network QaD = diag (QNb −(Nn −2) |QNb −(Nn −2) |, . . . , QNb |QNb |) the network
Minimal model R = [RcT RaT ]T Minimal model
softwares softwares
Proof and matrix construction: see paper.
Note: Qc ∈ RNb −(Nn −1) ⇒ State dimension reduced by (Nn − 1)!
The mine The mine

network
Conclusions network
References
Fundamentals
of Physical
Fundamentals
of Physical
1 L. Ljung and T. Glad, Modeling of Dynamic Systems, Prentice
modeling • A process typically involves to consider multiple physical modeling Hall Information and System Sciences Series, 1994.
Mechanical
Translation
domains: some general rules can be drawn and Mechanical
Translation 2 Noriaki Itagaki and Hidekazu Nishimura, “Gain-Scheduled
Mechanical Rotation
Flow Systems
appropriate tools for multi-physics modeling exist (e.g. Mechanical Rotation
Flow Systems
Vibration Control in Consideration of Actuator Saturation”, Proc.
Thermal Systems
bond graphs) Thermal Systems
of the IEEE Conference on Control Applications, pp 474- 479,
• Modeling and analysis of the mine ventilation network: vol.1, Taipei, Taiwan, Sept. 2-4, 2004.
network network
Kirchhoff’s laws • from the mine topology, establish the steady-state model: Kirchhoff’s laws 3 R.C. Dorf and R.H. Bishop, Modern Control Systems, 9th Ed.,
Series circuits
Parallel circuits
ventilation design and equilibrium behavior Series circuits
Parallel circuits
Prentice Hall, 2001.
Ventilation networks • simplify the graph to size the fans/forced ventilation Ventilation networks
Simple networks with Simple networks with 4 Hartman HL, Mutmansky JM, Ramani RV, Wang YJ. Mine
natural splitting apparatus natural splitting
Complex networks Complex networks Ventilation and Air Conditioning (3rd edn), Ch. 7. Wiley: New
• include the dynamics by considering the inertia of the air
Dynamics of Dynamics of York, 1997.
the network volume the network
Non-minimal model
Minimal model
• simulate the minimal model Non-minimal model
Minimal model
5 Y. Hu, O. Koroleva, and M. Krstic, “Nonlinear control of mine
Simulation • Several simulation softwares can be used Simulation
ventilation networks,” Systems and Control Letters, vol. 49, pp.
softwares softwares 239-254, 2003.
Conclusions Conclusions http://flyingv.ucsd.edu/papers/PDF/62.pdf
Air flow Air flow
modeling in modeling in
deep wells deep wells
Outline
Fluid dynamics Fluid dynamics

General form of a General form of a
conservation law conservation law
Convection-diffusion Convection-diffusion
Euler and Euler and
Navier-Stokes Navier-Stokes
1 The basic equations of fluid dynamics
Firn example Firn example
Cryogenics/QRL Cryogenics/QRL
Lumped Lumped
models models 2 From Euler equations to lumped models
0-D model: main 0-D model: main
characteristics characteristics
Bond graph modeling
overview Lesson 4: Air flow modeling in deep wells Bond graph modeling
overview
3 Fans, rooms and pollutant sources
Fans, rooms Fans, rooms
and pollutant and pollutant
sources sources
Simulation results
Emmanuel WITRANT Simulation results 4 Volume-averaging and estimation of the transport coefficients
Volume- emmanuel.witrant@ujf-grenoble.fr Volume-
averaging and averaging and
transport transport
coefficients Universidad Pedagógica y Tecnológica de Colombia, coefficients
Parameter estimation Parameter estimation
Time-delays Sogamoso, April 3, 2013 Time-delays
Air flow Air flow

deep wells
The basic equations of fluid deep wells
General form of a conservation law
dynamics [Hirsch 2007]
• Conservation: the variation of a conserved (intensive) flow
Convection-diffusion Convection-diffusion quantity U in a given volume results from internal sources
Euler and Euler and
Navier-Stokes
• Model from physics: subatomic, atomics or molecular, Navier-Stokes and the quantity, the flux, crossing the boundary
Cryogenics/QRL
microscopic, macroscopic, astronomical scale Cryogenics/QRL
• Fluxes and sources depend on space-time coordinates, +
Lumped Lumped
models • Fluid dynamics = study of the interactive motion and models on the fluid motion
characteristics
behavior of a large number of elements characteristics • Not all flow quantities obey conservation laws. Fluid flows
Bond graph modeling Bond graph modeling
overview overview fully described by the conservation of
• System of interacting elements as a continuum
Fans, rooms
and pollutant
Fans, rooms
and pollutant
1mass
sources
• Consider an elementary volume that contains a sufficiently sources 2momentum (3-D vector)
Simulation results
large number of molecules with well defined mean velocity Simulation results
3 energy
Volume- Volume-
averaging and and mean kinetic energy averaging and ⇒ 5 equations
transport transport
coefficients • At each point we can thus infer, e.g. velocity, temperature, coefficients • Other quantities can be used but will not take the form of a
Time-delays pressure, entropy etc. Time-delays conservation law
Air flow Air flow
modeling in Scalar conservation law modeling in
E. Witrant
Consider: E. Witrant
- a scalar quantity per unit Scalar conservation law (2)
General form of a volume U , General form of a
Provides the integral conservation form for quantity U :
Convection-diffusion - an arbitrary volume Ω fixed in Convection-diffusion
∂
Z I Z I
Euler and
Navier-Stokes
space (control volume)
Euler and
Navier-Stokes
Ud Ω + ~ · dS
F ~ = QV d Ω + ~ S · dS
Q ~
∂t Ω S Ω S
Cryogenics/QRL bounded by Cryogenics/QRL
Lumped
models
- a closed surface S (control Lumped
models
0-D model: main
characteristics
surface) crossed by the fluid 0-D model: main
characteristics
• valid ∀ fixed S and Ω, and any point in flow domain
Bond graph modeling
overview flow Bond graph modeling
overview
• internal variation of U depends only of fluxes through S ,
and pollutant
R and pollutant not inside
sources • Total amount of U inside Ω: Ud Ω with variation per unit sources
R Ω • no derivative/gradient of F : may be discontinuous and
Simulation results
∂ Simulation results
time ∂t Ω Ud Ω
Volume- Volume- admit shock waves
transport
• Flux = amount of U crossing S per unit time: transport
~ with d S
~ · dS ~ outward normal, and net total ⇒ relate to conservative numerical scheme ar the discrete
coefficients
Fn dS = F coefficients
Parameter estimation H
~ · dS ~ > 0 when entering the domain)
~ (F
Parameter estimation
level (e.g. conserve mass)
Time-delays
contribution − S F Time-delays
• ContributionHof volume and surface sources:
~ S · dS
~
R
Ω
QV d Ω + S
Q
Air flow Air flow

modeling in modeling in Convection-diffusion form of a convection law
Flux = convective transport + molecular agitation (even at rest)
• Convective flux:
Fluid dynamics
Differential form of a conservation law R Fluid dynamics
H
~ · dS
~ = ∇~ · Fd
~ Ω as: • amount of U carried away or transported by the flow
General form of a
Obtained using Gauss’ theorem S F General form of a
~ C = U~v
conservation law
Convection-diffusion
Ω conservation law
(velocity ~v ): F
Euler and Euler and ~ is the local
• for fluid density U = ρ, local flux through d S
Navier-Stokes
∂U ~ ~ ~ S ⇔ ∂U + ∇
~ ·Q ~ · (F ~ S ) = QV
~ −Q
Navier-Stokes
~ = dm
Firn example
+ ∇ · F = QV + ∇ Firn example mass flow rate: ρ~v · d S ~ (kg/s)
Cryogenics/QRL
∂t ∂t Cryogenics/QRL
• for U = ρu (u the quantity per unit mass),
Lumped Lumped
models models
~C · dS
F ~ = ρu~v · d S
~ = ud m~
characteristics
• the effective flux (F ~ S ) appear exclusively under the
~ −Q characteristics • Diffusive flux:
overview
gradient operator ⇒ way to recognize conservation laws
overview
• macroscopic effect of molecular thermal agitation
• from high to low concentration, in all directions,
sources • more restrictive than the integral form as the flux has to be sources proportional to the concentration difference
Simulation results
differentiable (excludes shocks) Simulation results
~ D = −κρ∇
• Fick’s law: F ~ u, where κ is the diffusion coefficient
Volume- Volume-
averaging and • fluxes and source definition provided by the quantity U averaging and (m2 /s)
transport transport
coefficients considered coefficients • Provides the transport equation:
∂ρu ~ ~ · (κρ∇
~ u) + QV + ∇ ~S
~ ·Q
Time-delays Time-delays
+ ∇ · (ρ~v u) = ∇
Conclusions Conclusions ∂t
⇒ Backbone of all mathematical modeling of fluid flow
phenomena
Air flow Air flow
modeling in Euler and Navier-Stokes equations modeling in
deep wells deep wells Example: solving the air continuity in polar firns and ice
E. Witrant • From the conservation of mass, E. Witrant
cores
Fluid dynamics momentum and energy: Fluid dynamics
From poromechanics, firn =
General form of a
conservation law
General form of a
conservation law
system composed of the ice
 ρ  ρ~v
     
∂ 0
Euler and

 ρ~v +∇· ρ~v T ⊗ ~v + p I − τ
 
 =  0

 ,
Euler and
lattice, gas connected to the
Firn example ∂t ρE ρ~v H − τ · ~v − k ∇T q̇ Firn example surface (open pores) and gas
trapped in bubbles (closed pores).
Lumped
models
with shear stress (Navier-Stokes Lumped
models Air transport is driven by:
0-D model: main
characteristics
only) 0-D model: main
characteristics
Bond graph modeling Bond graph modeling ∂[ρice (1 − ǫ)]
 λ  + ∇[ρice (1 − ǫ)~v ] = 0
   
overview
 τxx
   ux  overview
 ∂t
Fans, rooms  τxy  =  µ  (∇ · ~v ) + 2µ  0  Fans, rooms
and pollutant
τyy
    and pollutant
∂[ρogas f ]
sources
λ vy sources
+ ∇[ρogas f (~v + w
~ gas )] = −~r o →c
Simulation results Simulation results
∂t
Volume- Volume-
averaging and and viscosity [Stokes & Sutherland] averaging and ∂[ρcgas (ǫ − f )]
transport transport + ∇[ρcgas (ǫ − f )~v ] = ~r o →c
coefficients !3/2 coefficients ∂t
Parameter estimation 2 µ T Tsl + 110 Parameter estimation
Time-delays λ = − µ and = . Time-delays
with appropriate boundary and Scheme adapted from [Sowers
Conclusions 3 µsl Tsl T + 110 Conclusions
initial conditions. et al.’92, Lourantou’08].
• Discrete boundary conditions
(potential numerical instabilities).
Air flow Air flow
modeling in Firn example: from distributed to lumbed dynamics modeling in
e.g. eig(A) for CH4 at NEEM with dt = 1 month
E. Witrant • Defining q = ρcgas (ǫ − f ) and considering the 1-D case, we E. Witrant
Fluid dynamics
have to solve Fluid dynamics
∂q ∂
General form of a
+ [qv ] = r o →c General form of a
Imaginary part of eigenvalues

Convection-diffusion ∂t ∂z Convection-diffusion
3
FOU
Euler and Euler and 2 Central
Navier-Stokes
Firn example
• Approximate ∂[qv ]/∂z , i.e. on uniform mesh: Navier-Stokes
Firn example 1
FOU + central
FOU + 2nd order
ui −ui −1 FOU + 3rd order
• backward difference: (uz )i = ∆z
∆z + 2 (uzz )i
0
Lumped ui +1 −ui −1 2 Lumped
models • central difference: (uz )i = 2∆zi
− ∆6z (uzzz )i models
−1
0-D model: main 0-D model: main −2

characteristics • other second order: characteristics
−3
(uz )i = ui+1 +3u4i −∆5zui i−1 +ui−2 + ∆12z (uzzz )i
2
∆z 3
Bond graph modeling
overview − 8
(uzzzz )i Bond graph modeling
overview
−2 −1.5 −1 −0.5 0
Real part of eigenvalues
0.5 1 1.5 2
2u +3u −6u +u ∆z 3
Fans, rooms • third order: (uz )i = i+1 6i∆zi i−1 i−2 − 12
(uzzzz )i Fans, rooms
0.5

sources sources 0
Simulation results
• Provides the computable lumped model: Simulation results −0.5
Volume- Volume- −1
averaging and dq averaging and −1.5
transport = Aq + r o →c transport −2
coefficients dt coefficients
−2.5
Time-delays Time-delays −3
• The choice of the discretization scheme directly affects the 0 20 40 60 80
Depth (m)
100 120 140 160 180
definition of A and its eigenvalues distribution: need to
check stability and precision!
Air flow Air flow
modeling in modeling in e.g. eig(A) for CH4 at NEEM with dt ≈ 1 week, zoom
e.g. eig(A) for CH4 at NEEM with dt ≈ 1 week

General form of a General form of a 2 FOU

3
Central
Convection-diffusion Convection-diffusion FOU + central
FOU 1
Euler and 2 Central Euler and FOU + 2nd order
Navier-Stokes FOU + central Navier-Stokes FOU + 3rd order
Firn example 1 FOU + 2nd order Firn example
0
Cryogenics/QRL FOU + 3rd order Cryogenics/QRL
0
Lumped −1 Lumped −1
models models
0-D model: main −2 0-D model: main −2
−3
Bond graph modeling −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 Bond graph modeling
overview overview −15 −10 −5 0 5
Real part of eigenvalues −3
x 10
0.5

sources 0 sources 1

Simulation results −0.5 Simulation results
0.5
Volume- −1 Volume- 0
averaging and −1.5 averaging and
−0.5
transport −2
transport
coefficients coefficients −1
−2.5
−1.5
Time-delays −3 Time-delays
0 20 40 60 80 100 120 140 160 180
−2
Conclusions Depth (m) Conclusions
−2.5
−3
0 20 40 60 80 100 120 140 160 180
Depth (m)
Air flow Air flow

deep wells e.g. Impulse response (Green’s function) for CH4 at deep wells e.g. Impulse response (Green’s function) for CH4 at
E. Witrant NEEM with dt = 1 month E. Witrant NEEM with dt ≈ 1 week
Convection-diffusion 1.001 Convection-diffusion 1.001
Euler and FOU Euler and FOU
1 1
Norm of Green fct
Norm of Green fct

Navier-Stokes Central Navier-Stokes Central
FOU + central FOU + central
Firn example 0.999 Firn example 0.999
FOU + 2nd order FOU + 2nd order
Cryogenics/QRL FOU + 3rd order Cryogenics/QRL FOU + 3rd order
0.998 0.998
Lumped 0.997 Lumped 0.997

models 0.996 models 0.996
0.995 0.995
Bond graph modeling 0.994 Bond graph modeling 0.994
overview 0 20 40 60 80 100 120 140 160 180 overview 0 20 40 60 80 100 120 140 160 180
Depth (m) Depth (m)
and pollutant 25 and pollutant 25
sources sources
20 20
σ of Green fct
σ of Green fct
Volume- 15 Volume- 15
10 10
transport transport
coefficients coefficients
5 5
Time-delays 0 Time-delays 0
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
Conclusions Depth (m) Conclusions Depth (m)
Air flow Air flow
deep wells deep wells Method
E. Witrant Example 2: Travelling wave modeling cryogenics [Bradu, E. Witrant Assumptions:
Gayet, Niculescu, W’10]
Fluid dynamics Fluid dynamics • model using Euler equation
General form of a
conservation law
LHC sector 5-6 with the main cooling loops for the General form of a
conservation law
• flux according to the x direction only (the main flow
superconducting magnets: Convection-diffusion
Euler and
Navier-Stokes
Euler and
Navier-Stokes direction) : V = Vx and M = ρ · Vx ;
Cryogenics/QRL Cryogenics/QRL • straight line. The QRL curvature (radius of curvature of
Lumped
models
Lumped
models
4.3 km) has a negligible impact on the flow;
characteristics characteristics • in operational conditions, the kinetic component can be
Bond graph modeling
overview
Bond graph modeling
overview ~ |2 << P , which implies that
neglected: ρ · |V
ρ·V~T ⊗V ~ + P · I ≈ P · I.
sources sources
Simulation results Simulation results Euler equation expressed in 1D as:
Volume- Volume-
averaging and averaging and ∂X (x , t ) ∂X (x , t )
transport transport
+ F (X ) · = Q (x , t )
coefficients
coefficients
∂t ∂x
Conclusions Conclusions where X = [ ρ M E ]T is the state vector, F is the Jacobian

flux matrix and Q = [ 0 0 q ]T is the source vector.
Air flow Air flow

modeling in modeling in Temperature transport
Impact of convection heat, hydrostatic pressure and friction
Method (2): space discretization pressure drops:
Fluid dynamics
General form of a
From the Jacobian (empirical formulation for the helium internal Fluid dynamics
General form of a
conservation law
energy) conservation law
Euler and Euler and
 
Firn example
Cryogenics/QRL
 0 1 0  Firn example
Cryogenics/QRL
2
(γ−3)V
 
Lumped F =  2 − u0 γ̂ (3 − γ)V γ̂  Lumped
models γVE γE 2 models
γ̂V 3 − − γ̂( 3V2 + u0 ) γV
 
characteristics ρ ρ characteristics
overview overview
Fans, rooms
we obtain the state-space matrices from: Fans, rooms
sources sources
Ai (Xi ) Bi (Xi ) Ci (Xi )
Simulation results
Ẋi (t ) + Xi (t ) + Xi −1 (t ) + Xi +1 (t ) = Qi (t ) Simulation results
Volume- ∆x ∆x ∆x Volume-
transport h i transport
coefficients where i denotes the value at xi and Xi = ρi Mi Ei coefficients
Time-delays + add the interconnections with external inputs in Qi Time-delays
Air flow Air flow
modeling in modeling in 0-D model: main characteristics
deep wells
From Euler equations to lumped deep wells
models
General form of a
conservation law
General form of a
conservation law
• Energy approach: equivalent to finite
(0-D) Control-oriented model
Euler and
Euler and
volume method → physically consistent
Firn example • Non-dimensional modeling has an increasing use in Firn example averaging of the dynamics.
design, validation and tuning of control laws. • Hypotheses:

Lumped Lumped
models
• Two main advantages: models 1 only static pressure considered in
characteristics
• integration of as much physical properties as possible
characteristics energy conservation;
overview overview 2 impulsive term negligible compared to
(avoid data mapping);
Fans, rooms
• reduced computation: close to real-time, ≈ 10× slower in
Fans, rooms pressure in momentum conservation;
sources sources 3 momentum dynamics simplified using
worst cases.
Saint-Venant equations → algebraic
Volume- • Flow/effort model inferred from Euler equations: Volume-
relationship.
transport transport
coefficients
 ρ 
  
ρ~v
  
  0  coefficients ⇒ Algebro-differential model with
∂ 
 Parameter estimation
 ρ~v  + ∇ ·  ρ~v T ⊗ ~v + p I numerically robust ODE description.

 =  0  .
Time-delays
   Time-delays
∂t 
Conclusions
ρE ρ~v H Conclusions
 
0
Air flow
Bond graph modeling overview [V. Talon, PhD’04] Air flow
modeling in
deep wells
modeling in
deep wells
Bond graph model (2)
Rt
E. Witrant • Defining the mass flow Qm (t ) = L /S 0 ∆p (s )ds and the E. Witrant
• Power conjugate variables:
Fluid dynamics enthalpic flow Qh = Qm cp T , the conservation equations Fluid dynamics
conservation law write as     Conservation equation
conservation law Effort Flow Relation
 m   Qm,i − Qm,o  Convection-diffusion
Mass mass m flow rate Qm effort storage C
Euler and
Navier-Stokes d  Euler and
Navier-Stokes
Firn example
 Qm  =  S /L (pi − po )  . Firn example
Momentum (hydraulic) pressure p flow rate Qm flow storage I
Cryogenics/QRL dt  Cryogenics/QRL Energy (thermal) internal energy E enthalpic flow Qh effort storage C
U Qh ,i − Qh ,o
  
Lumped Lumped
models models
0-D model: main
• Pressure and temperature obtained from first Joule law 0-D model: main
Bond graph modeling (U = mcv T ) and the perfect gas relationship: Bond graph modeling
overview overview

and pollutant γ−1U U and pollutant
sources T= and p = (γ − 1) sources
Simulation results
R m v Simulation results
Volume- Volume-
averaging and • Pressure losses from Bernoulli’s equation (supposing averaging and
2
transport
coefficients incompressibility) as ∆P = ζ Qm /(2ρS 2 ), where ζ is a transport
coefficients
Time-delays
friction coefficient. Parameter estimation
Time-delays
γ+1
2
Conclusions
p 2(γ−1) Conclusions
• Saint Venant Qm = ρCd S γRT γ+1 : momentum
represented as a resistive element.
Air flow Air flow
Simulation results
deep wells
Fans, rooms and pollutant deep wells
Simulator properties:
sources • ventilation shafts ≈ 28 control volumes
(CV), 3 extraction levels
General form of a
conservation law • Additional features: friction losses and pollutant tracking. General form of a
conservation law
• regulation of the turbine and fans
Euler and
Navier-Stokes
• Turbine and fans: Euler and
Navier-Stokes
Firn example Firn example • flows, pressures and temperatures
Cryogenics/QRL • compressors that generate a Cryogenics/QRL
measured in each CV
Lumped flow depending on a pressure Lumped
models models
0-D model: main gradient and a rotational speed; 0-D model: main
• Computation time 34× faster than
Bond graph modeling Bond graph modeling real-time
overview • characteristics depend on overview
Fans, rooms specification maps → Fans, rooms Case study:

sources
• enthalpic flow Qh = Qm cp T and sources • 1st level fan not used (natural airflow), 2nd
Volume-
output temperature is obtained Volume-
operated at 1000 s (150 rpm) and 3rd runs
averaging and as ( " # ) averaging and continuously (200 rpm)
transport P γ−γ 1 transport
1
coefficients
To = Ti ηc
o
Pi
−1 +1 ; coefficients
• CO pollution injected in 3rd level
• measurement of flow speed, pressure,
Conclusions ⇒ capacitive and resistive Conclusions
elements. temperature and pollution at the surface

• Room included as (mostly) inertia elements and extraction levels
Air flow Air flow

modeling in Simulation results (2) modeling in

Convection-diffusion Convection-diffusion Simulation results (3)
Euler and Euler and
Firn example Firn example • Physical and chemical airflow properties:
• pressure losses = energy losses;
Lumped Lumped
models models • rooms ventilation rate = physical interconnections and
0-D model: main (a) Pressure losses in the inflow shaft (b) Extraction rooms ventilation rate 0-D model: main
characteristics characteristics importance of a global control strategy;
overview overview • temperature: geothermal effect and fans compression;
Fans, rooms Fans, rooms • pollutant transport: time-delay effect;
sources sources • Computation time 34× faster than real-time.
Volume- Volume-
transport transport
coefficients coefficients
(c) Extraction rooms temperature (d) CO pollutant concentrations in the

exhaust shaft
Air flow Air flow
deep wells
Volume-averaging and estimation deep wells
of the transport coefficients Observer-based online parameter estimation
Fluid dynamics Fluid dynamics Theorem (parameter estimation for affine PDE): Consider the
General form of a
conservation law
Volume-averaged model General form of a
conservation law class of systems
Euler and Euler and
pt = A(p , px , pxx , u, ϑ)ϑ

Navier-Stokes • Volume-averaged impact of momentum and density: Navier-Stokes 


a1 px (0, t ) + a2 p (0, t ) = a3



I I  a p (L , t ) + a p (L , t ) = a

1 1 4 x 5 6
Lumped
models
M̄ (t ) M (v , t )dv and ρ̄(t ) ρ(v , t )dv , Lumped
models
0-D model: main
V V V V 0-D model: main with distributed measurements of p (x , t ) and for which we
Bond graph modeling Bond graph modeling want to estimate ϑ. Then
overview • Energy losses = pressure losses (friction and exhausts): overview
Fans, rooms Fans, rooms ||p (x , t ) − p̂ (x , t )||22 = e −2(γ+λ)t ||p (x , 0) − p̂ (x , 0)||22

sources q̇ (x , t )R /cv = s (x , t ) + r (t )p (x , t ), sources
Simulation results Simulation results if 
Volume- Volume-

 p̂t = A(p̂ , p̂x , p̂xx , u, ϑ̂)ϑ̂ + γ(p − p̂ )
• Give the transport model with boundary (controlled) input:

a1 p̂x (0, t ) + a2 p̂ (0, t ) = a3




transport transport
a4 p̂x (L , t ) + a5 p̂ (L , t ) = a6


coefficients ( coefficients 

p̃t = c (t )p̃x + r (t )p̃ + s (x , t ),

ϑ̂ = A(p̂ , p̂x , p̂xx , u, ϑ̂)† [pt + λ(p − p̂ )]


Conclusions
p̃ (0, t ) = pin (t ) Conclusions
⇒ Given distributed measurements, estimate transport

coefficients and set feedback pin (t )
Air flow Air flow
modeling in
Example: comparison with gradient-descent algorithm modeling in Time-delay model
Consider the convective-resistive flow:
pt = d (t )pxx + c (t )px + r (t )p + s (t )pext (x , t )
Fluid dynamics
General form of a
Fluid dynamics
General form of a
pt (x , t ) − c (t )px (x , t ) = r (t )p (x , t )
with p (0, t ) = u(t ), p (x , 0) = ψ(x ). Applying the method of

−0.8
Euler and Euler and
−0.9 0.65
characteristics with the new independent variable θ as
Convection c(t)
Diffusion d(t)
Cryogenics/QRL −1 0.6 Cryogenics/QRL
Lumped −1.1 0.55 Lumped

models
−1.2 0.5
models p (θ) p (x (θ), t (θ))
−1.3 0.45
Bond graph modeling
overview
0 50 100 150 200 250 300 350 400
Gradient−based
450 500
0.4 estimation
0 50 100 150 200 250 300 350 400 450 500
Bond graph modeling
overview It follows that (solution including time axis)
−0.5 Reference 1
Fans, rooms Fans, rooms θf t
Observer−based estimation
−0.6 0.95
Z ! Z
sources
−0.7 0.9
sources p (L , t ) u(t − θf )exp r (η)d η with L = − c (η)d η
Resistance r(t)
−0.8 0.85
0 t −θf
Source s(t)

−0.9 0.8
−1 0.75
Volume- Volume- RL
The average pressure ξ(t ) 0 p (η, t )d η is provided by the
−1.1 0.7
−1.2 0.65
transport transport
−1.3
coefficients −1.4
0.6
0.55
coefficients Delay Differential Equation
−1.5 0.5
0 50 100 150 200 250 300 350 400 450 500 0 50 100 150 200 250 300 350 400 450 500
Time (s) Time (s) " Z θf !#
d
ξ = −c (t ) u(t ) − u(t − θf )exp r (η)d η + r (t )ξ(t )
dt 0
⇒ very accurate results, need to add a filter.
Air flow Air flow
deep wells
Conclusions deep wells
Main references
• Anderson, J.: Fundamentals of Aerodynamics, McGraw-Hill
Fluid dynamics Fluid dynamics Companies, 1991.
General form of a
conservation law
General form of a
conservation law
• C. Hirsch, Numerical Computation of Internal & External Flows: the
Fundamentals of Computational Fluid Dynamics, 2nd ed.
Euler and Euler and
Navier-Stokes Navier-Stokes Butterworth-Heinemann (Elsevier), 2007.
Firn example
Cryogenics/QRL
• Physical model provided by a nonlinear coupled partial Firn example
Cryogenics/QRL
• E. Witrant, K. Johansson, and the HynX team, “Air flow modelling in
Lumped differential equation, but provides physical understanding Lumped deep wells: application to mining ventilation”, IEEE CASE 2008, USA,
models models August 23-26, 2008.
0-D model: main • Physics can be partly conserved using energy-based 0-D model: main
http://www.gipsa-lab.grenoble-inp.fr/˜ e.witrant/papers/08_CASE0083_FI.pdf
Bond graph modeling
overview
model reduction Bond graph modeling
overview • E. Witrant and N. Marchand, ”Modeling and Feedback Control for Air
Fans, rooms • Volume-averaging provides the model structure for Fans, rooms Flow Regulation in Deep Pits”, in Scientific Monographs and Text
and pollutant and pollutant Books, Cambridge Scientific Publishers, 2008.
sources estimators / feedback strategies sources
http://www.gipsa-lab.grenoble-inp.fr/˜ e.witrant/papers/11_CSP_Witrant.pdf
Volume- • Flow convection represented by time-delays Volume-

• E. Witrant, S.I. Niculescu, “Modeling and Control of Large Convective
averaging and averaging and Flows with Time-Delays”, Mathematics in Engineering, Science and
transport transport
coefficients coefficients Aerospace, Vol 1, No 2, 191-205, 2010.
Parameter estimation Parameter estimation http://www.gipsa-lab.grenoble-inp.fr/˜ e.witrant/papers/10_Witrant_MESA.pdf
• B. Bradu, P. Gayeta, S.-I. Niculescu and E. Witrant, ”Modeling of the
very low pressure helium ow in the LHC Cryogenic Distribution Line
after a quench”, Cryogenics, vol. 50 (2), pp. 71-77, Feb. 2010.
http://www.gipsa-lab.grenoble-inp.fr/˜ e.witrant/papers/09_Simu_QRL.pdf
Extraction Extraction
rooms air rooms air
quality model quality model
Stratified flows
Aerodynamics Aerodynamics
of chambers of chambers
Turbulence Turbulence
Turbulent viscosity: Turbulent viscosity:
different models different models
Velocity fields Velocity fields
Fluid statics
Hydrostatic eq.
Mining Ventilation Control Fluid statics
Hydrostatic eq.
Buoyancy force Buoyancy force
Plumes in Plumes in
stratified flows stratified flows
Plume & fountain
Total ventilation rate
Lesson 5: Extraction rooms air quality model Plume & fountain
Elevated source Elevated source
Experiments Experiments
Pollutants Emmanuel WITRANT Pollutants

• Unknown environements
profile profile
Basic equations emmanuel.witrant@ujf-grenoble.fr Basic equations
Room model Room model • WSN automation
Fans and tubes Universidad Pedagógica y Tecnológica de Colombia, Fans and tubes
• Global conservation constraints - 0/1 D modeling
Tarpauline tube Tarpauline tube
Fan Sogamoso, April 3, 2013 Fan
WSN WSN
rooms air rooms air
quality model
Stratified flows and forced plumes quality model
Outline
Aerodynamics • Air/mist experiment for qualitative behavior Aerodynamics

Turbulent viscosity:
different models
different models 1 Gas dynamics in chamber-like mine workings
Fluid statics Fluid statics

Hydrostatic eq. Hydrostatic eq.
2 Fluid statics: buoyancy force
Plumes in Plumes in
Plume & fountain Plume & fountain
3 Stratified flows and forced plumes
Total ventilation rate Total ventilation rate
4 Constrained shape of the pollutants profile
profile profile
Basic equations Basic equations
Room model Room model 5 Peripheral dynamics induced by fans and tarpauline tubes
Fans and tubes Fans and tubes
Tarpauline tube • relatively slow process Tarpauline tube
Fan Fan
WSN • clear stratification when different gravities WSN
• extendable to temperature variation effect
rooms air rooms air Turbulence in jets
quality model
Gas dynamics in chamber-like quality model
mine workings [Kalabin et al., 1990]
different models
Velocity fields
• Estimates of the velocity field and turbulent viscosity over different models
Velocity fields
Fluid statics the volume of the underground chamber Fluid statics

Buoyancy force ⇒ analysis of the scattering and entrainment of harmful Buoyancy force
Plumes in impurities by ventilation jets Plumes in

stratified flows stratified flows https://engineering.purdue.edu/Engr/AboutUs/News/Publications/EngEdge/2002/environment_spotlight_s
Plume & fountain
• Complexity from recirculation regions and breakaway flows Plume & fountain
Elevated source
Experiments
• 2-D mathematical model provided by non-steady Elevated source
Experiments
Pollutants
Navier-Stokes (turbulent & incompressible) + continuity: Pollutants
profile ! profile
Basic equations ∂u ∂u ∂u 1 ∂p 1 ∂τ11 ∂τ12 Basic equations
Room model +u +v = − + + Room model
∂t ∂x ∂y ρ ∂x ρ ∂x ∂y
Fans and tubes ! Fans and tubes
Tarpauline tube ∂v ∂v ∂v 1 ∂p 1 ∂τ21 ∂τ22 Tarpauline tube
Fan
+u +v = − + + Fan
WSN
∂t ∂x ∂y ρ ∂y ρ ∂x ∂y WSN
Conclusions ∂u ∂v Conclusions
+ = 0
∂x ∂y Offset jets [Agelin-Chaab & Tachie, 2011] (left) and jets in rooms (right)
where τij is the Reynolds stress (viscosity-dependent).
rooms air The importance of turbulence control rooms air
Turbulent viscosity νt
1 Simplest model: νt constant for recirculation regions and
Aerodynamics Aerodynamics near-wall interactions
Turbulence
Turbulence
2 Small-scale motion of subgrid scale (Smagorinskii):
Velocity fields Velocity fields effective viscosity in terms of mean flow characteristics,
Fluid statics Fluid statics numerical determination
3 (k − L ): add turbulent kinetic energy k
Plumes in Plumes in
CD k 3/2
! !
Plume & fountain Plume & fountain ∂k ∂k ∂k ∂ νt ∂ k ∂ νt ∂ k
Total ventilation rate Total ventilation rate +u +v = + +νt G +
Elevated source Elevated source ∂t ∂x ∂y ∂ x σk ∂ x ∂ y σk ∂ y L
profile profile
where G = turbulent energy generation, L = scale of
√ vortices), σk - CD = numerical
turbulence (transverse
Room model Room model

constant, νt = Cµ k L where Cµ ≈ 1.
Fan Fan 4 Standard (k − ǫ): add energy dissipation rate ǫ and solve
WSN WSN
2 transport equations
+ appropriate boundary conditions
rooms air rooms air
quality model Example: (k − ǫ) after trucks CO pollution quality model
Example: (k − ǫ) after trucks CO pollution (2)

Plumes in Plumes in
profile profile

Fan Fan
WSN WSN
rooms air rooms air
quality model
Fluid statics and buoyancy force quality model
[Anderson, 1991]
Hydrostatic equation (2)
Aerodynamics Aerodynamics Integral form:
of chambers • Def.: force applied on the fluid in statics of chambers
different models
• On an infinitesimal element: Turbulent viscosity:
different models
Z p2 Z h2
Velocity fields Velocity fields dp = −ρg dy ⇔ p2 + ρgh2 = p1 + ρgh1
Fluid statics Net pressure force Fluid statics p1 h1
Hydrostatic eq.
Buoyancy force = dp (dx dz )
Hydrostatic eq.
Buoyancy force
⇔ p + ρgh = constant
!
Plumes in dp Plumes in
stratified flows − p+ dy (dx dz ) stratified flows e.g. on the walls of a container:
Plume & fountain dy Plume & fountain
dp
Elevated source
= − (dx dy dz ) Elevated source
Experiments
dy Experiments
profile
Gravity force profile
Basic equations
= −ρ(dx dy dz )g Basic equations

Fan Fan
WSN WSN
Conclusions dp Conclusions
⇒ Sum of forces =0 = − (dx dy dz ) − ρ(dx dy dz )g
dy
Provides the Hydrostatic equation dp = −ρgdy
rooms air rooms air
quality model
Buoyancy force
quality model
Stratified flows and forced plumes
Consider a body immersed in a stagnant fluid (ρ can vary), i.e. [Liu & Linden, 2006]
Turbulent viscosity: Force on body due to Turbulent viscosity:
Velocity fields pressure: Velocity fields

Hydrostatic eq.
F = (p2 − p1 )l × 1 Hydrostatic eq. • Heat source represented as a buoyant plume in a stratified
Plumes in
Z h2 Plumes in
(layered) environment [Morton, Taylor & Turner, 1956, Yih
stratified flows
Plume & fountain
= l (1) ρgdy stratified flows
Plume & fountain
1969]
h1
Elevated source
Elevated source
• Based on entrainement coefficients [Linden, Lane-Serff &
Experiments
R h2 Experiments
Smeed, 1990]
Pollutants Note: h ρgdy = weight of a column of unit area and height Pollutants
profile 1 profile • Air “thrown” with an initial momentum and higher
Basic equations
(h1 − h2 ), and F = weight of l columns side by side. Basic equations
Room model Room model temperature settles by negative buoyancy
Fans and tubes
Consequently (Archimedes principle): Fans and tubes
Fan Fan
WSN Buoyance force on body = weight of fluid displaced by body WSN
rooms air Experiment: 1 heat source + 1-2 diffusers rooms air
Plume equations
• Flow above heat source = plume with buoyancy B , mean
Aerodynamics
of chambers
Aerodynamics
of chambers
radius b , vertical velocity w (r , z ) and reduced gravity
Turbulence
Turbulence
g ′ = g (ρ0 − ρ)/ρs ∝ (T − T0 ), where ρs = reference, ρ0 =
different models
Velocity fields
different models
Velocity fields
ambient and ρ = plume densities
Fluid statics Fluid statics • Entrainment assumption [Morton’58]: the rate of
Hydrostatic eq.
Buoyancy force
Hydrostatic eq.
Buoyancy force
entrainment at the edge of a plume ∝ vertical velocity at
Plumes in Plumes in that height:
• gaussian distributions of vertical velocity
2 2
Total ventilation rate Total ventilation rate w (r , z ) = w̄ (z )e −r /b , and mean buoyancy
Elevated source Elevated source 2 2 2
Experiments Experiments g ′ (r , z ) = g¯′ (z )e −r /λ b , with characteristic length-scales b
Pollutants Pollutants and λ
profile profile
• resulting static space distributions:
d 2 d 2 2 d 2 2 ′
(b w̄ ) = 2αb w̄ , (b w̄ ) = 2λ2 b 2 g¯′ , (λ b w̄ ḡ ) = 0
Tarpauline tube
Fan
Tarpauline tube
Fan
dz dz dz
WSN WSN
where α = entrainment coef of the plume. Can be solved
analytically.
rooms air rooms air
quality model
Fountain equations quality model
• Fountain: flow from diffuser = negatively-buoyant turbulent

of chambers jet which rises a certain height until the negative buoyancy of chambers
Turbulent viscosity: reduces its upward momentum → 0. Then the flow Turbulent viscosity:
different models different models Additional issues for the interface
Velocity fields
reverses and falls down in an annular region outside the Velocity fields
Fluid statics
rising jet [Turner 1966]. Fluid statics • 2-layers stratification hypothesis from experimental
Buoyancy force
• Non-dimensional variables based on the initial source Buoyancy force observations
Plumes in
stratified flows
volume flux Q0 , momentum flux M0 and buoyancy flux F0 Plumes in
stratified flows • penetrative entrained volume flux Qe across density
of the fountain:
Plume & fountain
Plume & fountain
interface ∝ impinging volume flux Qi at density interface:
Elevated source
Experiments
z̃ = M0−3/4 F01/2 z , b̃ = M0−3/4 F01/2 b , ũ = M01/4 F0−1/2 w̄ , g̃′ = M05/4 F0−3/2 ḡ′ Elevated source
Experiments
Qe = Qi E where E = penetrative entrainement rate
profile defining the dimensionless fluxes of volume Q̃ = b̃ 2 ũ, profile
momentum M̃ = b̃ 2 ũ2 , and buoyancy F̃ = b̃ 2 ũg˜′ :


d d d 2 2 ′
Tarpauline tube
(b̃ 2 ũ) = 2αf b̃ ũ, (b̃ 2 ũ2 ) = 2λ2 b̃ 2 g̃′ , (λ b̃ ũg˜ ) = 0 Tarpauline tube
Fan
d z̃ d z̃ d z̃ Fan
WSN WSN
Conclusions where αf = entrainment coef of the fountain. Conclusions
rooms air Fixed heat load and total ventilation rate rooms air
Elevated heat source
E. Witrant • Consider the effect of multiple diffusers of the same air E. Witrant
Aerodynamics
• For one diffuser/heat source: Aerodynamics
Turbulence
• at interface height h , plume and each fountain carry Qp Turbulence
different models
and Qi , respectively Turbulent viscosity:
different models
Velocity fields • an amount of upper layer fluid Qe is entrained back into the Velocity fields
Fluid statics lower layer by the fountain above each cooling diffuser Fluid statics
Buoyancy force ⇒ net flow rate Q = total ventilation rate through the system. Buoyancy force
Plumes in • Multiple diffusers (n): momentum divided by n if same 2 Plumes in

Plume & fountain
flux source Plume & fountain
profile profile
Fans and tubes

Tarpauline tube
Fans and tubes
Tarpauline tube
Addapt the fow computation by including the distance of the
Fan
WSN
Fan
WSN
interface from the flow origin (h − hs ):
Qp = C [B (h − hs )5 ]1/3
rooms air rooms air
quality model quality model Experimental results (2): sources at different heights
E. Witrant Experimental results E. Witrant
Aerodynamics • Dyed salt solution injected in clear rectangular tanks filled Aerodynamics
Turbulence with fresh water. Measurements at steady-state. Turbulence
Velocity fields
• Buoyancy profiles with multiple diffusers with fixed heat Velocity fields
Fluid statics load and ventilation rate: Fluid statics

Plumes in Plumes in
profile profile

Fan Fan
WSN WSN
rooms air rooms air
quality model
Constrained shape of the quality model
pollutants profile [Witrant et al., 2010]
of chambers of chambers Problem structure
different models
different models
• Division into subsystems:
• fan / tarpaulin tube / extraction room / wireless sensor
network.
• Corresponding block diagram:
Plumes in Plumes in
profile profile
Basic equations • Seek model simplifications based on measurements from Basic equations
Fans and tubes

a wireless sensor network Fans and tubes
Tarpauline tube
Fan
• Constrain the dynamics from global (0-D) conservation Tarpauline tube
Fan
WSN
laws WSN
• Use an heuristic approximation of concentration
distributions
rooms air rooms air
quality model Main variables quality model
Basic equations for the concentration profiles
• Inputs to the system:
Aerodynamics Aerodynamics • Conservation law: conservation of mass
of chambers
• ρ: air density in vertical shaft; of chambers
Turbulence • P : air pressure in vertical shaft; Turbulence
Turbulent viscosity: Turbulent viscosity: ṁj (t ) = ṁj ,in (t ) − ṁj ,out (t ) − ṁj ,chem (t )
different models • ∆H : variation of pressure produced by the fan; different models
• ṁj ,in : incoming pollutant mass rate due to the engines;
Fluid statics Fluid statics • Constitutive relationship:
Hydrostatic eq. • ṁj ,chem : mass variation due to chemical reactions between Hydrostatic eq.
components; Z hroom
Plumes in Plumes in
stratified flows
• h : time-varying number of hops in WSN. stratified flows mj (t ) = Sroom cj (z , t )dz
Plume & fountain
• Outputs from the system: Plume & fountain 0
"Z hdoor #
Experiments
• cj (z , t ) pollutants (COx or NOx ) volume concentration Experiments = Sroom cj (z , t )dz + αj (t )∆h ,
Pollutants profiles, where z ∈ [0; hroom ] is the height in the extraction Pollutants 0
profile profile
Basic equations
room; Basic equations
Room model • uavg is the average velocity of the fluid in the tarpaulin tube; Room model
and hypothesis
Fans and tubes • mj pollutant mass in the room; Fans and tubes
Tarpauline tube
• τwsn delay due to the distributed measurements and Tarpauline tube αj (t )
Fan Fan
cj (z , t ) = .
WSN
wireless transmission between the extraction room and the WSN
1+ e j (t )(z −γj (t ))
−β
Conclusions fan. Conclusions
rooms air rooms air
quality model quality model Extraction room model
• shape parameters α, β and γ chosen as the state:
Aerodynamics
of chambers
Basic equations for the concentration profiles (2) Aerodynamics
of chambers
x (t ) = [α, β, γ]T ;
Turbulence The mass conservation equation sets the shape parameters Turbulence
• time derivative from mass conservation:
different models
dynamics with: Turbulent viscosity:
different models
 α̇j (t )
"Z hdoor #  

ṁj (t ) ≈ Sroom ċj (z , t )dz + α̇j (t )∆h , Ej  β̇j (t )  = ṁj ,in (t ) − Bj ufan (t − τtarp ) − Djk , with
 
0
γ̇j (t )

Z hdoor
Plumes in Qout Plumes in
stratified flows
ṁj ,out (t ) ≈ cj (z , t )dz , stratified flows
hdoor 0
Plume & fountain Plume & fountain  .
"Z hdoor # Total ventilation rate

  . .. ..  

ṁj ,chem (t ) = Sroom νjk cj (z , t )ck (z , t )dz + νjk αj (t )αk (t )∆h ,  . . .   T 
 ∂Cj ,i ∂Cj ,i ∂Cj ,i   ∆h
Elevated source Elevated source 
  
0  +  0
Ej Sroom Vint   
Pollutants Pollutants   ∂αj ∂βj ∂γj  
  
profile where Qout = St ηufan (t − τtarp ) is the volume rate of flow profile 
 ..

.. .. 
 0 


Room model leaving the room = Qin and νjk = −νkj is the chemical reaction Room model
. . .
 .  ..
rate
   
Fans and tubes Fans and tubes  .     
Tarpauline tube Tarpauline tube  .  .
1
   
Fan Fan
Bj Vint  Cj ,i  × Starp ν, Djk = Sroom Vint  ηjk ,i Cj ,i Ck ,i + η α α ∆ h
     
WSN WSN
jk j k
hdoor  
 .   
.. 
..

 
.
rooms air rooms air
quality model
Peripheral dynamics (fans, tubes quality model Physical delay in the tarpauline tube
and network) [Witrant et al., 2010] • Simplified flow transport equation:
of chambers of chambers " # " # " #
∂ u u r ∂ u
different models
different models + = 0,
Velocity fields Velocity fields ∂t Ttarp γTtarp u ∂x Ttarp
where r = 287.045 J /kg .K is the air gas constant and
Buoyancy force
Simplified model obtained considering: Buoyancy force
Plumes in Plumes in
γ = 1.4 is the ratio of specific heat coefficients.
stratified flows • Physical delay + energy losses for the tarpauline tube stratified flows
• Characteristic velocities v (x , t ) of the traveling waves:
Total ventilation rate • A locally regulated fan Total ventilation rate
−v + u r
q

Experiments
• Network delay + packet losses for the wireless sensor Experiments det
γTtarp −v + u = 0 ⇔ v1,2 (x , t ) = u(x , t ) ± r γTtarp (x , t ) .
profile network profile
• Down-flow time-delay approximated as:
Room model Room model Lt
τtarp (t ) ≈ q , where Lt is the length of the
ū(t ) + r γT̄tarp (t )
Fan Fan
WSN WSN tarpaulin tube, ū(t ) and T̄tarp (t ) are the space-averaged
Conclusions Conclusions flow speed and temperature, respectively.
rooms air Energy losses in the tarpauline tube rooms air
E. Witrant • Loss of airflow energy due to curves (concentrated losses E. Witrant
Aerodynamics
ξc ) and length (distributed losses ξd ) Aerodynamics
of chambers of chambers Energy losses in the tarpauline tube (2)
Turbulence • Conservation of energy (Bernoulli’s equation, Turbulence
• Supposing 2∆H /(ρufan2

different models incompressible):
different models ) << 1 (reasonable for
incompressible flow) and solving for uroom (t ):
Fluid statics 2 2 Fluid statics
uroom u
− ρ fan ,
Pfan − Proom = ρ∆H = ξd + ξc + ρ
Buoyancy force
2 2
Buoyancy force
uroom (t ) ≈ η × ufan (t − τtarp ),
Plumes in Plumes in s
stratified flows stratified flows !2
Plume & fountain Plume & fountain σLt 1
Total ventilation rate 2
• Darcy-Weisbach eqn: ξd = Lt ρuavg f /(2Dt ), where Dt = Total ventilation rate η 1− ,
Elevated source Elevated source Dt 1.82 ln Re − 1.64
Experiments tube diameter, f = friction losses, uavg = average velocity of Experiments
Pollutants the fluid in the tube. Pollutants

profile profile where η < 1 represents the energy losses associated with
• ξc introduced by the curves considering an effective length
the tarpaulin tube and Re ≈ 67280Dt × ufan .
Room model
Le = σLt , σ > 1 depends on curvature and diameter of the Room model
2
Fans and tubes
Tarpauline tube
tube ⇒ ξd + ξc = Le ρufan f /(2Dt ) Fans and tubes
Tarpauline tube
Fan • friction for turbulent flow (i.e. high Reynolds number Re ): Fan
WSN WSN
Conclusions !2 Conclusions
−2 1 Le ufan
f = (1.82 log10 Re − 1.64) ⇒ ξ = ξd +ξc = ρ
2 Dt 1.82 log10 Re − 1.64
rooms air Locally regulated fan rooms air Wireless sensor networks
Alternative: three-phase
E. Witrant E. Witrant Control to actuator signal
Typical models involve complex asynchronous motor model
Aerodynamics
datamaps: with fan as load to determine Aerodynamics
of chambers of chambers Path 2
Turbulence
the fan power consumption Turbulence
Turbulent viscosity: Turbulent viscosity: Control System
different models
Velocity fields
[Krause, 1986]. different models
Velocity fields
Path 1
Fluid statics Main principle: Fluid statics
Hydrostatic eq.
Buoyancy force • Let n2 be the speed of the
Hydrostatic eq.
Buoyancy force
Sensors to control signal
Plumes in motor, then: Plumes in
Plume & fountain n2 = 60pf1 (1 − s ) where f1 = Plume & fountain Network characteristics: Delay associated with each
Elevated source supply frequency and p = Elevated source • time-varying channel and node i :
number of motor winding network topology; αi : time to wait before sending
profile poles. profile • dynamic selection of h (t ) a packet (random);
hops; F : propagation and
Fans and tubes
• slip factor of the motor is defined as s (n1 − n2 )/n1 , Fans and tubes • next node ensures progress transmission;
Tarpauline tube
Fan
where n1 is the speed of the rotating magnetic field. Tarpauline tube
Fan toward destination; βi : Automatic Repeat reQuest
WSN
• Local feedback control loop set by varying the supply WSN
• i.e. random sleep to save (ARQ): retransmission;
frequency while maintaining a constant voltage-frequency energy. h (t )
X
ratio: scalar V/Hz control ⇒ τ(t ) = h (t )F + (αi + βi )
i =1
rooms air Clusters and dynamic selection of hops rooms air
1, 2 or 3 clusters
Network setup and experimental measurements
Aerodynamics Emitter Receiver Aerodynamics
of chambers tk+1 of chambers
tk
Turbulent viscosity: (Clusters + 1) hops Turbulent viscosity: • 20 packets/s
Fluid statics
• Randomized Protocol
Hydrostatic eq.
- Sleeping strategy depending selection within a cluster, Fluid statics
Hydrostatic eq.
Buoyancy force
on traffic and network - High freq. = transmissions Buoyancy force
Plumes in Plumes in
stratified flows
conditions, between nodes. stratified flows
Plume & fountain
- Optimize energy Plume & fountain
Elevated source
consumption in a clustered Elevated source • Tmotes Sky nodes
Experiments 300 Experiments
environment [Bonivento & • radio controller

profile 250
profile Chipcon CC2420
al.’06],
(2.4GHz)
Room model
⇒ Dynamical organization in 200 Room model
• IEEE 802.15.4
Tarpauline tube
clusters [P.G. Park MS’07]. 150
Tarpauline tube
Fan
WSN
Three possible behaviors: 100
Fan
WSN
Conclusions - Low freq. = changes in the Conclusions

50
number of clusters,
- Medium freq. = node 0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
rooms air rooms air
quality model
Conclusions quality model
References
1 G.V. Kalabin, A.A. Baklanov and P.V. Amosov, “Calculating the Aerogas
Aerodynamics Aerodynamics Dynamics of Chamber-Like Mine Workings on the Basis of
Mathematical Modeling”, Mine Aerodynamics, 1990.
different models
different models 2 M. Agelin-Chaab, M.F. Tachie Characteristics and structure of turbulent
3D offset jets International Journal of Heat and Fluid Flow, Vol. 32(3),
Fluid statics • Aerodynamics in chambers driven by turbulence vs. main Fluid statics pp. 608-620, June 2011.
Buoyancy force flow Buoyancy force
http://dx.doi.org/10.1016/j.ijheatfluidflow.2011.03.008
Plumes in Plumes in 3 Anderson, J.: Fundamentals of Aerodynamics, McGraw-Hill
stratified flows • Plumes and fountains lead to a stratified distribution stratified flows Companies, 1991.
Total ventilation rate • Simplified grey-box model of the 2-layers profile Total ventilation rate 4 Liu Q, Linden P. The fluid dynamics of an underfloor air distribution
system. Journal of Fluid Mechanics, 554:323-341, 2006.
Experiments
• Periferal dynamics included with energy losses (tube), Experiments
Pollutants Pollutants 5 E. Witrant, A. D’Innocenzo, G. Sandou, F. Santucci, M. D. Di Benedetto,

profile time-delay (tube + network) and locally regulated fan profile A. J. Isaksson, K. H. Johansson, S.-I. Niculescu, S. Olaru, E. Serra, S.
Room model Room model Tennina and U. Tiberi, ”Wireless Ventilation Control for Large-Scale
Fans and tubes Fans and tubes Systems: the Mining Industrial Case”, International Journal of Robust
Tarpauline tube Tarpauline tube and Nonlinear Control, vol. 20 (2), pp. 226 - 251, Jan. 2010.
Fan Fan
WSN WSN 6 E. Witrant, P. Gun Park, M. Johansson, C. Fischione and K.H.
Conclusions Conclusions Johansson, ”Control over wireless multi-hop networks”, IEEE
Conference on Control Applications, Singapore, Oct. 2007.
http://www.gipsa-lab.grenoble-inp.fr/˜ e.witrant/papers/07_IeeeCCA.pdf
Control Control
strategies for strategies for
mining
ventilation
mining
ventilation
Motivation: 3 control problems in
mining ventilation
network network
Controls in all Controls in all
branches branches
Controls in co-tree Controls in co-tree
only
In deep wells
Mining Ventilation Control only
In deep wells
Fast MPC Fast MPC
Time-delay approach Time-delay approach
PDE control PDE control
Extraction Lesson 6: Principals control strategies for mining Extraction

rooms rooms
MPC vs. hybrid
control
ventilation MPC vs. hybrid
control
Hybrid Control Hybrid Control
Strategy Strategy
Safety Control
Example
Emmanuel WITRANT Safety Control
Example
Comparison
emmanuel.witrant@ujf-grenoble.fr Comparison
Universidad Pedagógica y Tecnológica de Colombia,

Sogamoso, April 4, 2013
Control Control
mining
ventilation
Outline mining
ventilation
Control of the ventilation network
[Hu et al. 2003]
network network
branches
Controls in co-tree
branches
Controls in co-tree
• Historically based on multivariable linear models (allows
only only
1 Nonlinear control of the ventilation network for classical control)
In deep wells In deep wells
Fast MPC Fast MPC • New objective: exploit non-linear tools to prevent the
PDE control 2 Distributed dynamics control in deep wells PDE control sensitivity to operating conditions
rooms rooms • Two different control objectives:
MPC vs. hybrid
control 3 Predictive and hybrid control of the extraction rooms MPC vs. hybrid
control 1 actuation in all branches of the network and global
Strategy Strategy regulation
Safety Control Safety Control
Example Example
2 actuation only in branches not belonging to the tree of the
Comparison Comparison
graph (independent, link branches) and regulation around
Conclusions Conclusions the operating point
• System peculiarity: control inputs always multiplied by
quadratic nonlinearities (cf. Lesson 3)
Control Control
strategies for
mining
Design with controls in all branches strategies for
mining
ventilation
• Based on the link/tree
ventilation Theorem: closed-loop system properties and stability
partitionning (resp. c/a) with For the system described by
network resistance (Rc and Ra ) and network
Controls in all Controls in all Q̇ = −K (I − YRQ )QD2 R + KYQ Q + KYd d
branches fan pressure drop (d ) branches
Controls in co-tree
only
Controls in co-tree
only H = YRQ QD R + YQ Q + Yd d
• Ra and d are auxiliary inputs
Fast MPC (not necessary but improve Fast MPC
and the previous control laws, the following results hold:
PDE control efficiency) and Rc is sufficient PDE control
T T T
http://www.geeksforgeeks.org/applications-of- i. H (t ) Hr = [Hcr , Har ]
Extraction to control the system Extraction
T T T
rooms
MPC vs. hybrid
Choose control laws as: rooms
MPC vs. hybrid
ii. Q = Qr = [Qcr , Qar ] is exponentially stable
control control
Hybrid Control
2 −1
Hybrid Control iii. suppose that Qi (0) ≥ 0, Qir > 0 and λ < mini Ki Rir Qir , then
Strategy
Safety Control
Rc = (Kc QcD ) (Kc Hcr + λQce ) Strategy
Safety Control
Ri (t ) > 0, ∀t ≥ 0, where i = 1, . . . , n.
Example
2 −1 Example
Comparison Ra = (Ka QaD ) (Ka Har + λQae ) Comparison
Conclusions Conclusions Proof (main idea): use of the alebraic properties of the system,
d = Hmr + Rm Qm
Kirchhoff’s law and input-to-state linearization.
where r = reference (equilibrium) values of controlled heads Note: min R = branch resistor when the actuator “door” is fully
(m = fan branch); deviations from equilibrium quantities: open.
Qc /a e = Qc /a − Qc /a r ; λ = constant; K = inverse of inertance.
Control Control
strategies for Design with controls in co-tree only strategies for
mining mining
• Control objective using only Rc (constant Ra and d ):
2 −1
Ventilation Rc = (Kc QcD ) (Kc Hc + λQce ) Ventilation
network
2 −1
network
Theorem: feasibility region
Controls in all
Ra = (Ka QaD ) Ka Har Controls in all
Let F = {Qc ∈ R N −nc +1 |Rci (Qc ) ≥ Rcimin , i = 1 . . . N − nc + 1}

branches branches
only
d = Hmr + Rm Qmr only
be the feasible control set, where Rcimin is the minimum feasible

Fast MPC Fast MPC
control values. Define also the set Br = {||Qe || ≤ r }.
Time-delay approach • Implies the dynamics Time-delay approach
PDE control PDE control Let r ∗ be the largest r such that Br ∈ F .
Extraction
rooms Q̇c 2
= −Kc QcD Rc + Kc Hc
) n Extraction
rooms
Then, Q = Qr is exponentially stable with the region of
MPC vs. hybrid
2 ⇒ Q̇c = −λQce MPC vs. hybrid
attraction that includes Br ∗ .
control Q̇a = −Ka QaD Ra + Ka Ha control
Strategy
Safety Control
Strategy
Safety Control
Proof (idea): exponential convergence from the analysis of the
Example Exponential stability is obtained BUT the constraint R > 0 Example Lyapunov function V = 12 ||Qce ||2 .
may not be satisfied (need implementable feedback)
• Refined feedback (Q̇a = λEQc Qce and model prop.):
2 −1
Rc (Qc ) = (Kc QcD ) [λ(I − Kc SHa Ka−1 EQc )Qce + Kc SHa QaD
2
(Qc )Rar
+Kc Sd dr − Kc Rm Sd eQmc Qc ]
where eQmc and S describe the connections topology.

Control Control
mining mining
Example: without auxiliary controller
E. Witrant Example: Mine ventilation networksystem with 4 E. Witrant
Ventilation branches Ventilation

network network
branches branches
only only

Fast MPC Fast MPC
rooms rooms
MPC vs. hybrid MPC vs. hybrid
control control
Strategy Strategy
Example Example
Conclusions Model obtained by choosing branches 3 and m as the tree of Conclusions
the network
Disturbance due to actuator saturation but the controller
eventually recovers to bring the system in its feasibility region.
Control Control
strategies for Example: with auxiliary controller strategies for
mining
ventilation
mining
ventilation
Controlling the distributed
dynamics in deep wells: coping
Ventilation
network
Ventilation
network with convection and resistance
branches branches
only only

Fast MPC Fast MPC
Time-delay approach Time-delay approach - Well known topology
- Essential impact of fluid
rooms rooms dynamics
control
Hybrid Control
control
Hybrid Control
- Control air transport
Strategy
Safety Control
Strategy
Safety Control
and energy dissipation
Example
Comparison
Example
Comparison
- Infinite-dimentional
- due to particular choice of initial conditions analysis (PDE,
(made to saturate the inputs in previous
Time-delays)
case), auxiliary control R3 not active
- d allows R1 and R2 not to saturate
- d varies in wide range and R1 and R2 have
trends opposite to previous case
Control Control
mining mining
E. Witrant Problem formulation E. Witrant Fast model predictive control (MPC): principles [W. &
Marchand 2008]
Ventilation • Hypothesis 1: heater operated based on atmospheric Ventilation
network network
Controls in all conditions (i.e. external input) and regulated input = Controls in all
• Predictive control: widely used in process industry for its
branches branches
Controls in co-tree
turbine downflow pressure (a local control loop is set on Controls in co-tree
robustness and slow dynamics of the processes.
only only
the turbine to adjust its rotational speed according to a + Control of multi-variable coupled dynamical systems
Fast MPC Fast MPC + Handling constraints on the state and on the control input
Time-delay approach
desired pressure). Time-delay approach
+ Express optimality concerns
Extraction
• Control problem: ensure a minimum pressure within the Extraction + Conceptually easy handling of nonlinearities in the system
rooms shaft (at each extraction level) based on the turbine rooms model.
control
actuation. control - Find the extremum of an optimization problem that can
Strategy Strategy hardly be guaranteed (i.e. for PDEs).
Safety Control • As pressure ր as we go deeper, control = ensuring a Safety Control
Example
Comparison minimum pressure at the bottom of the pit.
Example
Comparison
• Fast MPC: enables an analytical solution to the predictive
Conclusions Conclusions control problem without any optimization by using the
• Hypothesis 2: distributed pressure measurements are
structure of the system.
available to set a state feedback control law.
Control Control
strategies for Fast MPC for pressure distribution strategies for Fast MPC for pressure distribution (2)
mining mining
ventilation Hyp.: slowly varying coeff. and neglect diffusion ⇒ ventilation Approximate the pressure open-loop (OL) profile p (x , t + τ),
E. Witrant pt (x , t ) = cpx (x , t ) + rp (x , t ) + spext (x ). MPC = E. Witrant τ ∈ [0, +∞], by:
Ventilation • find an open-loop control profile τ → p (0, τ) at t such that the Ventilation
p̃ (x , t + τ) = pe −αx e −βτ + pst (x , p (0, ∞))
network network
Controls in all
solution of the dynamical system has “some” properties (e.g. Controls in all
branches
stability, robustness, optimality . . . ) on a time horizon [t , t + w ] branches • when τ ր, p̃ (x , t + τ) → pst that ensures desired p (L , ∞)
only
where w is the prediction horizon (may be ∞) only
• at t (τ = 0), take: p = e αL [p (L , t ) − pst (L , p (0, ∞))] to ensure
Fast MPC • open-loop control profile applied at its first instant p (0, 0) and Fast MPC
the initial condition at x = L
Time-delay approach
PDE control
the scheme is repeated at the next time instant t + dt . Time-delay approach
PDE control
• to ensure p̃ solution of the PDE, impose β = αc − r .
Extraction • control objective: give inflow pressure control profile p (0, τ) to Extraction α is a free parameter, but:
rooms rooms
MPC vs. hybrid ensure a given down pressure p (L , ∞) in closed loop; MPC vs. hybrid
• α > r /c necessary: OL trajectory → p (L , ∞)
control control
Hybrid Control
Strategy
• once steady state achieved, a corresponding constant inflow Hybrid Control
Strategy • convergence speed ∝−1 difference between α and r /c
Safety Control pressure p (0, ∞) ensures the desired down pressure p (L , ∞) Safety Control
Example Example • the closer α is to r /c , the closer p̃ (x , t ) is to p (x , t ) (exact
Comparison • hyp. at t : start with some constant initial inflow p (0, t ), p (L , t ) at Comparison
solution of PDE)
Conclusions the bottom and an established steady state pressure profile: Conclusions
Z x
Resulting OL control law at t ∀τ > 0, is therefore:
− cr x s − cr (x −z )
pst (x , p0 ) = p0 e − e pext (z )dz
c 0 p̃ (0, t + τ) = (p (L , t ) − p (L , ∞)) e −α(x −L ) e −βτ + p (0, ∞)
" Z #
Note: p (L , ∞) = pst (L , p (0, ∞)) and, as soon as the system is r
L s L − r (L − z )
in a steady state regime at t , p (L , t ) = pst (L , p (0, t )). p (0, ∞) = e c p (L , ∞) + e c pext (z )dz
c 0
Control Control
mining
ventilation
mining
ventilation
Simulation results (2)
E. Witrant
Simulation results E. Witrant
1.06
x 10
5 Turbine pressure p(0,t)
1.06
x 10
5 Turbine pressure p(0,t)
network
• Two scenarii are compared: network
1.05
1.04
1.05
1.04
Controls in all
1 the inflow pressure is set at some value that ensures a Controls in all
p(0,t) (Pa)
p(0,t) (Pa)
1.03 1.03
branches branches
Controls in co-tree
only
down pressure of 1 hPa for the initial ventilation topology; Controls in co-tree
only
1.02 desired turbine pressure
effective turbine pressure
1.02 desired turbine pressure
effective turbine pressure
1.01 1.01
In deep wells 2 the inflow pressure is automatically computed and adapted In deep wells 1 1
Fast MPC
Time-delay approach
online according to the above control law using the down Fast MPC
Time-delay approach
0.99
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0.99
0 200 400 600 800 1000 1200 1400 1600 1800 2000
t (s) t (s)
PDE control pressure measurement. PDE control
5 Down pressure p(L,t) 5 Down pressure p(L,t)
x 10 x 10
Extraction
• In both cases, consider test case in Lesson 4 with 2nd Extraction 1.015 1.015
desired down pressure
rooms rooms 1.01 effective down pressure 1.01
MPC vs. hybrid
control level fan operated at t = 1000 s instead of t = 2000 s . MPC vs. hybrid
control 1.005 1.005
p(L,t) (Pa)
p(L,t) (Pa)
1 1
Strategy
Safety Control
• Control not directly set on the system but acts as a Strategy
Safety Control 0.995 0.995
Example
Comparison
reference to the turbine outflow pressure (shaft inflow Example
Comparison
0.99 0.99 desired down pressure
effective down pressure
Conclusions pressure set using the corresponding turbine speed by Conclusions

0.985
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0.985
0 200 400 600 800 1000 1200 1400 1600 1800 2000
t (s) t (s)
means of a local PI control law on the turbine).

Down pressure regulation in a mine with (right) and without
(left) fast predictive control
Control Control
strategies for Time-delay compensation approach [W. & Niculescu, 2010] strategies for Application to the mine ventilation process
mining mining
E. Witrant • Recall from Lesson 4: convective-resistive flow modelled E. Witrant Reference and effective (filtered) turbine output pressure:
5
Ventilation
with a delay (functional) differential equation Ventilation
1.14
x 10
1.13
network network
Controls in all
• Control objective: design a feedback such that the Controls in all
1.12
Turbine pressure (Pa)

branches
1
RL branches 1.11
Controls in co-tree
only
average distributed pressure p̄ (t ) = L 0
p (x , t )dx tracks Controls in co-tree
only
1.1
reference p̄r (t ), with

1.09
Effective
In deep wells In deep wells 1.08
Fast MPC Fast MPC 1.07 Reference

Time-delay approach " Z θf !# Time-delay approach
˙p̄ (t ) = − c (t ) u(t ) − u(t − θf )exp

1.06
r (η)d η + r (t )p̄ (t ). 1.05
Extraction
rooms
L 0
Extraction
rooms
1.04
1.03
0 200 400 600 800 1000 1200 1400 1600 1800 2000
MPC vs. hybrid MPC vs. hybrid Time (s)
control
Hybrid Control
• Achieved if (fixed point theorem): control
Hybrid Control
Strategy
p̄˙ (t ) − p̄˙ r (t ) + λ(p̄ (t ) − p̄r (t )) = 0 Strategy Feedback tracking error:
Safety Control Safety Control 4
10
Tracking error (Pa)

Example Example
Comparison • Using the previous DDE and solving for u(t ), it follows that Comparison 10
2
0
Conclusions Conclusions 10
L 10
−2
u(t ) = [r (t )p̄ (t ) + λ(p̄ (t ) − p̄r )] + p (L , t )

c (t )
−4
10
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Time (s)
ensures |p̄ (t ) − p̄r | = |p̄ (0) − p̄r |e −λt ⇒ Sensitivity to initial conditions and some numerical
integration errors but exponential convergence verified!
Control Control Dynamic Boundary Stabilization of Quasi-Linear
mining mining Hyperbolic Systems [Castillo, W., Prieur, Dugard, 2012]
Let n be a positive integer and Ω be an open non-empty
E. Witrant
Application to the mine ventilation process (p̄r = 1.1 hPa ) E. Witrant
convex set of Rn . Consider the general class of strict
network x 10
5
network quasi-linear hyperbolic systems of order n defined as follows:
Pressures after exhaust fans (Pa)

Controls in all
1.1 Controls in all
branches branches
1.09
Controls in co-tree
only
1.08
Controls in co-tree
only ∂t s (x , t ) + F (s (x , t ))∂x s (x , t ) = 0 (1)
rd
In deep wells 3 extraction level In deep wells
1.07
Fast MPC
Time-delay approach 1.06
Fast MPC
Time-delay approach
where s (x , t ) ∈ Ω and F : Ω → Rn×n (continuously
PDE control
1.05 2 nd extraction level PDE control
differentiable function). Consider system (1) in Riemann
Extraction 1 st extraction level Extraction
rooms 1.04 rooms coordinates (coupled transport equations):
MPC vs. hybrid 1.03 MPC vs. hybrid
control control
Hybrid Control
Strategy
1.02 Hybrid Control
Strategy ∂t ξ(x , t ) + Λ(ξ)∂x ξ(x , t ) = 0 ∀ x ∈ [0, 1], t ≥ 0
Safety Control 1.01 Safety Control
Example 0 200 400 600 800 1000 1200 1400 1600 1800 2000 Example
Comparison Time (s) Comparison
where ξ ∈ Θ, Λ is a diagonal matrix function Λ : Θ → Rn×n such
Conclusions
Pressures provided to the extraction rooms at the three levels. Conclusions
that Λ(ξ) = diag (λ1 (ξ), λ2 (ξ), ..., λn (ξ)) with
λ (ξ) < ... < λm (ξ) < 0 < λm+1 (ξ) < ... < λn (ξ), ∀ξ ∈ Θ
|1 {z } | {z }
ξ− ξ+
Control Control
strategies for strategies for Initial Condition
mining mining
ventilation Boundary Conditions h iT
ventilation
• Consider that there exists a δ0 > 0 and a continuously
State partition: ξ = ξ− ξ+ where ξ− ∈ Rm and ξ+ ∈ Rn−m . differentiable function ξ 0 : [0, 1] → Θ that satisfies the
Ventilation
network
Most approaches consider the following static boundary control Ventilation
network
zero-order and one-order compatibility conditions such
Controls in all
branches
for (??): ! !
Controls in all
branches
that |ξ 0 |H 2 ((0,1,R n )) < δ0 . Then, the initial condition can be
Controls in co-tree
ξ− (1, t ) ξ− (0, t ) Controls in co-tree
defined for system (1) as:
only
=G only
In deep wells ξ+ (0, t ) ξ+ (1, t ) In deep wells

Fast MPC | {z } | {z } Fast MPC
Time-delay approach
PDE control
Yc Yξ Time-delay approach
PDE control
ξ(x , 0) = ξ 0 (x ), Xc (0) = Xc0 , ∀ x ∈ [0, 1] (3)
Extraction
rooms
where the map G : Rn×n → Rn×n vanishes at 0. Extraction
rooms
MPC vs. hybrid
control
Instead, we consider the dynamic boundary conditions defined: MPC vs. hybrid
control
• The control problem is to find a control gain K such that
Strategy Strategy
Safety Control
X˙c = A Xc + B K Yξ Safety Control the system (1) with the boundary condition (2) is Lyapunov
Example
(2) Example
stable ∀ ξ ∈ Ξ ⊆ Θ.
Yc = Xc
n×n n ×n n ×n 7 9ABCDE
where A ∈ R , B ∈ R are given matrices and K ∈ R is 8 5
to be designed
123415647 9AFCDE
5
Control Control
mining mining
ventilation Theorem: exponential stability ventilation Euler Equations
E. Witrant Consider the system (1) with BC (2) and IC (3). Assume that E. Witrant
Consider the Euler equations (quasi-linear hyperbolic system)
Ventilation there exists a diagonal positive definite matrix Q ∈ Rn×n and a Ventilation expressed in terms of the primitive variables (density (ρ),
network network
Controls in all matrix Y ∈ Rn×n such that the following LMI is satisfied Controls in all velocity (u) and pressure (p )):
branches branches
Controls in co-tree ∀i ∈ [1, ..., Nϕ ] Controls in co-tree
only only
∂V ∂V
In deep wells " T
# In deep wells + A(V) + C(V) = 0
Fast MPC QA + AQ + Λ(wi )Q BY Fast MPC ∂t ∂x
≺0  
YT BT
Time-delay approach Time-delay approach    
PDE control −Λ(wi )Q PDE control
 ρ  u ρ 0   0 
   1  ; C =  
Extraction
rooms
Extraction
rooms
V = u  ; A = 0 u ρ   G 
MPC vs. hybrid Let K = YQ −1 , then there exist two constants α > 0 and M > 0 MPC vs. hybrid
 
p

0 a 2ρ u

(γ − 1)ρ(q + uG )
control control
Hybrid Control
Strategy
such that, for all continuously differentiable functions Hybrid Control
Strategy
Safety Control ξ 0 : [0, 1] → Ξ satisfying the zero-order and one-order Safety Control
q
γp
Example Example a= ρ is the speed of sound, γ is the specific heat ratio and
Comparison compatibility conditions, the solution of (1), (2) and (3) Comparison
Conclusions satisfies, for all t ≥ 0, Conclusions C (V ) is a function that describes the friction losses and heat
exchanges. As the isentropic case is analyzed, then C (V ) = 0.
||Xc (t )||2 + ||ξ(x , t )||L 2 (0,1) ≤ Me −αt ||Xc0 ||2 + ||ξ 0 (x )||L 2 (0,1) Video: A change of reference from Vref = [1.16, 20, 100000]T
to Vref = [1.2, 30, 105000]T is introduced.
Control Control
mining
ventilation
Predictive and hybrid control of mining
ventilation Problem formulation
E. Witrant
the extraction rooms [IJRNC 2010] E. Witrant
• Control objective: minimize the fan energy consumption
network network while ensuring an acceptable air quality in the extraction
branches branches room
only only
• Height-dependent model → objective rephrased as
In deep wells - Barely known In deep wells
guaranteeing a maximum pollutant concentration at a
Fast MPC Fast MPC
Time-delay approach topology Time-delay approach
given height zr :
Extraction
- Involve turbulence Extraction max yj (t ) ≤ ȳj ,
rooms and flow sources in a rooms ∀j
control
Hybrid Control
stratified distribution control
Hybrid Control where ȳj is the threshold value on pollutant j (i.e. COx or
Strategy Strategy
Safety Control Safety Control NOx ) and zr = highest height where the air quality has to
Example Example
Comparison Comparison be guaranteed and around which the sensors should be
Conclusions - Simplified grey-box model of the 2-layers profile Conclusions placed
- Periferal dynamics included with energy losses (tube),
• Communication constraints (delays, timeout, packet
time-delay (tube + network) and locally regulated fan
losses and bandwidth limitations) should also be included.
Control Control
mining
ventilation
Adequacy of linear controllers? mining
ventilation
Adequacy of linear controllers (2)
network network
Controls in all Test: PI tuned to regulate CO at 0.028 Controls in all
But, changing the reference to 0.035
branches branches
only only

Fast MPC Fast MPC
rooms rooms
control control
Strategy Strategy
Example Example
⇒ Non linearities of the system have to be carefully taken into

Very satisfactory !! But. . . account
Control Control
strategies for strategies for Problem: robustness against pollutant emissions
mining mining
ventilation NLMPC: Receding horizon control ventilation predictions
E. Witrant
• Control objectif: regulate the level of pollutants
E. Witrant
→ Need of a closed loop control architecture
network yj (t ) → ydes ,j (t ) network
branches • Constraints to be satisfied: maxt yj (t ) ≤ ȳj branches
only only

Fast MPC Fast MPC
Scheduling algorithm based
Time-delay approach
PDE control
Time-delay approach
PDE control
on prediction model
Extraction Extraction (maximum value of the
rooms rooms
MPC vs. hybrid MPC vs. hybrid delay)
control control
Hybrid Control Hybrid Control ⇒ constrained MPC with
Strategy • Control tuning parameters: Strategy
Safety Control Safety Control on-line solution of
Example • N : prediction horizon, long enough for transient behavior Example
Comparison Comparison successive optimization
• Nu : number of degrees of freedom: precision vs.
Conclusions
complexity
Conclusions problems
• λ: weight control effort vs. tracking performances
⇒ Intuitive tuning based on system response
Control Control
mining
ventilation
Hybrid Control Strategy mining
ventilation
Ventilation Affine hybrid model → threshold control Ventilation

Safety Control
network network
branches (x1 £ gl,1 – e) Ú branches
Controls in co-tree
L
x1 = a1x1 + b1
L
q1 (x2 ³ gh,2 + e)Ú Controls in co-tree
• Automatically verify if for a given control strategy the
only x2 = a2x2L + bL2 (x3 ³ gh,3 + e)Ú only
In deep wells
L
x3 = a x + b
3 3
L
3 (x1 > gl,1 – e)Ù
(x4 > 0)
In deep wells
hybrid automaton satisfies Safety and Comfort properties
x4 = 1 (x2 < gh,2 + e)Ù
x1 := x1
Fast MPC
x2 := x2
(x3 < gh,3 + e)Ù
(x4 > 0) x1 := x1
Fast MPC • Unfortunately, model checking in general undecidable
PDE control
x3 := x3
x4 := 0
x2 := x2
x3 := x3
PDE control even for affine hybrid automata
x4 > d x4 := 0
Extraction Extraction • construct an abstraction of a hybrid automaton with affine
rooms rooms
x1 = a1xH1 + bH1 x1 = a1x1L + bL1
MPC vs. hybrid H
x2 = a2x + b
2
H
2
L
x2 = a x + b
2 2
L
2
MPC vs. hybrid dynamics, which preserves temporal properties expressed
control control
Hybrid Control
Abstraction of hybrid x4 = 1
H
x3 = a3x 3 + b3
H
x3 = a3x3L + bL3
x4 = 1
Hybrid Control by CTL and TCTL formulae (temporal logics constraints)
Strategy Strategy
• abstract model belongs to a subclass of timed automata,
Safety Control
Example
automaton with affine x4 < d q4 x4 < d q2 Safety Control
Example
called durational graph
Comparison
dynamics, which preserves x1 := x1
x2 := x2
x4 > d Comparison
Conclusions x3 := x3 Conclusions
temporal properties x4 := 0 x1 = a1xH1 + bH1 q3 x1 := x1
x2 = a2xH2 + bH2 x2 := x2
expressed by CTL and x3 = a3xH3 + bH3 x3 := x3
(x1 < gh,1 + e)Ù
(x1 ³ gh,1 + e)Ù x4 = 1 x4 := 0
TCTL formulae (temporal (x2 £ gl,2 – e)Ù
(x2 > gl,2 – e)Ù
(x3 > gl,3 – e)Ù
(x3 £ gl,3 – e)Ù (x4 >0)
logics constraints) (X4 >0)
Control Control
mining
ventilation
Example mining
ventilation
Comparison between the two
E. Witrant
Automatic verification procedure on
E. Witrant
approaches
Ventilation the hybrid model using the following Ventilation
network
Controls in all
set of thresholds (high and low, 3 network
Controls in all Test case (CO and NO concentrations)
branches gases) branches
only only
In deep wells gl 1 = 0.2955, gh 1 = 0.2975 [Kg /m 3] In deep wells

−3
Fast MPC
gl 2 = 0.5 × 10 , gh 2 = 2.5 × 10−3 [Kg/m 3] Fast MPC
3
Number of trucks and gas concentration (g/m3)

PDE control gl 3 = 0.0885, gh 3 = 0.091 [Kg /m 3] PDE control
2.5
rooms rooms
MPC vs. hybrid MPC vs. hybrid 2
control control
Strategy Strategy 1.5
Example Example
Comparison Comparison 1
0.5
0
0 500 1000 1500
time (s)
Safe original hybrid system and maximum time of
uncomfortable air < 62 s
Control Control
strategies for strategies for Comparison between the two approaches
mining
ventilation
Comparison: regulation efficiency mining
ventilation −4
x 10
80
E. Witrant E. Witrant 2.5
70
2 60
CO [kg/m3 ]
ufan [m/s]
Ventilation Ventilation 50
network
−4 network 1.5
40
x 10
1 30
branches branches
80
MPC (solid line):
CO Concentation[kg/m ]
Control input ufan [m/s]

3
Controls in co-tree Controls in co-tree 20
only only 0.5

10
2 60
MPC with underestimated pollutant source (1 truck)
In deep wells • direct trade-off In deep wells 0
0 500 1000 1500
0
0 500 1000 1500
40 −3 MPC with parameter uncertainty (20%) 6
x 10 x 10
Fast MPC
1
Time-delay approach
regulation efficiency Fast MPC
Time-delay approach
6
Thresholds with guards 80/90 % → 85/95 %
3.5
20 5 3
PDE control
vs. energy min. PDE control
2.5
Energy [MJ]
NO [kg/m3 ]
4
Extraction
0 0 Extraction
0 500 1000 1500 0 500 1000 1500 2
rooms Time[s] Time[s] • better ratio rooms 3
1.5
−3 6 2
control
x 10 x 10 control
6
Hybrid Control
6 Threshold (dashed line): Hybrid Control
1
1
0.5
Strategy Strategy
NO Concentation[kg/m ]
3
5 5
Safety Control
4 4
• easy tuning Safety Control 0
0 500
Time [s]
1000 1500
0
0 500 1000 1500
Example Example Time [s]
Energy [J]
3 3
• find the tighter band
Conclusions
2 2 Conclusions
for guards Thresholds: MPC:
1 1
O(1) complexity, time consuming but still realistic (RT/2.5)
0 0
0 500
Time[s]
1000 1500 0 500
Time[s]
1000 1500 comparisons betw. meas. and hard optimization problems
safety thresholds
can be embedded on WSN centralized control
fails safety if too tight guards finer tuning and more robust
Control Control
mining
ventilation
Conclusions mining
ventilation
Main references
E. Witrant E. Witrant • Y. Hu, O. Koroleva, and M. Krstic, “Nonlinear control of mine ventilation
networks,” Systems and Control Letters, vol. 49, pp. 239-254, 2003.
network network http://flyingv.ucsd.edu/papers/PDF/62.pdf
branches branches • E. W. and N. Marchand, ”Modeling and Feedback Control for Air Flow
Controls in co-tree
only
• Variety of control problems Controls in co-tree
only Regulation in Deep Pits”, Cambridge Scientific Publishers, 2008.
In deep wells • Necessitate appropriate control laws considering, i.e. In deep wells http://www.gipsa-lab.grenoble-inp.fr/˜ e.witrant/papers/11_CSP_Witrant.pdf
Fast MPC
Time-delay approach • algebraic structure from the graph and nonlinearities for
Fast MPC
Time-delay approach
• E. W., S.I. Niculescu, “Modeling and Control of Large Convective Flows
PDE control PDE control with Time-Delays”, Mathematics in Engineering, Science and
network control
Extraction Extraction Aerospace, Vol 1, No 2, 191-205, 2010.
rooms • distributed dynamics (PDE) for large convective flows rooms
http://www.gipsa-lab.grenoble-inp.fr/˜ e.witrant/papers/10_Witrant_MESA.pdf
MPC vs. hybrid
• peripheral dynamics and height-dependent distributions for MPC vs. hybrid
control control
• E. W., A. D’Innocenzo, G. Sandou, F. Santucci, M. D. Di Benedetto, A.
Hybrid Control
Strategy rooms Hybrid Control
Strategy
J. Isaksson, K. H. Johansson, S.-I. Niculescu, S. Olaru, E. Serra, S.
Example • Preliminary mathematical and simulation analysis that Example Tennina and U. Tiberi, ”Wireless Ventilation Control for Large-Scale
motivate experimental validation Systems: the Mining Industrial Case”, International Journal of Robust
Conclusions Conclusions and Nonlinear Control, vol. 20 (2), pp. 226 - 251, Jan. 2010.
http://www.gipsa-lab.grenoble-inp.fr/˜ e.witrant/papers/09_IJRNC_r23.pdf
• F. Castillo Buenaventura, E. W., C. Prieur and L. Dugard, ”Dynamic

Boundary Stabilization of Hyperbolic Systems”, Proc. of the 51st IEEE
Conference on Decision and Control, Maui, Hawaii, December 10-13,
2012. http://www.gipsa-lab.grenoble-inp.fr/˜ e.witrant/papers/13_cdc_castillo.pdf
Feedback Feedback
control control
Outline
Background Background
Goals Goals
Communication Communication 1 Background

Architecture Architecture
Uniform radio network
Hybrid wired-wireless
Mining Ventilation Control Uniform radio network
architecture
Extended mobile
architecture
Extended mobile
2 Motivation & Goals
wireless architecture wireless architecture
Models, Models,
Algorithms and Lesson 7: Application of wireless sensors to Algorithms and 3 Communication Architecture
Tools Tools
Distributed Estimation
by Wireless Sensors
mining ventilation Distributed Estimation
by Wireless Sensors
Latency/Energy Latency/Energy
Models in Multi-hop Models in Multi-hop 4 Models, Algorithms and Tools
WSN WSN
Advanced
Emmanuel WITRANT Advanced
Services emmanuel.witrant@ujf-grenoble.fr Services
5 Advanced Services
Universidad Pedagógica y Tecnológica de Colombia,
Sogamoso, April 4, 2013
Feedback Feedback
control control Mine topology
principles
Background principles
Goals Goals
Communication Communication
Architecture • Typical existing ventilation control systems are either poor Architecture
Uniform radio network Uniform radio network
architecture
or non existing, a continuous monitoring of air quality is Hybrid wired-wireless
architecture
Extended mobile
wireless architecture
absent, and the only communication capability is simply Extended mobile
Models, voice over walky-talkie Models,

Algorithms and Algorithms and
Tools • The amount of air pumped in the rooms is manually Tools
Distributed Estimation Distributed Estimation
by Wireless Sensors
Latency/Energy
controlled in open-loop, and ventilation energy can be by Wireless Sensors
Latency/Energy
Models in Multi-hop
WSN optimized while guaranteeing operators safety Models in Multi-hop
WSN
Advanced
Services
• Reference works: [Fischione, Pomante, Santucci et al., 2008, Advanced
Services
W. et al. 2009]
Feedback Feedback
control control
principles
Motivation & Goals principles
Communication Architecture
• The primary need for automation is sensing:
Goals • Advanced communication architectures for distributed Goals • to obtain information about air pressure and temperature
Communication sensing/actuation leads a mining operating site to become Communication to apply a proper control strategy
• for the secondary system, gas concentrations are also
Uniform radio network an interconnected system Uniform radio network
architecture
architecture
required
Extended mobile
• Advanced control strategies allow to save energy and Extended mobile
wireless architecture wireless architecture • Basic sensing infrastructure composed of:
improve safety
Models, Models,
Tools
• Provide advanced services Tools
- fixed wireless sensor
by Wireless Sensors
Latency/Energy
Wireless sensing and communication architecture: by Wireless Sensors
Latency/Energy nodes in the ventilation
Models in Multi-hop Models in Multi-hop
WSN
• Design space exploration keeping into account present WSN
shaft, tunnels and
Advanced Advanced
Services and future needs able to increase safety and efficiency of Services extraction rooms
the whole system - mobile wireless
• Set of models and algorithms needed to validate the sensors associated to
proposed approaches by means of simulations entities working in the
secondary system
Feedback Feedback
control control Uniform radio network
E. Witrant E. Witrant • One radio technology (e.g. 802.15.4) is adopted in the

system: interaction with the existing infrastructure is
minimal
Goals Communication Architecture (2) Goals
Communication Communication • Topology variation induces an impact in scalability of

Architecture • The sensing network has to be complemented by a Architecture
solutions, since a larger size implies larger number of
architecture
communication network portion, which is in charge of Hybrid wired-wireless
architecture
hops, longer delays and larger traffic to be supported by
Extended mobile
wireless architecture delivering information over longer ranges, up to controllers Extended mobile
relaying nodes
Models, and actuators Models,
• the nodes deployed on portions of tunnels where electrical
Algorithms and Algorithms and cabling is absent should be battery powered
Tools • Different opportunities can be considered, i.e. Tools
Distributed Estimation Distributed Estimation • mobile nodes is not subjected to energy issues, since they
by Wireless Sensors
• Uniform radio network by Wireless Sensors
Latency/Energy Latency/Energy can be maintained by people when they come to ground
Models in Multi-hop
WSN
• Hybrid wired-wireless architecture Models in Multi-hop
WSN
Advanced • Extended mobile wireless architecture Advanced

Services Services
Feedback Feedback
control Hybrid wired-wireless architecture control
principles principles Extended mobile wireless architecture
E. Witrant • Since the system foresees the presence of some cabling it E. Witrant The basic setup depicted so far can be enhanced by further
is interesting to consider Power Line Communication exploiting the mobile entities:
devices for setting up the network • An interesting evolution exploits hand-helds and on-truck nodes
Goals Goals
• provide a wired backbone along the power line to connect to build a dynamic wireless backbone
Architecture the Control Room with an 802.15.4 WSN Architecture
• Such a network would help to make energy constraints on some
• could be very useful for relay nodes Uniform radio network
architecture
architecture lower tier wireless sensor nodes less stringent
Extended mobile
• A proper PLC/802.15.4 Gateway is needed: acts also as a Extended mobile
Models, sink node for the patches described before Models,

Tools Tools
by Wireless Sensors
by Wireless Sensors
• Moreover, the
Latency/Energy
Models in Multi-hop
Latency/Energy
Models in Multi-hop
introduction of a
WSN WSN
802.11 network
Advanced Advanced
Services Services
would provide a
larger bandwidth
Conclusions Conclusions support for
introducing
advanced services
Feedback Feedback
control control
Models, Algorithms and Tools
E. Witrant Extended mobile wireless architecture with mobile E. Witrant
Background gateway Background
Goals Goals
Hybrid wired-wireless Hybrid wired-wireless
architecture architecture Different approaches by means of simulations have been
Extended mobile Extended mobile
developed as a set of models and algorithms related to the
Models, Models,
main features of the system:
Tools Tools
Distributed Estimation Distributed Estimation • Distributed Estimation by Wireless Sensors
by Wireless Sensors by Wireless Sensors
Latency/Energy
Models in Multi-hop
Latency/Energy
Models in Multi-hop
• Latency/Energy Models in Multi-hop WSN
WSN WSN
Advanced Advanced
Services Services
Feedback Feedback
control Distributed Estimation by Wireless Sensors control
E. Witrant • Objective: use a set of distributed measurements to obtain E. Witrant

Latency/Energy Models in Multi-hop WSN
Background
an estimate of the air quality Background
Considering the uniform radio network architecture the wireless
Goals • These approaches are in contrast to the traditional ones Goals
network can be interpreted as a clustered network:
Communication
Architecture
with sensors that provide raw data to a fusion center Communication
Architecture
• deal with network delay for closed-loop analysis and with
• Results have been achieved for a movable grid of sensors Uniform radio network
architecture
architecture
energy consumption concerns for network lifetime with
and some experimental results have been obtained in lab
Extended mobile
Extended mobile
wireless architecture limited maintenance
Models,
facilities with light sensing elements Models,
→ high level model for end-to-end delay that can be further
Tools • A practical example: reconscruction of a known profile (of Tools
refined introducing some topological details
by Wireless Sensors light) by means of a network of MICAz nodes Distributed Estimation
by Wireless Sensors
Latency/Energy
Models in Multi-hop
Latency/Energy
Models in Multi-hop
→ model for the energy spent in the Transmission /
WSN WSN
Reception / Idle states and during the transitions
Advanced Advanced
Services Services • need a framework to support energy/performance analysis
Conclusions Conclusions of control algorithms by means of discrete event network
simulators
Feedback Feedback
control control
Advanced Services
Background Latency/Energy Models in Multi-hop WSN (2) Background
Goals Goals
Communication
• 2 discrete event simulator have been setup (OMNeT++, Communication The presence of mobile nodes in the system leads to consider
DESYRE) to evidence energy depletion and latency: Uniform radio network other services
Hybrid wired-wireless Hybrid wired-wireless
architecture
Extended mobile
architecture
Extended mobile
• A localization service is of particular interest and it would
be limited, at a first instance, to provide the tunnel/room
Models, Models,
Algorithms and Algorithms and coordinates
Tools Tools
by Wireless Sensors
by Wireless Sensors
• A perspective view is to deploy a localization services that
Latency/Energy
Models in Multi-hop
Latency/Energy
Models in Multi-hop
allows to track the position within the whole mining area
WSN WSN
Advanced Advanced
• Other advanced services require the extended
Services Services
architecture (e.g. 802.11), e.g. IP-based voice/video
Conclusions Conclusions service
Feedback Feedback
control control
Conclusions
Background A longer term view Background
Goals Goals
architecture
architecture • Architecture definition and algorithms design for a WSN in
Extended mobile Extended mobile
wireless architecture wireless architecture the mining scenario
Models, Models,
Algorithms and Algorithms and • Need analysis of network architectures of increasing
Tools Tools
complexity, with the goal of supporting distributed
by Wireless Sensors by Wireless Sensors
Latency/Energy Latency/Energy estimation and fulfilling reliability and latency requirements
Models in Multi-hop Models in Multi-hop
WSN WSN
for a ventilation control application
Advanced Advanced
Services Services
Feedback
control
principles
Reference
E. Witrant
Background
Goals 1 C. Fischione, L. Pomante, F. Santucci, C. Rinaldi, S.

Communication Tennina, “Mining Ventilation Control: Wireless Sensing,
Architecture
Uniform radio network Communication Architecture and Advanced Services”, in
architecture
Extended mobile
Proc. of IEEE Conference on Automation Science and
Engineering (IEEE CASE 08), Washington, DC, USA,
Models,
Algorithms and August 2008.
Tools
by Wireless Sensors
2 E. Witrant, A. D’Innocenzo, G. Sandou, F. Santucci, M. D.
Latency/Energy
Models in Multi-hop
Di Benedetto, A. J. Isaksson, K. H. Johansson, S.-I.
WSN
Niculescu, S. Olaru, E. Serra, S. Tennina and U. Tiberi,
Advanced
Services ”Wireless Ventilation Control for Large-Scale Systems: the
Conclusions
Mining Industrial Case”, International Journal of Robust
and Nonlinear Control, vol. 20 (2), pp. 226 - 251, Jan.
2010. http://www.gipsa-lab.grenoble-inp.fr/˜ e.witrant/papers/09_IJRNC_r23.pdf

13 MVC Sogamoso

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

13 MVC Sogamoso

Uploaded by

Copyright:

Available Formats

Mining Ventilation Control

Dr. Emmanuel WITRANT

Universidad Pedagogica y Tecnologica de Colombia

Provide technical skills and research insights for mining

• C. Hirsch, Numerical Computation of Internal & External Flows:

criteria gases, quality of diesel fuel and lubricating oil, and

Airflows in during the combustion process Airflows in

Mathematical • EC:TC Ratio: fraction of EC in TC. Mathematical

Mathematical number, mine Hv = V 2 /2g, not cumulative but appears as Mathematical

Design and Design and

Mathematical • Laminar flow in leakage through doors and stoppings in Mathematical

Mathematical gradient difference at a given flow velocity, first Mathematical

Design and Design and

dimensionality and numerical issues [N. Petrov];

Control system Control system

Time and Time and 1 The process of control system design

The feedback The feedback

Stability and Stability and 4 Stability and controllability

Stability and 3 Control: performance specifications & actuators Stability and

Robustness vs. Robustness vs.

Control system Control system

Robustness vs. Robustness vs.

Control system Control system

Time and 2. Automation design Time and

The feedback The feedback

Robustness vs. Robustness vs.

Control system 4. Simulation Control system

Tracking error |Tdi−Ti| (°C)

Stability and Stability and 0.5

Control system Control system

• Uncertain G , Gd : need for robustness

Time and Time and

The feedback The feedback • Insights from frequency-dependent plot

Control system ẋ1 (t ) = −a1 x1 (t ) + x2 (t ) + β1 u(t ) Control system

Stability and Stability and

E. Witrant E. Witrant • Defining the scaling factors

1 Formulate nonlinear state space model from physics, i.e.

Conclusions Conclusions δ˙x = A δx + B δu, δx (0) = x (0) − x ∗

E. Witrant Continuous-time linear system E. Witrant

Control system Control system

The feedback The feedback

Stability and Stability and y [k ] = Cx [k ] + Du[k ]

E. Witrant E. Witrant Sampled systems (4)

• For more general systems, Control configurations

The feedback The feedback

Stability and q−1 Stability and

Control system Control system y = G (s )u + Gd (s )d

Plant input L = GK loop transfer function

Conclusions • SSd small for disturbance rejection Conclusions

Time and stability of linear systems Time and

Control system Control system Scaling and performance

Robustness vs. Robustness vs.

E. Witrant E. Witrant Stability for nonlinear systems [Marchand, 2009]

Time and Time and

Robustness vs. Robustness vs.

E. Witrant Graphical interpretation E. Witrant

E. Witrant Lyapunov theory: Lyapunov functions E. Witrant

Control system Definition: V : Rn → R is a Lyapunov function if continuous and Control system

Time and Time and

Time and Time and

Stability and Stability and

Robustness vs. Robustness vs.

Time and Time and