metals

Article
An Efficient Fluid-Dynamic Analysis to Improve
Industrial Quenching Systems
Manuel de J. Barrena-Rodríguez 1 , Marco A. González-Melo 1 , Francisco A. Acosta-González 1, *,
Eddy Alfaro-López 2 and Francisco A. García-Pastor 1
1 Centro de Investigación y de Estudios Avanzados del I.P.N., Industria Metalúrgica # 1062, Parque Industrial
Saltillo-Ramos Arizpe, Ramos Arizpe 25900, Mexico; mbarrena85@gmail.com (M.d.J.B.-R.);
marco.gonzalez@cinvestav.edu.mx (M.A.G.-M.); francisco.garcia@cinvestav.edu.mx (F.A.G.-P.)
2 Rassini Suspensiones S.A. de C.V., Puerto Arturo # 803, Col. Bravo, Piedras Negras 26040, Mexico;
ealfarol@rassini.com
* Correspondence: andres.acosta@cinvestav.edu.mx; Tel.: +52-844-438-9600

Academic Editor: Murat Tiryakioglu

Received: 20 March 2017; Accepted: 17 May 2017; Published: 25 May 2017

Abstract: This paper addresses the problem of understanding the relationship between fluid flow
and heat transfer in industrial quenching systems. It also presents an efficient analysis to design or
optimize long standing quenching tanks to increase productivity. The study case is automotive leaf
springs quenched in an oil-tank agitated with submerged jets. This analysis combined an efficient
numerical prediction of the detailed isothermal flow field in the whole tank with the thermal
characterization of steel probes in plant and laboratory during quenching. These measurements were
used to determine the heat flow by solving the inverse heat conduction problem. Differences between
laboratory and plant heat flux results were attributed to the difference in surface area size between
samples. A proposed correlation between isothermal wall shear stress and heat flux at the surface
of the steel component, based on the Reynolds-Colburn analogy, provided the connection between
thermal characterization and computed isothermal fluid flow. The present approach allowed the
identification of the potential benefits of changes in the tank design and the evaluation of operating
conditions while using a much shorter computing time and storage memory than full-domain fluid
flow calculations.

Keywords: steel quenching; quenching heat flux; quenching boiling curve; heat transfer coefficient;
quenching fluid flow; oil quenching

1. Introduction
The generation of a quantitative understanding of fluid dynamics and heat transfer during
quench tank production is a research area that offers wide space for technical contributions. Many of
commercially important steels and aluminum alloys are heat-treated to achieve the target hardness and
tensile strength. This treatment consists of heating the metallic pieces to their dissolution temperature
and then quenching them by fast cooling. Water, aqueous polymer, oil, molten salt or gas is generally
used as a quenching medium, and the choice depends on the alloy chemical composition and the
desired mechanical properties of the product. Commercial alloys have been properly formulated
for suitable quenching to reach specific property values. However, long-standing production tanks
introduce a major challenge: controlling agitation, its intensity and uniformity. Agitation plays a key
role in determining the rate of heat transfer from the workpieces to the quenching medium. When
a hot solid piece is immersed into the liquid, boiling occurs. First, a vapor film develops over the
surface, and the rate of heat transfer is controlled by the isolating properties of this stable vapor
film. Once the surface cools to reach the so-called Leidenfrost temperature, the vapor film collapses,

Metals 2017, 7, 190; doi:10.3390/met7060190 www.mdpi.com/journal/metals

Metals 2017, 7, 190 2 of 29

and the liquid directly contacts the solid surface, generating a swarm of bubbles detaching from
the surface. This leads to a fast increase in the heat flux to reach a maximum value, the so-called
critical heat flux. During cooling, solid-phase transformations occur in the metal that create one or
more new phases with required mechanical properties. For example, austenite (a solid solution of
carbon in gamma iron with face-centered cubic lattice) is transformed into hard phase martensite
(a supersaturated solid solution of carbon in alpha iron with body-centered tetragonal lattice) when
steel is quenched from austenitization temperature to a temperature below Ms (martensite start
temperature). The specific volume of martensite is ~3.5% higher than the corresponding value of
austenite, which may lead to an adverse effect during non-uniform cooling. The internal stresses in the
solid that are generated from this phase transformation combine with stresses promoted by thermal
gradients, and the result may produce disastrous consequences in terms of distortion and/or crack
formation in the product. Furthermore, spot areas where the stable vapor film remained for a long
time would probably lead to low spotty hardness as result of an insufficient cooling rate to generate
the desired solid-phase transformation.
Industrial quenching practice has largely remained on a semi-empirical basis because there
is a minimal quantitative understanding of the quenching process itself in many heat treating
shops [1]. Heat treating engineers and equipment suppliers struggle to find design and operating
solutions without any industry-wide guidelines for quench system design, except for some company
specifications. Although it is well known that fluid flow varies greatly as a function of position within
industrial tanks, fluid velocity distribution monitoring is rarely found in production quench systems.
An important reason for this is that properly designed flow devices with sufficient sensitivity and
ruggedness for use in a heat treating shop are not fully commercially available, as mentioned in [2].
This reference presents a useful compilation of the existing flow process sensors and methods.
The state of the art in the application of fundamental knowledge to understand the complex
quenching phenomena is better appreciated from a review of a number of publications presenting
mathematical models of this process. Table 1 presents the evolution of the predictive power of
mathematical models developed for metal quenching using a liquid. This is not a comprehensive list;
its purpose is merely to offer a notion of the general trend in modeling according with the purpose
of this work. The reader can find a complete and detailed review of the quench numerical models
in [3]. Table 1 shows that from the early 1990s to the beginning of this 21st century, computational fluid
dynamics (CFD) focused on isothermal fluid flow calculations. In these works [4–9], the Navier-Stokes
and continuity equations were numerically solved together with turbulence equations under steady
state conditions. The pioneering work of Totten et al. [4] showed that flow is not uniform in quenching
tanks. Garwood et al. [5] validated their fluid flow calculations for an industrial tank using a laboratory
physical model in which quantitative velocity measurements were carried out using Laser Doppler
anemometry (LDA). Halva and Volný [6] computed isothermal fluid flow to examine the homogeneity
of velocity distribution as a function of agitator placement. The effect of the location of spray eductors
around a rack of aluminum panels on flow homogeneity was studied by Bogh [7]. Isothermal fluid
flow simulations were computed by Kernanzhitskiy [8], who used these results combined with
Taguchi partial factorial statistical analysis methodology to provide innovative design concepts for
a new quenching system design. Kumar et al. [9] evaluated flow uniformity around automotive
pinion gears during quenching. The success of fluid flow simulations was further boosted by more
powerful and available computer facilities. In the present decade, new contributions appeared that
solved, simultaneously with CFD equations, the transient heat conduction equation to determine the
temperature evolution of a solid body within the computational domain [10–18]. References [13,14,18]
also solved the corresponding solid-phase transformation kinetic equations, and in [13,18], the residual
stress and deformation were additionally computed. Reference [10] presented simulations of two
systems of agitation in a quench process. The authors assumed a non-isothermal fluid flow in
contact with a solid with time-dependent temperature. However, boiling was not considered in their
formulation. The heat transfer coefficient was computed from the temperature profile in a turbulent
Metals 2017, 7, 190 3 of 29

boundary layer. In this way, the authors could obtain a heat transfer coefficient that changed with both
the position on a solid surface and time.

Table 1. Evolution of predictive power of mathematical models developed for metal quenching.

Solid-Phase Residual Stress and

Author (s), Year CFD Heat Conduction
Transformation Distortion
Totten et al., 1993 - - -
Garwood et al., 1992 - - -
Halva and Volny, 1993 Isothermal fluid - - -
Bogh, 1994 flow - - -
Kernazhitskiy, 2003 - - -
Kumar et al., 2007 - - -
Non-isothermal
R.-Bineli et al., 2010 Non-isothermal solid - -
but no-boiling flow
Isothermal and
Krause et al., 2010 - -
non-isothermal solids
Mixed fluids
Srinivasan et al., 2010 - -
Gao et al., 2011 Non-isothermal solid yes yes
Computed
Ricci et al., 2011 yes -
interface, VOF
Computed
El-Kosseifi, 2012 Non-isothermal solid - -
interface, LS
Yang et al., 2013 - - -
Isothermal flow
Banka et al., 2013 Non-isothermal solid yes yes
Hardness prediction
Ko et al., 2013 - using Quenching Factor
Mixed fluids Non-isothermal solid Analysis (QFA)
Kopun et al., 2014 - -
Passarella et al., 2014 yes yes

References [11–13,16–18] moved a step forward by simulating boiling heat transfer using the
mixed fluids approach. This method considers two intermixed or interpenetrated fluids (vapor
and liquid) with the space-dependent void fraction of every phase. Momentum, continuity and
energy equations are written for each phase and solved simultaneously. The rates of exchange of
mass, momentum and energy between phases are included as source terms in the corresponding
equations and are calculated from empirical expressions. Therefore, this approach requires to define the
corresponding empirical constants. Krause et al. [11] considered both isothermal and non-isothermal
solid bodies. The numerical results obtained for the former case correspond to a steady-state boiling
heat transfer for which there is a considerable number of reported works. Therefore, the authors
validated their model against literature results. Srinivasan et al. [12] implemented their laminar flow
model in the commercial code AVL-FIRE (the AVL-designed 3D CFD software package) to simulate the
temperature evolution of a solid plate dipped at a given velocity in static water in a laboratory container.
The quenching of an “S”-shaped part was reported in [13]. The authors computed the solid-fluid heat
flux from a balance between the heat required to evaporate, quench and facilitate liquid convection
according with the heat flux partitioning model, the so-called Rensselaer Polytechnic Institute (RPI)
model of subcooled boiling, implemented in the Fluent code. They presented a computed heat
transfer coefficient that changes with time and the position on the solid surface and the corresponding
computed deformation and stress distribution of the part. A more fundamental CFD approach was
presented in [14,15] in which multiphase boiling heat transfer was computed by considering a single
fluid with space-dependent properties to calculate the interface position where the physical properties
jump from liquid to gas values. Level set (LS), Volume of Fluid (VOF) and Phase Field are examples of
methods in this category. The results from this approach provide a detailed picture of the actual boiling
phenomena, such as the rates of nucleation, growth and detachment of bubbles. However, it demands
considerable computational effort because the time and space steps for numerical integration should
Metals 2017, 7, 190 4 of 29

be very small. Isothermal flow calculations gained importance in [19,20]. Yang et al. [19] combined the
CFD results with artificial neural network (ANN) analysis to determine an optimal design of water flow
distribution in a quenching tank for large and complicated aluminum alloy workpieces. A relatively
small number of fluid flow simulations provided enough data to the ANN algorithm to determine the
optimal position of directional flow baffles and agitation speed. Reference [20] evaluated the effect of
orientation of a hot ring with respect to the direction of a fluid flowing around it. Heat transfer was
computed assuming a constant solid surface 100 ◦ C above the fluid temperature. However, the authors
computed the temperature evolution in the solid and its kinetics of solid-phase transformation and
ring distortion.
The previous literature review indicates that predictions of solid-phase transformation kinetics and
steel distortion and hardness depend on the knowledge of the rate of heat transfer during quenching.
Typically, the heat transfer coefficient is used to establish a thermal boundary condition on the solid
surface. However, available data of this coefficient are seldom related to the specific fluid dynamic
conditions in industrial tanks and therefore simplifying assumptions are adopted for its determination.
The present work proposes a correlation between isothermal wall shear stress and heat flux at the
surface of a quenching steel component. This correlation is then used to evaluate more accurately the
solutions proposed from the isothermal fluid-dynamic analysis during steel quenching in industrial
tanks. In contrast to the commonly found full-domain calculations in which the momentum and
mass equations are solved simultaneously in the whole domain, the present work shows fluid flow
calculations that were efficiently carried out in “cascade downstream sequence”. This method allowed
a high level of detail to be obtained in the flow field by using an order of magnitude fewer volume
elements of fluid and a much shorter computation time compared with full-domain calculations.
Therefore, it was possible to carry out multiple numerical simulations in a relatively short time to
search for a new design that would improve the oil velocity around the leaf springs.

2. Materials and Methods

2.1. Materials Properties

Table 2 shows the chemical composition of the leaf springs considered for this study and Table 3
presents the thermophysical properties of the steel and the quenching oil. Chemical composition
was obtained from a spark optical emission spectrometry analysis. C, S, O and N were determined
from combustion with infrared absorption detection analysis; O and N additionally required thermal
conductivity analysis. Knowledge of the dependence of steel density, thermal conductivity and heat
capacity with temperature is mandatory for a heat flow analysis. Figure 1 shows the corresponding
property plots determined from JMatPro® software data base (Version 9, Sente Software Ltd., Surrey
Technology Center, Guildford, UK, 2016) using the chemical composition of steel and an austenitic
grain size of ASTM 9. These property values were determined considering the cooling of austenite
to form martensite. The leaf springs are quenched at rates of cooling that essentially produce only
martensite throughout their whole cross section area. The table shows the density; viscosity; boiling
point, Tsat ; and temperatures for the start of martensite formation, Ms , and 50% and 90% of martensite
formation, M50 and M90 , respectively.

Table 2. Chemical composition of DIN 51CrV4 steel employed.

Chemical Composition C Si Mn P S Cr V Sum of Others 1

Mass % 0.50 0.24 0.97 0.04 0.008 1.04 0.14 <0.281
1 Analyzed impurity elements are: Cu (0.14), Ni, Mo, Al (0.07), N, O, Ti, and Nb. Balance is Fe.
 Metals 2017, 7, 190 5 of 29

Metals 2017, 7, Table

190 3. Thermophysical properties of materials. See Figure 1 for k, ρ and Cp of steel. 5 of 29

Material kt (W/m·K) ρ (kg/m3) Cp (J/kg·K) μ (mPa·s) Tsat (°C) Ms (°C) M50 (°C) M90 (°C)
Steel: 51CrV4 k(T)3. Thermophysical
Table ρ(T) properties
Cp(T) - Figure 1 -for k, ρ and
of materials. See 288Cp of steel.
251 166
Oil: FTR
Material kt (W/m·K) ρ (kg/m3 ) Cp (J/kg·K) µ (mPa·s) T sat (◦ C) M s (◦ C) M 50 (◦ C) M 90 (◦ C)
(Equiquench
Steel: 51CrV4 - k(T)827 at 60 °C
ρ(T) - Cp (T) - °C
7.3 at 60 -
290 288- 251 - 166 -
Oil: FTR
770) Houghto-
(Equiquench 770) - 827 at 60 ◦C - 7.3 at 60 ◦C 290 - - -
Quench®
Houghto-Quench®

Figure
Figure11 shows
shows anan abrupt
abrupt change
change of
of properties
properties at
at temperature
temperature of
of martensite
martensite start,
start, M
Mss==288
288°C.
◦ C.
Density decreases, while thermal conductivity and heat capacity increase. This later property
Density decreases, while thermal conductivity and heat capacity increase. This later property includes
includes
the latentthe latent
heat heat of solid-phase
of solid-phase transformation
transformation from austenite
from austenite to martensite.
to martensite.

Figure
Figure 1.1. Thermophysical
Thermophysicalproperties
propertiesofofsteel
steelDIN
DIN51CrV4
51CrV4estimated
estimatedfrom
from JMatPro
JMatProsoftware
® ®
softwareforfor
a
cooling rate of 56 ◦
°C/s from austenizing to ambient temperature. Notice how
a cooling rate of 56 C/s from austenizing to ambient temperature. Notice how properties changeproperties change
abruptly
abruptly at at temperature
temperature of of martensite
martensite start,
start, M
Mss == 288 ◦ C. Heat
288 °C. Heat capacity
capacity considers
considers the
the solid-phase
solid-phase
transformation
transformationlatentlatentheat.
heat.

The heat capacity of the transforming steel was computed from the following equation,
The heat capacity of the transforming steel was computed from the following equation,
∂ x H +x H
Cp = a a m m , (1)
∂( xa Ha ∂T+ xm Hm )
Cp = , (1)
where x is the mass fraction, H is the enthalpy per ∂Tunit mass, and subscripts “a” and “m” refer
austenite
where x is and
themartensite, respectively.
mass fraction, H is theNote that both
enthalpy quantities
per unit mass, depend on temperature;
and subscripts “a” and “m”the mass
refer
fraction
austenitedepends on the kinetics
and martensite, of transformation,
respectively. while
Note that both enthalpydepend
quantities depends onon the heat capacity
temperature; the mass of
the corresponding
fraction depends on phase.
the kinetics of transformation, while enthalpy depends on the heat capacity of the
corresponding phase.
2.2. Plant Temperature Measurements
2.2. Plant Temperature Measurements
2.2.1. The Quenching System
2.2.1. The Quenching System
The study case was an oil quenching tank with a carousel provided with tools to shape and hold
The springs,
steel leaf study case
17 was
mm an oil quenching
thick, tank and
120 mm length with75a carousel
mm wide, provided withare
while they tools to shapeinand
immersed thehold
oil,
steel leaf springs, 17 mm thick, 120 mm length and 75 mm wide, while they are immersed
as shown schematically in Figure 2a. In preparation for quenching, two preheated straight leaf in the oil, as
shown schematically
springs are manuallyinaccommodated
Figure 2a. In preparation
in a shaperfortable
quenching, two preheated
at position straight leaf
9 of the carousel. springs
A head of
are manually accommodated in a shaper table at position 9 of the carousel. A head
downholders then descends over the leaf springs, holding them against the shaper table, as shown in of downholders
thendrawings
the descendsofover the leaf
Figure 2b. springs, holding
The carousel themand
rotates against the shaper
submerges the table, as shown
springs in the
in the oil, drawings
stopping at
of Figure 2b. The carousel rotates and submerges the springs in the oil, stopping at position number
Metals 2017, 7, 190 6 of 29
Metals 2017, 7, 190 6 of 29

position
1. numbercharges
The operator 1. The another
operatortwo charges anotherattwo
leaf springs leaf springs
position at position
9 and repeats 9 and repeats
the process. There arethe
process.
five Therepositions
carousel are five carousel
immersed positions immersed
in the oil; see numbersin the1oil;
to see
5 innumbers
Figure 2a.1 to 5 in Figure
When the leaf2a. When
springs
the leaf
reach springs
the sixth reach thehead
one, the sixth of
one, the head of downholders
downholders releasessprings,
releases the shaped the shapedandsprings,
a robotand
arma takes
robot
arm takes
them out tothem out to a transporting
a transporting band. The band. The dimensional
dimensional specifications
specifications of the quenching
of the quenching system aresystem
given
areTable
in given4.inATable 4. Aoil
constant constant
flowrate oilenters
flowrate entersthe
through through
feeder the
tube feeder tubeflows
and then and then flowstwo
through through
rows
two
of rows of
nozzles tonozzles
generatetoimmersed
generate immersed
jets that feedjetsthe
thattank.
feed The
the tank. Theout
oil flows oil flows outtank
from the fromtothe tank to
a cooling
a cooling
tower, tower,
which whichitprevents
prevents it from overheating.
from overheating. When this tank When this tankthe
increased increased the productivity
productivity of automotive of
automotive leaf springs, an unacceptable distortion of
leaf springs, an unacceptable distortion of the product was found. the product was found.

(a)

(b)
Figure 2.
Figure 2. (a)
(a) Schematic
Schematic representation
representation of
of the
the oil
oil quenching
quenching tank.
tank. Carousel
Carousel is
is fed
fed with
with leaf
leaf springs
springs at
at
position 9. (b) Sequence for shaping a leaf spring before immersion in the tank.
position 9. (b) Sequence for shaping a leaf spring before immersion in the tank.

Table 4.
Table 4. Dimensional
Dimensional specifications
specifications of
of the
the quenching
quenching system.
system.

Tank Component Length (m) Width (m) Height (m) Diameter (m)
Tank Component
Feeder, tube Length- (m) Width
- (m) Height- (m) Diameter (m)
0.197
Feeder, hexahedron
Feeder, tube 1.78- 0.20- 0.16
- 0.197 -
Feeder, hexahedron
Nozzle 1.78
0.045 0.20
- 0.16- -0.0254
Nozzle
Tank (main body) 0.045
2.96 2.11- -
2.74 0.0254-
Tank (main
Virtual boxbody) 2.96
0.22 (thickness) 2.11
1.46 2.74
0.57 - -
Virtual box
Tank exit orifice 0.22 (thickness)
- 1.46
- 0.57- - 0.250
Tank exit orifice - - - 0.250

Previous research work [21] presented some clues to determine the reason for this distortion.
Previous
The objective of research workwork
the present [21] presented some an
was to develop clues to determine
efficient the reason
fluid-dynamic for this
analysis distortion.
aimed to gain
The objective ofofthe
understanding thepresent
coolingwork was
process ofto
thedevelop an efficient
leaf springs. fluid-dynamic
This knowledge analysis
was used aimedchanges
to analyze to gain
understanding
in the tank designof the cooling process
to improve of the leaf springs.
its fluid-dynamics This knowledge
and ultimately was
the rate of used
heat to analyze
transfer changes
from the steel
in the
leaves. tank design to improve its fluid-dynamics and ultimately the rate of heat transfer from the
steel leaves.
2.2.2. Diagnostic Thermal Analysis
2.2.2. Diagnostic Thermal Analysis
The first step in this approach was to characterize the thermal behavior of the leaf spring during
The first step
its quenching. in this
Figure 3a approach was to
schematically characterize
shows the of
a drawing thermal
a leaf behavior of the
spring that wasleaf
cutspring during
into three its
parts:
quenching.
fixed, looseFigure 3a schematically
and center shows
sections. They werea drawing of a leafbyspring
instrumented that was
screwing cut thermocouples
K-type into three parts: to
fixed,
the
loose and center sections. They were instrumented by screwing K-type thermocouples
surface of the leaf in the indicated points. The tip of the thermocouple was hold against the surface to the surface of
of the spring. An instrumented whole leaf spring had an important drawback when manually
removing it from the furnace to the carousel. The thermocouple wires delayed this operation, and the
Metals 2017, 7, 190 7 of 29

the leaf in the indicated points. The tip of the thermocouple was hold against the surface of the spring.
Metals 2017, 7, 190
An instrumented whole leaf spring had an important drawback when manually removing it 7from of 29

the furnace to the carousel. The thermocouple wires delayed this operation, and the leaf temperature
leaf temperature dropped excessively before the section was immersed in the oil. In contrast, the
dropped excessively
smaller before
sections were the section
heated was immersed
in a portable in the
muffle and oil.moved
easily In contrast,
to the the smaller
carousel. sections
Figure were
3b shows
heated in a portable muffle and easily moved to the carousel. Figure 3b shows a photograph
a photograph during accommodation of the preheated and instrumented fixed section onto the during
accommodation
shaper table.ofFigure
the preheated and instrumented
3c is a photograph captured fixed
duringsection onto the shaper
the preheating table. Figureloose
of the instrumented 3c is
a photograph captured
section in a muffle. during the preheating of the instrumented loose section in a muffle.

(a)

(b) (c)

FigureFigure
3. (a)3.Schematic
(a) Schematic representation
representation of sectioned
of sectioned leafleaf spring
spring indicatingthermocouple
indicating thermocouplelocations
locations in
in
millimeters, and photographs of (b) preheated fixed section being placed in shaper table and
millimeters, and photographs of (b) preheated fixed section being placed in shaper table and (c) loose (c) loose
section being preheated in a muffle.
section being preheated in a muffle.

The leaf spring temperature measurements were carried out three times for every leaf section.
The
Sinceleaf spring
there were temperature measurements
eight thermocouples were
in the three carriedthen
sections, outwethree times24
obtained forcooling
every curves.
leaf section.
Since there were eight thermocouples in the three sections, then we obtained 24 cooling curves.
2.3. Laboratory Temperature Measurements
2.3. Laboratory Temperature Measurements
Laboratory thermal analyses were carried out also on DIN 51CrV4 steel samples quenched with
Laboratory
the same oil thermal analyses
used in plant and were
undercarried
controlledout and
alsorecorded
on DIN fluid
51CrV4 steel samples
velocities. Figure 4a quenched
shows a
with schematic
the samerepresentation
oil used in plant of the and underrig
laboratory controlled
to determine andthe recorded fluid
rate of heat velocities.
removal Figure
from steel during4a
shows a schematic This
oil-quenching. representation
system wasofinspired
the laboratory rig toJominy
in the classic determine the rate of
End-Quench testheat
[22]removal
which usesfroma
steel controlled
during oil-quenching.
water flow rateThis system was
to characterize inspired
steel in the classic
hardenability. Jominy
This system End-Quench
seems test [22]
to be a convenient
which choice
uses to relate the already
a controlled water flowknownrateisothermal fluid flow
to characterize steelfield of jets impinging
hardenability. over aseems
This system flat surface
to be
with the choice
a convenient rate of to
heat transfer
relate during quenching,
the already as it is discussed
known isothermal fluid flow in field
Section 2.3. impinging
of jets In our case,over
the
experimental set-up includes an oil reservoir that feeds a 2 HP (1.49 kW) gear
a flat surface with the rate of heat transfer during quenching, as it is discussed in Section 2.3. In our pump with an oil
flowrate capacity of 40 L/min. The target oil flowrate was controlled
case, the experimental set-up includes an oil reservoir that feeds a 2 HP (1.49 kW) gear pump withusing two valves and was
an oilmeasured
flowrate using a turbine
capacity flowmeter
of 40 L/min. The(FLR6115D-BSPP,
target oil flowrate Omega Engineering
was controlled Inc., two
using Norwalk,
valvesCT,
and USA).
was
An electric resistance of 4000 W and a K-type thermocouple are immersed
measured using a turbine flowmeter (FLR6115D-BSPP, Omega Engineering Inc., Norwalk, CT, USA). in the oil reservoir. Both
are connected to a controller unit which regulates the power consumed by the resistance to maintain
An electric resistance of 4000 W and a K-type thermocouple are immersed in the oil reservoir. Both are
the oil temperature at a set-point value. A second K-type thermocouple is connected to the line to
connected to a controller unit which regulates the power consumed by the resistance to maintain the
register the oil temperature at a location near the nozzle. This nozzle is 12.7 mm internal diameter
oil temperature at a set-point value. A second K-type thermocouple is connected to the line to register
and its tip is separated from the sample surface by the same distance. The sample is a disk having a
the oil temperature at a location near the nozzle. This nozzle is 12.7 mm internal diameter and its tip is
diameter of 25.4 mm and a thickness of 8.5 mm, which corresponds to the semi-thickness of the
separated from the sample surface by the same distance. The sample is a disk having a diameter of
studied leaf spring.
25.4 mm and a thickness
Figure of 8.5amm,
4a also shows which
moving corresponds
baffle that diverted to the
the semi-thickness
upcoming oil while of the thestudied leaf moved
probe was spring.
from the muffle to a position on top of the nozzle. Once the probe was in this position, the baffle axis
Metals 2017, 7, 190 8 of 29

Figure
Metals 2017, 7,4a
190also shows a moving baffle that diverted the upcoming oil while the probe was moved 8 of 29
from the muffle to a position on top of the nozzle. Once the probe was in this position, the baffle
waswas
axis rotated and the
rotated andoil
theimpacted the pre-heated
oil impacted steel disk.
the pre-heated steelTemperature was measured
disk. Temperature with a K-type
was measured with
athermocouple
K-type thermocouple(TJ36-CAXL-116-G-12, Omega Engineering
(TJ36-CAXL-116-G-12, Inc., Norwalk,
Omega Engineering Inc.,CT, USA) that
Norwalk, CT,wasUSA)hold by
that
compression
was fittings against
hold by compression the bottom
fittings againstofthe
a hole machined
bottom from
of a hole the backfrom
machined of thethe
disk.
back Theof bottom
the disk.of
this hole was at a nominal distance of 1.8 mm from the impact surface. The thermocouple
The bottom of this hole was at a nominal distance of 1.8 mm from the impact surface. The thermocouple sheath
diameter
sheath was 1.59
diameter mm
was andmm
1.59 theand
diameter of the drill
the diameter wasdrill
of the 1.7 mm.
wasThe instrumented
1.7 mm. disk was heated
The instrumented disk
up in a cylindrical muffle locally built over the platform of a ceramic radiant
was heated up in a cylindrical muffle locally built over the platform of a ceramic radiant full full cylinder heater
(CRFC-46/240-C-A,
cylinder Omega Engineering
heater (CRFC-46/240-C-A, Omega Inc., Norwalk, Inc.,
Engineering CT, USA) and CT,
Norwalk, withUSA)
the aid
andofwith
a temperature
the aid of
acontroller.
temperature controller.

(a) (b)
Figure4.4.(a)
Figure (a)Schematic
Schematicrepresentation
representationofoflaboratory
laboratoryoil
oilclosed-loop
closed-loopforfordetermination
determinationof ofthe
therate
rateofof
heat removal during oil-quenching steel samples; (b) detailed view of the probe that holds a steel
heat removal during oil-quenching steel samples; (b) detailed view of the probe that holds a steel disc disc
(marroncolor)
(marron color)and
andwith
withsub-superficial
sub-superficialthermocouple
thermocouple(black
(blackcolor).
color).Dimensions
Dimensionsareareininmm.
mm.

The sample was hold by an artifact that is shown schematically in Figure 4b. It consists of a
The sample was hold by an artifact that is shown schematically in Figure 4b. It consists of a holder
holder that was machined to be connected to a tube, on one side, and to hold the steel disk on the
that was machined to be connected to a tube, on one side, and to hold the steel disk on the other side,
other side, using three screws that pressed the lateral surface of the disk.
using three screws that pressed the lateral surface of the disk.
The steel disk was intended to represent a semi-thickness portion of the leaf spring. Therefore,
The steel disk was intended to represent a semi-thickness portion of the leaf spring. Therefore,
its lateral and back surface areas should be adiabatic, leaving only the front surface for heat removal.
its lateral and back surface areas should be adiabatic, leaving only the front surface for heat removal.
After testing several different holder designs, this artifact allowed us to approach this adiabatic
After testing several different holder designs, this artifact allowed us to approach this adiabatic
condition, as it will be shown in Section 4.
condition, as it will be shown in Section 4.
The test procedure started to verify co-axial alignment between nozzle and holder. An
The test procedure started to verify co-axial alignment between nozzle and holder. An aluminum
aluminum hollow cylinder was machined ad-hoc to fit the diameters of nozzle and holder. Therefore,
hollow cylinder was machined ad-hoc to fit the diameters of nozzle and holder. Therefore, this cylinder
this cylinder allowed us to adjust co-axial alignment as well as distance between nozzle and holder.
allowed us to adjust co-axial alignment as well as distance between nozzle and holder. Thereafter,
Thereafter, the instrumented steel disk was heated up to austenitization temperature. At the same
the instrumented steel disk was heated up to austenitization temperature. At the same time, the oil was
time, the oil was also being heated up at a desired set-point. Before the probe reached the target
also being heated up at a desired set-point. Before the probe reached the target temperature, the pump
temperature, the pump was on to establish a given oil flowrate. The baffle was located above of the
was on to establish a given oil flowrate. The baffle was located above of the nozzle, diverting the oil
nozzle, diverting the oil flow. The disk reached the wished temperature and was hold at this
flow. The disk reached the wished temperature and was hold at this temperature for 3 min. The probe
temperature for 3 min. The probe was taken out from the muffle to place it above the nozzle and the
was taken out from the muffle to place it above the nozzle and the baffle axis was rotated. Oil impacted
baffle axis was rotated. Oil impacted the preheated disk and the thermal history was recorded. The
the preheated disk and the thermal history was recorded. The Plexiglas box that contained the nozzle
Plexiglas box that contained the nozzle was closed from the top to prevent oil flaming during the test.
was closed from the top to prevent oil flaming during the test. The experimental conditions include the
The experimental conditions include the study of the effect of initial steel temperature, oil velocity
study of the effect of initial steel temperature, oil velocity and oil temperature on the heat flow curve,
and oil temperature on the heat flow curve, and are shown in Table 5. Initial temperature of steel is
and are shown in Table 5. Initial temperature of steel is just at the start of oil-quenching, i.e., after the
just at the start of oil-quenching, i.e., after the air-cooling period. The impact velocity of oil jet on the
air-cooling period. The impact velocity of oil jet on the surface of the disk, V I , was corrected for the
surface of the disk, VI, was corrected for the gravity acceleration, g, because the flow is upward. The
expression was obtained from Bernoulli’s equation and is given by,
1/2
VI = V2N - 2gh , (2)
Metals 2017, 7, 190 9 of 29

gravity acceleration, g, because the flow is upward. The expression was obtained from Bernoulli’s
equation and is given by,
 1/2
VI = VN2 − 2gh , (2)

where V N is the oil velocity at the exit of the nozzle (=Q/AN ) which depends on the oil flowrate, Q,
and the cross section area of the nozzle, AN ; and h is the distance between the nozzle exit and the
impact surface.

Table 5. Laboratory conditions for oil-quenching experiments.

Initial Temperature T I (◦ C) Impact Velocity V I (m/s) Oil Temperature T f (◦ C)

800
850
0.9 60
880
900
0.4
0.9
850 60
1.3
2.0
50
850 0.9
60

2.4. Connection between Fluid Flow and Heat Transfer

Relationships between the rate of fluid flow and the associated rate of heat transfer between
a solid surface and a fluid are not new. Reynolds-Colburn analogy between shear stress and heat
transfer is very useful in obtaining a first approximation for heat transfer in cases in which the shear
stress is “known”. This analogy is not a physical law but it is a convenient analytical tool based on the
hypothesis of the mechanism of heat transfer and shear stress [23]. It states the following relationship
between convection and transferred to wall quantities:

heat flux to wall momentum flux to wall

= , (3)
convection heat flux convection momentum flux
which can be represented by the equation,

qw τw
= 2, (4)
ρvf Cp ( Tf − Tw ) ρvf

where ρ, vf , Cp and Tf are fluid magnitudes: density, bulk velocity, heat capacity and bulk temperature,
respectively; while qw , τ w and Tw are the heat flux, shear stress and temperature at the wall. In this
analogy, the shear stress is unrelated to heat transfer, that means it can be computed from isothermal
fluid flow. The use of this analogy to approach boiling heat transfer in quenching tanks has not
been reported before, at least according with the best knowledge of the authors. In this paper, it is
proposed an efficient analysis aimed to elucidate the fluid flow field and the rate of heat flux removed
from steel pieces quenched in long standing tanks. The isothermal fluid flow calculations predicted
detailed shear stress distributions on the surface of leaf springs, while the analyses of heat fluxes
determined from plant and laboratory data were used to propose an empirical equation based on the
Reynolds-Colburn analogy.

3. Formulation of the Model

As it was explained in Section 2.4, an analysis based on Reynolds-Colburn analogy was developed
in this work. Therefore, heat transfer and oil vapor evolution during quenching were sacrificed in the
numerical model for the sake of an efficient algorithm. However, boiling heat transfer was addressed
Metals 2017, 7, 190 10 of 29

using plant and laboratory measurements. The objective of this model was to obtain a detailed picture
of the isothermal oil flow field in the tank, particularly near the surface of the leaf springs. The shear
stress and pressure imposed by the flowing oil on the wall of the spring are results that depend on the
oil velocity profile near its surface.

3.1. Governing Differential Equations

The model represents the numerical solution of the Three-Dimensional (3D) momentum and
mass equations for isothermal flow under steady state conditions. Turbulence was also considered by
computing time-averaged quantities through the following governing equations,
Continuity:
∂ρu j
= 0, (5)
∂x j
Navier-Stokes:
∂

ρui u j + pδij − τij − λij = ρg j , (6)
∂x j
Turbulence kinetic energy, k:
"   #
∂ µt ∂k
ρku j − µ + = G − ρε, (7)
∂x j σk ∂x j

Rate of energy dissipation, ε:

" #
ε2
 
∂ µt ∂ε
ρεu j − µ + = (ρC1 S)ε − (ρC2 ) √ , (8)
∂x j σε ∂x j k + υε

where u is the time-averaged local velocity; λij is the Reynolds stress tensor; x is the coordinate position;
ρ, µ, and υ are the density, viscosity, and kinematic viscosity of the fluid, respectively; and G is the rate
of generation of turbulence kinetic energy due to the mean velocity gradients and is defined together
with other quantities in Table 6. The oil properties are included in Table 3.

Table 6. Variables and constants appearing in turbulence equations [24].

C1 C2 G η S Sij σk σε
h i q  
η 1 ∂u j
max 0.43, η+5 1.9 µt S2 S εk 2Sij Sij 2 ∂xi + ∂ui
∂x j 1 1.2

The Reynolds stress is obtained from the Boussinesq approach and considers incompressible flow.
The respective equation is:
2
λij = 2µt Sij − ρkδij , (9)
3
where δij is the Kronecker delta (=0 for i 6= j) and (=1 for i = j). Turbulence viscosity, µt , was determined
from the following equation:
k2
µt = ρCµ , (10)
ε
This quantity was computed using the realizable k-ε model [25]. This model differs from the
standard k-ε model in two important ways: (1) The realizable model contains an alternative formulation
for the turbulent viscosity, and (2) a modified transport equation for the dissipation rate, ε, has been
derived from an exact equation for the transport of the mean-square vorticity fluctuation. The term
“realizable” means that the model satisfies specific mathematical constraints on the Reynolds stresses,
consistent with the physics of turbulent flows. The realizable model considers that Cµ is not a constant
but is a function of the main strain and rotation rates and the angular velocity of the system rotation and
Metals 2017, 7, 190 11 of 29

the turbulence fields, k and ε [24]. Neither the standard model nor the RNG (ReNormalization Groups)
model is realizable. This model has been extensively validated for a wide range of flows [25,26],
Metals 2017, rotating
including 7, 190 homogeneous shear flows, free flows including jets and mixing layers, channel 11 of 29
and boundary layer flows, and separated flows. For all these cases, the performance of the model has
model.
been foundEspecially noteworthybetter
to be substantially is thethan factthat
thatofthetherealizable
standard k-ε model resolves
model. the round-jet
Especially noteworthy anomaly;
is the
that is, it predicts the spreading rate for axisymmetric jets and that for
fact that the realizable model resolves the round-jet anomaly; that is, it predicts the spreading rate for planar jets. This is particularly
useful in our case
axisymmetric to obtain
jets and a realistic
that for planarflow jets. field
This in is the immerseduseful
particularly oil jetsincoming
our case from multiple
to obtain nozzles.
a realistic
The in
flow field computational
the immerseddomain includedfrom
oil jets coming onlymultiple
the space occupied by the oil inside the feeder and
nozzles.
tank.TheThis domain was divided into three
computational domain included only the space occupied subdomains: (1) the feeder, by(2)thethe
oiltank
insideandthe (3)feeder
five virtual
and
boxes at the five carousel positions where leaves are immersed. The
tank. This domain was divided into three subdomains: (1) the feeder, (2) the tank and (3) five virtual reason for this is to improve the
computer
boxes at theefficiency
five carousel by solving
positions each subdomain
where leaves are oneimmersed.
at a time rather than the
The reason forfull
thisdomain in a single
is to improve the
computer efficiency by solving each subdomain one at a time rather than the full domain in a that
run. This approach is valid because the oil flows downstream. This supports the assumption single oil
velocity
run. This profile
approach at the exit ofbecause
is valid nozzlesthe does oil not
flowsdepend on the velocity
downstream. pattern the
This supports in the tank but depends
assumption that oil
only on the flowrate at the inlet cross section area of the tube feeder.
velocity profile at the exit of nozzles does not depend on the velocity pattern in the tank but depends Therefore, our first step was to
solve for the fluid flow in the feeder by itself, whose boundary conditions
only on the flowrate at the inlet cross section area of the tube feeder. Therefore, our first step was to and meshing are shown in
Figure
solve for5a.
theInfluid
a second
flow in step,
thethe velocity
feeder profiles
by itself, whose computed
boundary at the exit of the
conditions andrespective
meshing are nozzles
shown werein
imposed as an inlet boundary condition when solving the governing
Figure 5a. In a second step, the velocity profiles computed at the exit of the respective nozzles were equations for the whole tank.
During this
imposed as an step,
inlettheboundary
shaper table, the head
condition when of downholders
solving the governingand the leaf springsfor
equations weretheabsent
wholeexcept
tank.
for simple flat plates that replaced them. The justification for this
During this step, the shaper table, the head of downholders and the leaf springs were absent except for was to numerically capture the
overall fluid flow in the tank by carefully considering the most evident
simple flat plates that replaced them. The justification for this was to numerically capture the overall flow resistance and free flow
areas.flow
fluid Figurein the5b tank
shows bythe corresponding
carefully considering boundary
the most conditions
evident flowand meshing.
resistanceTo and compute
free flow theareas.
fluid
flow field
Figure at thethe
5b shows neighborhood
corresponding of boundary
leaf springs, a third subdomain
conditions and meshing. was Toformed
compute bythefive virtual
fluid flowboxes
field
that contained the immersed springs together with the detailed
at the neighborhood of leaf springs, a third subdomain was formed by five virtual boxes that contained geometry of the shaper table and
downholders.
the immersed springs The sizetogether
of these withboxesthe was chosengeometry
detailed to includeofconveniently
the shaper tablethe leaf and springs together
downholders.
with the shaping tools. This choice was also validated by showing
The size of these boxes was chosen to include conveniently the leaf springs together with the shaping that computed results were
independent from box size. The computed velocity distribution
tools. This choice was also validated by showing that computed results were independent from from the tank solution was
considered
box size. Theascomputed a boundary condition
velocity for the from
distribution facesthe that hadsolution
tank a net positive inflow, whereas
was considered as a boundarya zero
reference pressure was considered a boundary condition for those faces
condition for the faces that had a net positive inflow, whereas a zero reference pressure was considered that had a net outflow. Figure
a5c indicatescondition
boundary the corresponding
for those boundary
faces that conditions and meshing.
had a net outflow. FigureNotice that numerical
5c indicates meshing is
the corresponding
active only
boundary on the subdomain
conditions and meshing. that isNotice
being solved for, thusmeshing
that numerical avoidingisdiscretization
active only on of the
the subdomain
solid bodies
to save computer memory and time. Regarding the non-slip wall
that is being solved for, thus avoiding discretization of the solid bodies to save computer memory boundary condition, the adopted
strategy
and time. withRegarding the normal
the non-slip dimensionless
wall boundary distance fromthe
condition, the wall, strategy
adopted y+, waswith to theconsider
normala
hydrodynamically smooth wall with roughness height, K s = 0, and default roughness constant Cs =
dimensionless distance from the wall, y+, was to consider a hydrodynamically smooth wall with
0.5 that had
roughness no effect
height, Ks = 0,inand thedefault
computation.
roughness These coefficients
constant Cs = 0.5 are
thatdefined for the
had no effect Law-of-the
in the computation.Wall
Modified
These for Roughness
coefficients are defined[27].for the Law-of-the Wall Modified for Roughness [27].

Zero pressure at nozzle outlets Zero shear stress and zero normal
velocity at oil free surface

Zero pressure
at tank exit

Velocity at
nozzles outlet
Inlet velocity

(a) (b)

Figure 5. Cont.
Metals 2017, 7, 190 12 of 29
Metals 2017, 7, 190 12 of 29

Zero pressure for

outlet faces

Velocity distribution at inlet faces

(c)
Figure 5.
Figure 5. Subdomains
Subdomains andand boundary
boundary conditions:
conditions: (a) discretized feeder,
(a) discretized (b) discretized
feeder, (b) discretized tank,
tank, and
and (c)
(c)
discretized virtual box, all with cut view planes. The footprint of shaper table, downholders and
discretized virtual box, all with cut view planes. The footprint of shaper table, downholders and leaf leaf
spring is
spring is shown.
shown.

3.2. Solution Method

3.2. Solution Method
The governing equations were solved by the finite volume method [28] implemented in the
The governing equations were solved by the finite volume method [28] implemented in the
ANSYS FluentTM TM (version 16, ANSYS Inc., Canonsburg, PA, USA, 2015), code. Every subdomain was
ANSYS Fluent (version 16, ANSYS Inc., Canonsburg, PA, USA, 2015), code. Every subdomain
divided into several regions to reduce skewness, improving mesh quality. The underrelaxation factor
was divided into several regions to reduce skewness, improving mesh quality. The underrelaxation
was manually changed throughout the solution process to reduce the ‘scaled’ residuals for mass,
factor was manually changed throughout the solution process to reduce the ‘scaled’ residuals for mass,
momentum and turbulence equations [28] to values that were less than 10−3.
momentum and turbulence equations [28] to values that were less than 10−3 .

4. Results
4. Results and
and Discussion
Discussion

4.1. Laboratory
4.1. Laboratory Thermal
Thermal Analysis
Analysis
Laboratory thermal
Laboratory thermal data
data for each experimental condition were obtained from at least 5 replicas.
Everycurve
Every curvewas
was numerically
numerically smoothed
smoothed before
before any further
any further processing.
processing. It was It wastheused
used the average
moving moving
average method, choosing 5 points and averaging them according with
method, choosing 5 points and averaging them according with the equation: the equation:
j=i+2
i = ∑jj ==i -i 2+w2j Tj ,
Tav (11)
Tiav = ∑ j = i − 2 w j Tj , (11)
where i and j are integers that denote time node, the weighting factors wj are equal to 0.1, 0.2, 0.4, 0.2
and 0.1i and
where for j j=are
i − integers
2, i − 1, i,that
i + 1denote
and i +time node, the weighting
2, respectively. factors wj sweeps,
After 12 smoothing are equalthe to temperature
0.1, 0.2, 0.4,
0.2 and did
curves 0.1 for = i − 2, appreciably
notj change i − 1, i, i + 1 and i + 2,
but the respectively.
cooling Afterimproved
rate curves 12 smoothing sweeps, the
substantially, bytemperature
minimizing
curves
their numerical noise. The cooling rate was estimated from the temperature smoothedminimizing
did not change appreciably but the cooling rate curves improved substantially, by data using
their numerical noise. The cooling rate was estimated from the temperature
central finite differences and no further smoothing was applied to these curves. Temperature curves smoothed data using
central finite differences
for all replicas were smoothed and noand further smoothing
then they was applied
were averaged; a 95%to confidence
these curves. Temperature
interval curves
was computed
for all replicas were smoothed and then they were averaged; a 95% confidence
based on standard deviation and a t-student distribution. Figure 6 shows the average temperature- interval was computed
based on standard
time curve together deviation
with itsand 95% a t-student
confidence distribution.
interval. TheFigure 6 shows
limits the average
(average ± σ) aretemperature-time
indicated by the
curve together with its 95% confidence interval. The limits (average
upper and lower curves. These results correspond to an oil impact velocity and temperature± σ) are indicated by the upperofand0.9
lower curves. These results correspond to initial
an oil impact velocity andof temperature of 0.9 m/sshows ◦ C,
and 60that
m/s and 60 °C, respectively, and with steel temperature 850 °C. The figure
respectively, ◦ C. The figure shows that temperature decreases
temperatureand with initial
decreases verysteelfasttemperature
until it reachesof 850~300 °C. This is expected since the oil boiling
very fast until is
it reaches ◦ is 290 ◦by
temperature 290 °C. ~300
BelowC.this Thistemperature,
is expected since the oil
boiling boiling
stops andtemperature
cooling proceeds C. Below this
one-phase
temperature, boiling stops and cooling proceeds by one-phase convection.
convection. The corresponding cooling rate curves are also shown in the figure. They reach a The corresponding cooling
rate curves are
maximum valuealsoatshown
a high in the figure. They
temperature. reach shoulder
The early a maximum valuebefore
formed at a high
thetemperature. The early
peak is attributed to a
shoulder formed before the peak is attributed to a brief film boiling regime.
brief film boiling regime. Notice that uncertainty for the cooling rate is larger at temperatures above Notice that uncertainty
for ◦ C, at which the cooling rate reaches its
700the
°C, cooling
at whichrate the is largerrate
cooling at temperatures
reaches its maximum above 700 value.
maximum value.
Metals 2017, 7, 190 13 of 29
Metals
Metals2017,
2017,7,7,190
190 13
13ofof29
29

Figure
Figure6.
Figure 6.6.Average thermal
Averagethermal
Average evolution
thermalevolution and
evolutionand its
itscorresponding
andits cooling
coolingrate
correspondingcooling
corresponding rateobtained
rate obtainedfrom
obtained fromoil-quenching
from oil-quenching
oil-quenching
laboratory
laboratory experiments.
laboratoryexperiments. Impact
experiments.Impact
Impactoiloil velocity
oilvelocity and
velocityand oil
andoil temperature
oiltemperature were
temperaturewere 0.9
were0.9 m/s
0.9m/s and
m/s and 60
and 60 °C,
60 °C,
◦ C,respectively.
respectively.
respectively.
AA95%
A 95% confidence
95%confidence interval
confidenceinterval isisindicated
intervalis indicated for
indicatedfor both
forboth curves,
bothcurves, temperature
curves,temperature and
temperatureand cooling
andcooling rate.
coolingrate.
rate.

Figure
Figure 777 shows
shows
shows the the effect
the effect of
effect of the
of the jet
the jet impact
jet impact velocity
impact velocity
velocity onon the
on the average
the average temperature
average temperature evolution
temperature evolution
evolution
Figure
determined
determined fromfrom laboratory
from laboratory data.
laboratory data. Similarly,
data. Similarly,
Similarly, toto Figure
to Figure 6,
Figure 6, each
6, each curve
each curve includes
curve includes its
includes its 95%
its 95% confidence
95% confidence
confidence
determined
interval.
interval.ItItcan
can be
be seen
seen ininFigure
Figure 77that
that during
during the
theinitial
initialcooling
cooling period
period all
allcurves
curves are
are very
veryclose each
closeclose
each
interval. It can be seen in Figure 7 that during the initial cooling period all curves are very
other.
other. However,
However, these
these curves
curves clearly
clearly separate
separate each
each other
other at
at temperatures
temperatures below
below ~350
~350 °C
°C where
where even
even
each other. However, these curves clearly separate each other at temperatures below ~350 ◦ C where
their
their confidence
confidence intervals
intervals do do not
not mutually
mutually intersect.
intersect. Finally,
Finally, curves
curves become
become closer
closer each
each other
other atat
even their confidence intervals do not mutually intersect. Finally, curves become closer each other at
temperatures
temperatures below
below 100
100 °C
°C
◦ as
asthey
they asymptotically
asymptotically approach
approach the
theoil
oiltemperature.
temperature. The
The effects
effectsof
ofthe
theoil
oil
temperatures below 100 C as they asymptotically approach the oil temperature. The effects of the
temperature
temperature and and thethe initial
initial steel
steel temperature
temperature were were not
not clear
clear from
from the
the temperature-time
temperature-time curves.
curves.
oil temperature and the initial steel temperature were not clear from the temperature-time curves.
However,
However,the theboiling
boilingcurves
curveswerewereaabetter
betterchoice
choiceto toappreciate
appreciatethese
theseeffects.
effects.
However, the boiling curves were a better choice to appreciate these effects.

Figure
Figure 7.7. Effect
Effect of
Effect of jet
of jet impact
jet impact velocity
impact velocity on
velocity on the
on the average
the average thermal
average thermal evolution
thermal evolution and
evolution and corresponding
corresponding 95%
corresponding 95%
confidence
confidenceinterval.
interval.
Metals 2017, 7, 190 14 of 29

The determination of the heat flux removed by the oil jet, impacting one face of the disk sample,
was carried out solving the IHCP (Inverse Heat Conduction Problem) for one-dimensional transient
heat conduction using Beck’s algorithm [29] implemented in the Fortran code CONTA (Sandia National
Laboratories, Albuquerque, NM and Michigan State University, East Lansing, MI, USA, 1986).
This analysis is valid when lateral surface of the disk is adiabatic. Furthermore, the boundary
condition on the back face should be known. The adiabatic condition is the most convenient choice
for our case. Therefore, the sample represented the thermal behavior of the semi-thickness of the leaf
spring. Verification of each solution was carried out by solving the Direct Heat Conduction Problem
Metals 2017, 7, 190 14 of 29
(DHCP) imposing this evolving heat flux as a boundary condition and solving the heat conduction
equationThe using the finite volume method [28] implemented in the Fortran code CONduction in DUCTs
determination of the heat flux removed by the oil jet, impacting one face of the disk sample,
(CONDUCT)
was carried (CONDUCT,
out solving the SuhasIHCP V. (Inverse
Patankar, University
Heat Conduction of Minnesota,
Problem) for Minneapolis,
one-dimensional MN,transient
USA, 1988).
The heat
computed temperature evolution agreed with both, measured
conduction using Beck’s algorithm [29] implemented in the Fortran code CONTA (Sandiaand IHCP solution, specially above
120 ◦National
C. Laboratories, Albuquerque, NM and Michigan State University, East Lansing, MI, USA,
Figure
1986) 8a analysis
. This presentsis thevalideffect
whenoflateral
oil velocity
surface on thedisk
of the boiling curves Furthermore,
is adiabatic. determined the from laboratory
boundary
tests.condition
For eachonoilthe velocity,
back face the average
should temperature
be known. evolution
The adiabatic and itsis95%
condition confidence
the most convenientinterval
choice were
for our case. Therefore, the sample represented the thermal behavior
used to compute the corresponding heat flux curves. Therefore, the figure shows an average heat of the semi-thickness of the leaf
flux spring. Verification
curve together with of each
its 95% solution was carried
confidence out by
interval. Thesolving
figure the Directthat
shows Heathigher
Conduction Problem
heat fluxes result
(DHCP) imposing this evolving heat flux as a boundary condition
from higher impact velocities, as expected. The curves present some noise at temperatures below and solving the heat conduction
300 ◦equation
C. The reason using the finite volume method [28] implemented in the Fortran code CONduction in
for this is not clear, but it may be due to the evolution of latent heat during solid
DUCTs (CONDUCT) (CONDUCT, Suhas V. Patankar, University of Minnesota, Minneapolis, MN,
phase transformation that could rise the steel temperature during quenching. Figure 8b shows the
USA, 1988). The computed temperature evolution agreed with both, measured and IHCP solution,
effect of oil temperature on the heat flux. A significate difference between heat fluxes was found in the
specially above 120 °C.◦
temperature Figure range of 300 Cthe
8a presents 450 ◦ C.
toeffect of The heat flux
oil velocity onfortheoil at 60 ◦curves
boiling C is above the onefrom
determined obtained using oil
laboratory
◦
at 50tests.
C. Presumably because oil viscosity decreases with temperature. A lower oil viscosity allows
For each oil velocity, the average temperature evolution and its 95% confidence interval were
an easier
used to growth
compute andthedetachment
corresponding of heat
oil vapor bubbles
flux curves. from the
Therefore, thesteel
figuresurface,
shows an increasing
average heat theflux
rate of
supply
curve of together
cold liquidwithto itsthe
95%hot surface interval.
confidence [30]. Figure 8c shows
The figure shows the effect
that higherof heat
initial steelresult
fluxes temperature
from
just higher
before impact
jet impact. It is seen
velocities, that these
as expected. Thecurves
curveshave
presentthesome
samenoise
shape at but they are shifted
temperatures below 300 to the
°C. left
sideThefor reason for thistemperatures.
lower initial is not clear, butThe it maywidebe due to the evolution
confidence intervalofatlatent heatflux
the heat during solid phase
maxima makes it
transformation
difficult to claim anthat could
effect rise thetemperature
of initial steel temperature during quenching.
over maximum Figure
heat flux; 8b shows
however, thetemperature
in the effect of
rangeoiloftemperature
200 ◦ C to 300 on ◦the
C, aheat
lowerflux.
heatA flux
significate difference
is associated between with
to samples heat afluxes
lowerwas found
initial in the
temperature,
temperature
as expected. range of 300 °C to 450 °C. The heat flux for oil at 60 °C is above the one obtained using
oil at 50 °C. Presumably because oil viscosity decreases with temperature. A lower oil viscosity allows
Regarding validity of the adiabatic boundary condition at non-active faces of the steel disk,
an easier growth and detachment of oil vapor bubbles from the steel surface, increasing the rate of
Figure 4b shows that sample is touched by three screws only. These screws cool down simultaneously
supply of cold liquid to the hot surface [30]. Figure 8c shows the effect of initial steel temperature just
withbefore
the sample and therefore the rate of heat input is minimum. Lateral and back faces cool down
jet impact. It is seen that these curves have the same shape but they are shifted to the left side
by air natural convection
for lower initial temperatures. and radiation
The widebut the rateinterval
confidence of heatattransfer to air
the heat flux is lessmakes
maxima than 10 kW/m2 at
it difficult
a surface temperature ◦
to claim an effect ofof 500 temperature
initial C and decreases with thisheat
over maximum temperature. It is in
flux; however, seenthein all plots ofrange
temperature Figure 8
that of
heat fluxes are ~100 kW/m 2 at a surface temperature of 150 ◦ C. Therefore, the heat flux that may
200 °C to 300 °C, a lower heat flux is associated to samples with a lower initial temperature, as
crossexpected.
through non-active surfaces is negligible.

(a) (b)

Figure 8. Cont.
Metals 2017, 7, 190 15 of 29
Metals 2017, 7, 190 15 of 29

(c)
Figure 8.
Figure 8. Boiling
Boiling curves
curves obtained
obtained from
from laboratory
laboratory data,
data, effect
effect of:
of: (a)
(a) impact
impact jet
jet velocity, (b) oil
velocity, (b) oil
temperature and (c) initial steel temperature.
temperature and (c) initial steel temperature.

Regarding validity of the adiabatic boundary condition at non-active faces of the steel disk,
4.2. Plant versus Laboratory Thermal Analyses
Figure 4b shows that sample is touched by three screws only. These screws cool down simultaneously
withThethe temperature error for plant
sample and therefore measurements
the rate of heat inputwas larger thanLateral
is minimum. for the and
corresponding laboratory
back faces cool down
data.
by airHowever, fixed and and
natural convection looseradiation
sections but
generally showed
the rate of heatlarger cooling
transfer to airrates than
is less those
than measured
10 kW/m 2 at a

for the center section. Figure 9a shows a comparison of the temperature evolution measured
surface temperature of 500 °C and decreases with this temperature. It is seen in all plots of Figure 8 in plant
versus
that heatthefluxes
one measured
are ~100 in laboratory.
kW/m Plant temperatures
2 at a surface temperature ofare150
representative of measurements
°C. Therefore, for each
the heat flux that may
leaf
crosssection,
through center, fixed and
non-active loose.
surfaces The figure also shows the times at which the leaf springs reach
is negligible.
every position in the carousel: 0 s, 36 s, 72 s, 108 s, and 144 s, leaving out the tank after 180 s. Laboratory
4.2. Plant versusinclude
measurements Laboratory Thermalfor
the results Analyses
the lowest and highest oil impact velocities. It is useful to recall
that leaf thickness is 17 mm while laboratory sample thickness is 8.5 mm. As it was mentioned before,
The temperature error for plant measurements was larger than for the corresponding laboratory
this sample represents the leaf semi-thickness since only one face is active for oil cooling. Further, it
data. However, fixed and loose sections generally showed larger cooling rates than those measured
should be recalled the position of thermocouples: in plant, they were screwed against the leaf surface,
for the center section. Figure 9a shows a comparison of the temperature evolution measured in plant
while in laboratory they were inserted through a hole in the back of the disk to a depth of 1.8 mm
versus the one measured in laboratory. Plant temperatures are representative of measurements for
from the oil cooled surface. In the figure, plant temperatures decrease very fast in comparison with
each leaf section, center, fixed and loose. The figure also shows the times at which the leaf springs
laboratory measurements, except for the center section of the leaf. The slopes of plant temperature
reach every position in the carousel: 0 s, 36 s, 72 s, 108 s, and 144 s, leaving out the tank after 180 s.
curves change abruptly just below 400 ◦ C, while the slopes of laboratory temperature curves change
Laboratory measurements include the results for the lowest and highest oil impact velocities. It is
more smoothly until reaching 300 ◦ C. Figure 9a indicates the temperatures for start and end of
useful to recall that leaf thickness is 17 mm while laboratory sample thickness is 8.5 mm. As it was
martensitic transformation, Ms and M90 , respectively. In the figure, it is seen that plant temperatures
mentioned before, this sample represents the leaf semi-thickness since only one face is active for oil
barely reach the end of martensite formation after 180 s. In contrast, laboratory temperatures decrease
cooling. Further, it should be recalled the position of thermocouples: in plant, they were screwed
well below their plant counterparts after only 100 s. This is attributed to the oil velocity: In laboratory
against the leaf surface, while in laboratory they were inserted through a hole in the back of the disk
tests, the oil impact velocity is constant during the whole test; but in plant tests the oil velocity changes
to a depth of 1.8 mm from the oil cooled surface. In the figure, plant temperatures decrease very fast
depending upon the leaf position in the tank, as it will be shown in the next section. To properly
in comparison with laboratory measurements, except for the center section of the leaf. The slopes of
compare the previous results, the corresponding boiling curves were determined using the solution of
plant temperature curves change abruptly just below 400 °C, while the slopes of laboratory
the IHCP as before. Figure 9b shows the heat flux as a function of the computed surface temperature
temperature curves change more smoothly until reaching 300 °C. Figure 9a indicates the
for all plant measurements and for the lowest and highest oil velocity laboratory cases. The major
temperatures for start and end of martensitic transformation, Ms and M90, respectively. In the figure,
difference between these results is at high temperatures. The magnitude and dispersion of plant heat
it is seen that plant temperatures barely reach the end of martensite formation after 180 s. In contrast,
fluxes are much larger than the corresponding values for laboratory tests. The center section had
laboratory temperatures decrease well below their plant counterparts after only 100 s. This is
maximum heat fluxes as low as 1200 kW/m2 while loose section heat fluxes almost reach 2800 kW/m2 .
attributed to the oil velocity: In laboratory tests, the oil impact velocity is constant during the whole
These results contrast with maximum heat fluxes between 1400 kW/m2 and 1550 kW/m2 obtained in
test; but in plant tests the oil velocity changes depending upon the leaf position in the tank, as it will
laboratory tests. On the other hand, at surface temperatures below 300 ◦ C the laboratory heat fluxes
be shown in the next section. To properly compare the previous results, the corresponding boiling
surpass the corresponding plant values. There is only some heat flux perturbation that was found
curves were determined using the solution of the IHCP as before. Figure 9b shows the heat flux as a
in thermocouples T4 loose section and T2 fixed section. Analyzing the oil velocity in both systems,
function of the computed surface temperature for all plant measurements and for the lowest and
laboratory test should have had larger heat fluxes, at least for the impact velocity of 2.0 m/s. In plant,
highest oil velocity laboratory cases. The major difference between these results is at high
the oil mean velocity at the exit of the nozzles is only 1.67 m/s as it can be estimated from data from
temperatures. The magnitude and dispersion of plant heat fluxes are much larger than the
corresponding values for laboratory tests. The center section had maximum heat fluxes as low as 1200
kW/m2 while loose section heat fluxes almost reach 2800 kW/m2. These results contrast with
maximum heat fluxes between 1400 kW/m2 and 1550 kW/m2 obtained in laboratory tests. On the other
hand, at surface temperatures below 300 °C the laboratory heat fluxes surpass the corresponding
plant values. There is only some heat flux perturbation that was found in thermocouples T4 loose
Metals 2017,
section and7,T190
2 fixed section. Analyzing the oil velocity in both systems, laboratory test should 16 of 29
have
had larger heat fluxes, at least for the impact velocity of 2.0 m/s. In plant, the oil mean velocity at the
exit
Tableof 4the
andnozzles
recallingis that
onlyinput
1.67 m/s as it can
oil velocity be m/s.
is 0.5 estimated
Once thefrom oildata from
enters Tableits4velocity
the tank and recalling that
decreases
input oil velocity is 0.5 m/s. Once the oil enters the tank its velocity decreases
very fast with distance, as it will be shown in the next section. The authors attribute these larger heatvery fast with distance,
asflux
it will be shown
values to the in the next
Taylor section. The
wavelength. Thisauthors
conceptattribute
explains these
the larger heat of
difference flux values
heat fluxesto based
the Taylor
on
wavelength. This concept explains the difference of heat fluxes based on
the surface area size of the samples. During the transitional boiling regime (at temperatures above the surface area size of the
samples.
those forDuring
maximum the transitional boiling
heat flux) the regime
hot solid (at temperatures
surface is almost blanketed above those
with for maximum
vapor. The vaporheattries
flux)
the hot solid surface is almost blanketed with vapor. The vapor tries to
to buoy up, and it does in big slugs forming unstable columns or vapor jets that form and collapse buoy up, and it does in big
slugs forming unstable
intermittently under thecolumns or vapor
liquid pressure. ThejetsTaylor
that form and collapse
instability intermittently
is the collapse process ofunder the liquid
these emerging
pressure. The Taylor instability is the collapse process of these emerging
jets and the corresponding Taylor wavelength is the distance between jets that for a one-dimensionaljets and the corresponding
Taylor
wave is wavelength
given by theisfollowing
the distance between
equation [31]: jets that for a one-dimensional wave is given by the
following equation [31]:
 1
2
√ γ
λ = 2π 3   1
  , (12)
g ρf γ− ρg 2 (12)
λ = 2π√3 ,
g ρf - ρg
where γ is the surface tension of oil, reported by the supplier as 0.033 N/m, and densities of fluid
where
and gasγ isare,
the ρsurface tension3 of oil, reported by 3the supplier as 0.033 N/m, and densities of fluid and
f = 827 kg/m and ρg = 4.3 kg/m . The Taylor wavelength was calculated as 2.2 cm.
gas are,
This ρf = 827
means that kg/m 3 and ρg = 4.3sample,
the laboratory kg/m3. The
2.54 Taylor wavelength
cm diameter, was 1.2
was only calculated as 2.2than
times larger cm. This
λ. Wemeans
can
that the laboratory sample, 2.54 cm diameter, was only 1.2 times larger than λ. We can imagine
imagine that there will only be one single vapor jet detaching from the surface. In contrast, the leaf that
there willwas
spring only 7.5becm
onewide,
singleconsidering
vapor jet detaching from
its shortest the surface.
dimension. In contrast,
This the leaffor
accommodates spring
morewas 7.5
than
cm3 times the value of λ. Therefore, several vapor jets would be forming in the plant sample, improvingof
wide, considering its shortest dimension. This accommodates for more than 3 times the value
λ.heat
Therefore,
transfer.several vapor jets would be forming in the plant sample, improving heat transfer.

(a) (b)

(c)
Figure
Figure9.9.Comparison
Comparisonofofplant
plantand
andlaboratory
laboratory results:
results: (a)
(a) measured temperatureevolution,
measured temperature evolution,(b)
(b)boiling
boiling
curves
curvesand
and(c)
(c)boiling
boilingcurves
curveswith
with Taylor wavelength correction.
Taylor wavelength correction.
Metals 2017, 7, 190 17 of 29

Our laboratory results can be corrected for this effect. Lienhard IV and Lienhard V [31] presented
a plot in their book, which shows that the maximum pool boiling heat flux, for a given system, is
independent from the surface area size of heater if this size is at least three times λ. Below this limit,
experimental data indicate a decrease in the maximum heat flux.
For our laboratory sample, we read in that plot a heat flux that is half the value the expected
one for an “infinite” surface. Considering that this full correction is applicable for the maximum heat
flux, and that the Taylor instability does not exist during one-phase cooling, we propose the following
empirical corrections for laboratory boiling curve.
For temperatures Ta ≤ T ≤ Tmax ,
T − T a ma
   
qcorr = qorig 1 + , (13)
Tmax − T a
and for temperatures Tb ≥ T > Tmax ,
( mb )
Tb − T

corr orig
q = q 1 + , (14)
Tb − T max

where Tmax is the temperature at the peak heat flux, qorig and qcorr are the original and corrected heat
fluxes, respectively; Ta and Tb are the lowest and highest temperatures where correction applies. In our
case, they were taken as 350 ◦ C and the initial sample temperature, respectively. The exponents ma
and mb were both fit equal to 1/3. The proposed empirical equations multiply only the peak heat flux
by a factor of 2 and modify the rest of the curve by a smaller factor that decreases in proportion to the
temperature difference, |T − Tmax |, reminding a level rule. Figure 9c shows the data of Figure 9b
but with corrected laboratory heat fluxes. The magnitude of heat flux is now above the corresponding
plant values, as it is expected from the low oil velocities in most of the tank. The maxima heat fluxes in
plant are shifted towards lower temperatures. This is a result of the observed large initial slopes in the
temperature-time curves and their abrupt change just below 400 ◦ C, see Figure 9a. Further research is
required to elucidate this behavior.

4.3. Fluid Flow Analysis

This section presents the results of fluid flow in the original system using both methods,
the proposed “cascade” calculations versus the full domain calculations. Numerical simulations
were then carried out using only the “cascade” calculations for the proposed cases shown in Table 7.
This table includes, in addition to the original system operating at its current oil flowrate, the system
operating at twice and at ten times this flowrate. Another modification consisted of placing baffles
along the periphery of the carousel. It was thought that they would divert the oil to flow parallel to
the carousel faces rather than crossing them. A further modification was to shorten the tank depth.
The idea behind this change was to reduce the cross-sectional area for fluid flow, which would lead to
an increase in oil velocity in the tank.

Table 7. Characteristics of the analyzed quenching systems.

Case Characteristics
Feeder with two rows of nozzles pointing upward.
Original
Oil average velocity at feeder tube, 0.5 m/s.
Original feeder.
Higher flowrate
Oil average velocity at feeder tube, 1 and 5 m/s.
Baffles installed around carousel Original feeder and oil velocity at feeder tube, 0.5 m/s.
Original feeder and oil velocity at feeder tube, 0.5 m/s.
Shallow tank
Depth of tank in carousel area, 1.73 m.
Feeder with nozzle rows at 90◦ from each other.
Nozzle rows at 90◦
Oil average velocity at feeder tube, 0.5 m/s.
Metals 2017, 7, 190 18 of 29

Metals 2017, 7, 190 18 of 29

Finally, a separate case considered a modification in the location of nozzles. The original feeder
Finally, a separate case considered a modification in the location of nozzles. The original feeder
has two parallel rows of nozzles pointing upward, but the modified one has two rows of nozzles at
has two parallel rows of nozzles pointing upward, but the modified one has two rows of nozzles at
90◦ , one pointing upward
90°, one pointing and the
upward and other pointing
the other pointinghorizontally.
horizontally.

4.3.1. Mesh Quality

4.3.1. Mesh Quality
The sensitivity of the of
The sensitivity total
the mass flowrate
total mass with
flowrate thethe
with maximum
maximumcellcell size andnumber
size and numberofof elements for
elements
for the feeder and tank in the original case is represented in Figure 10. Mesh
the feeder and tank in the original case is represented in Figure 10. Mesh refining at the feeder refining at the feeder
inlet and
inlet and nozzle outlets improved the accuracy to compute the mass flowrate. The feeder was
nozzle outlets improved the accuracy to compute the mass flowrate. The feeder was discretized with
discretized with a total of 1,354,849 elements to obtain mesh-independent results. Mesh quality is
a total of 1,354,849 elements to obtain mesh-independent results. Mesh quality is given by an average
given by an average skewness of 0.228 with a standard deviation of 0.119. Figure 10 shows that the
skewnesstankof 0.228
resultswith
were aallstandard deviationThis
mesh-independent. of 0.119.
subdomainFigurewas10divided
showsinto thata the
totaltank results were
of 6,637,275
all mesh-independent.
elements, which have an average skewness of 0.211 and standard deviation of 0.128. Finally, thewhich
This subdomain was divided into a total of 6,637,275 elements, third have
an average skewness of 0.211 and standard deviation of 0.128. Finally, the third sub-domain an
sub-domain includes the 5 virtual boxes. Each one was discretized with 3,434,308 elements having includes
average
the 5 virtual skewness
boxes. Eachof one0.230
wasand standard deviation
discretized of 0.123. elements having an average skewness of
with 3,434,308
0.230 and standard deviation of 0.123.

Figure 10. Sensitivity

Figure analysis
10. Sensitivity in terms
analysis of computed
in terms of computedpercent oftotal
percent of totaloiloil flowrate
flowrate through
through feederfeeder
and and
tank. Thetank.
number
The number of used elements is indicated in every point. Refined mesh at feeder inlet and nozzlenozzle
of used elements is indicated in every point. Refined mesh at feeder inlet and
outlets
outlets improved improved calculations
calculations accuracy.
accuracy.

In contrast, the full system was discretized into ~48.6 million elements to obtain a similar element
In contrast, the fulland
size distribution system wasskewness
average discretized into ~48.6
compared with million elements to
all the subdomains obtain
used a similar
for the proposedelement
size distribution
“cascade” and average
computing skewness
sequence. Table compared with all of
8 shows a comparison thethesubdomains used for
number of elements and the proposed
the total
“cascade” computing
computing timesequence.
between these methods.
Table 8 shows The a“cascade”
comparison calculations
of the require
number a maximum of 6.6
of elements and the
million elements per run, whereas the full system needs a total of 48.6 million
total computing time between these methods. The “cascade” calculations require a maximum of elements. This makes
a tremendous difference in terms of computer memory, file management and storage requirements.
6.6 million elements per run, whereas the full system needs a total of 48.6 million elements. This makes
The total running time takes 21.7 h in “cascade” method, whereas this run lasts 168 h in the full system
a tremendous difference in terms of computer memory, file management and storage requirements.
calculations to obtain similar convergence and results.
The total running time takes 21.7 h in “cascade” method, whereas this run lasts 168 h in the full system
calculations to obtain
Table similar
8. Number convergence
of elements and results.
and computing time for fluid flow simulation in a quench tank using
“cascade” calculations versus full system calculations. Computer runs were carried out on an 8-core
microprocessor Intel Xeonand
Table 8. Number of elements ESG45, 2.4 GHz and
computing time32 for
GB of RAM.
fluid flow simulation in a quench tank using
“cascade” calculationsMillions
versus of
full system Running
calculations.
Time— Computer Millions
runs wereof carried out onTime—
Running an 8-core
Subdomain
Elements
microprocessor Intel Xeon ESG45, 2.4 GHz “Cascade”
and 32 GB(h)of RAM. Elements Full System (h)
Feeder 1.4 1.1 48.6 168

Millions of Running Millions of Running Time—Full

Subdomain
Elements Time—“Cascade” (h) Elements System (h)
Feeder 1.4 1.1
Tank 6.6 12.2 48.6 168
Box 1 3.4 1.9
Metals 2017, 7, 190 19 of 29

Table 8. Cont.

Millions of Running Millions of Running Time—Full

Subdomain
Metals 2017, 7, 190 Elements 19 of 29
Time—“Cascade” (h) Elements System (h)
BoxTank
2 3.4 6.6 1.4 12.2
BoxBox
3 1 3.4 3.4 1.6 1.9
BoxBox
4 2 3.4 3.4 1.7 1.4
BoxBox
5 3 3.4 3.4 1.8 1.6
Box 4 3.4 1.7
Box 5 3.4 1.8
4.3.2. Choice of Virtual Box Size
4.3.2.
The Choicebox
virtual of Virtual Box Size
size was selected to include both leaf springs, with dimensions to match or
exceed the shaping tools surface
The virtual box size was boundaries. Three box
selected to include bothsizes were considered
leaf springs, to carry
with dimensions to out a sensitivity
match or
analysis, smallthe
exceed (1458 × 434tools
shaping × 221), selected
surface (1458 ×Three
boundaries. 570 × 221)
box andwere
sizes (1818 × 570
largeconsidered × 286),
to carry outalla in mm.
sensitivity
Figure 11a showsanalysis, small (1458
the computed maps× 434
of ×shear
221), selected
stress on(1458
the ×surface
570 × 221) and large
of both leaf (1818 × 570
springs × 286), using
obtained
all in mm. Figure 11a shows the computed maps of shear stress on the surface of both leaf springs
three different virtual box sizes. It can be noticed that the small and selected boxes led to mutually
obtained using three different virtual box sizes. It can be noticed that the small and selected boxes led
similar results, while the large box results did not fully agree with them.
to mutually similar results, while the large box results did not fully agree with them.

L1 Front
Sup
L2 Back

Small Selected Large

(a)

(b) (c)
Figure 11. (a) Comparison of computed wall shear stress maps on the leaf spring surfaces for the first
11. (a) Comparison
Figure carousel of computed wall shear stress maps on the leaf spring surfaces for the
position, with small, selected, and large virtual boxes. (b) Average wall shear stress per face
first carousel position,
for three box sizeswith small,
and (c) selected,
average and
pressure perlarge virtual
face for boxes.
three box (b)
sizes. L1Average wall
= Leaf 1, L2 shear
= Leaf stress per
2, box
face forsize
three
1 = box sizes
small, and (c)3average
2 = selected, = large. pressure per face for three box sizes. L1 = Leaf 1, L2 = Leaf 2,
box size 1 = small, 2 = selected, 3 = large.
Metals2017,
Metals 2017,7,7,190
190 20
20of
of29
29

Figure 11b,c are plots of the average wall shear stress and pressure, respectively. The averages
were Figure 11b,c each
taken over are plots of the
of four leafaverage
faces. wall
Theseshear
plotsstress and pressure,
confirm respectively.
the agreement betweenThe theaverages
results
obtained with small and selected boxes. The faces of the selected box size are close enough toobtained
were taken over each of four leaf faces. These plots confirm the agreement between the results the leaf
with small
surface and selected
to properly boxes.
transmit theThe faces of the
oil velocity selected box
information sizesurface,
to this are closeand
enough to the
are also far leaf surface
enough fromto
properly transmit the oil velocity information to this surface, and
the leaf body to include a better representation of the tools that hold it. are also far enough from the leaf
body to include a better representation of the tools that hold it.
4.3.3. Fluid Flow Maps for Original Case
4.3.3. Fluid Flow Maps for Original Case
Figure 12a–d show a comparison between the results obtained with the “cascade” and full
Figure 12a–d show a comparison between the results obtained with the “cascade” and full system
system methods. Figure 12a represents the shear stress distribution computed using the “cascade”
methods. Figure 12a represents the shear stress distribution computed using the “cascade” method at
method at the surface of the leaf springs at carousel position number 1, and Figure 12b shows the
the surface of the leaf springs at carousel position number 1, and Figure 12b shows the corresponding
corresponding results but using the full domain calculations. In both cases, the wall shear is larger at
results but using the full domain calculations. In both cases, the wall shear is larger at the spring eyes
the spring eyes because the oil flows freely there. In addition, a shadow effect in the wall shear
because the oil flows freely there. In addition, a shadow effect in the wall shear distribution is present.
distribution is present. This is a result of directional flow passing through the leaf springs. Figure
This is a result of directional flow passing through the leaf springs. Figure 12c,d show the wall shear
12c,d show the wall shear stress element frequency obtained with the full domain and “cascade”
stress element frequency obtained with the full domain and “cascade” methods for first and second
methods for first and second carousel positions. In these plots, the frequency represents the number
carousel positions. In these plots, the frequency represents the number of volume elements which wall
of volume elements which wall shear is within one class number. Each class number is equivalent to
shear is within one class number. Each class number is equivalent to 0.01 Pa. These figures show that
0.01 Pa. These figures show that there is a fair agreement between “cascade” and full solutions. The
there is a fair agreement between “cascade” and full solutions. The “cascade” solution was sensitive
“cascade” solution was sensitive enough to recreate the response of the shear stress distribution to
enough to recreate the response of the shear stress distribution to both, leaf and carousel positions.
both, leaf and carousel positions. The analysis presented in this paper was obtained using the
The analysis presented in this paper was obtained using the “cascade” method.
“cascade” method.

(a)

(b)

Figure 12. Cont.

Metals 2017, 7, 190 21 of 29
Metals 2017, 7, 190 21 of 29

(c) (d)
Figure 12.
Figure 12. Computed
Computed wallwall shear
shear stress
stress maps
maps on
on the
the leaf
leaf spring
spring surfaces
surfaces for
for the
the first
first carousel position
carousel position
in the original quench system using (a) “cascade” and (b) full system methods. Comparison of
in the original quench system using (a) “cascade” and (b) full system methods. Comparison wall
of wall
shear stress cell frequency for (c) first carousel position, box 1 and (d) second carousel position,
shear stress cell frequency for (c) first carousel position, box 1 and (d) second carousel position, box box
2.
2. One class number is equivalent to
One class number is equivalent to 0.01 Pa.0.01 Pa.

4.3.4. Effect of Oil Flowrate

4.3.4. Effect of Oil Flowrate
Figure 13a–c show the computed velocity norm maps at the frontal face of the 5 virtual boxes,
Figure 13a–c show the computed velocity norm maps at the frontal face of the 5 virtual boxes,
according with Figure 2a starting from position 1 at the top, when feeding oil at three different
according with Figure 2a starting from position 1 at the top, when feeding oil at three different velocities
velocities 0.5 m/s, 1 m/s and 5 m/s, respectively. The frontal face of a box is the one that sees radially
0.5 m/s, 1 m/s and 5 m/s, respectively. The frontal face of a box is the one that sees radially out,
out, and it receives the oil that comes from the feeder. The projecting shaper table is shown by the
and it receives the oil that comes from the feeder. The projecting shaper table is shown by the white
white rectangular spots. Figure 13a corresponds to the original case and shows that the first box is
rectangular spots. Figure 13a corresponds to the original case and shows that the first box is essentially
essentially the only one that receives the oil at a relatively higher velocity, ~0.04 m/s. Boxes 2 to 5
the only one that receives the oil at a relatively higher velocity, ~0.04 m/s. Boxes 2 to 5 receive oil at
receive oil at very low velocities. This is consistent with the measured plant temperature evolution
very low velocities. This is consistent with the measured plant temperature evolution presented in
presented in Figure 9a. In these results, the measured rate of cooling in the first carousel position was
Figure 9a. In these results, the measured rate of cooling in the first carousel position was the highest.
the highest. Figure 13b shows the velocity maps when the oil is fed at 1 m/s. Once again, the first box
Figure 13b shows the velocity maps when the oil is fed at 1 m/s. Once again, the first box receives
receives the oil at the highest velocity, leaving the other boxes with the lowest flow velocity. Figure
the oil at the highest velocity, leaving the other boxes with the lowest flow velocity. Figure 13c is
13c is an extreme case in which oil is input at 5 m/s. The oil velocity at the first box increases at values
an extreme case in which oil is input at 5 m/s. The oil velocity at the first box increases at values of
of approximately 0.40 m/s. Velocity is also increased in all other boxes. This solution might work for
approximately 0.40 m/s. Velocity is also increased in all other boxes. This solution might work for
improving productivity. However, the high and non-uniform oil velocity at the first box may lead to
improving productivity. However, the high and non-uniform oil velocity at the first box may lead to
localized high cooling rates with undesirable quality results.
localized high cooling rates with undesirable quality results.
Metals 2017, 7, 190 22 of 29
Metals
Metals 2017,
2017, 7, 190
7, 190 2229
22 of of 29

(a)(a) (b)(b) (c)(c)

Figure
Figure 13.13. Computed
Computed velocity
velocity maps
maps at frontal
at frontal face
face areas
areas of virtual
of virtual boxes.
boxes. Original
Original system
system operating
operating
Figure 13. Computed velocity maps at frontal face areas of virtual boxes. Original system operating at
at inlet
at inlet oil oil velocity
velocity of (a)
of (a) 0.50.5 m/s,
m/s, (b)(b) 1 m/s
1 m/s andand
(c) (c) 5 m/s.
5 m/s.
inlet oil velocity of (a) 0.5 m/s, (b) 1 m/s and (c) 5 m/s.
4.3.5.
4.3.5. Effect
Effect of of Baffles
Baffles at Carousel
at Carousel Periphery
Periphery
4.3.5. Effect of Baffles at Carousel Periphery
Figure
Figure 14a,b
14a,b show
show thethe computed
computed velocity
velocity norm
norm maps
maps ononthethe frontal
frontal faces
faces of of
thethe virtual
virtual boxes
boxes forfor
theFigure
the 14a,b
original
original case show
case
and andthefor
for computed
thethe velocity
modified
modified norm
carousel
carousel maps
with
with onThe
baffles.
baffles. the
The frontal
original
original faces ofwas
feeder
feeder the
was virtual
usedused boxes
in in both for
both
thecases.
original
cases.
The case
The andshow
figures
figures for the
show modified
that
that there
there carousel
is
is no no withdifference
baffles.
significant
significant The
difference original
between
between feeder
these
these was
cases.
cases. used
This
This isin both cases.
attributed
is attributed
Thetofigures
to
thethe show
depth
depth of that
of
thethethere
tank.
tank. is Although
no significant
Although thethe difference
baffles
baffles were
were between
placed
placed to tothese
block
block cases.
thethe This
oiloil
flowflowis attributed
through
through thethetointerior
the depth
interior
of the
of oftank.
thethe Although
carousel,
carousel, thethethe
still baffles
still
relativelywere
relatively
largeplaced
large to block the
cross-sectional
cross-sectional oil
of flow
area
area of
thethe through
tanktank
ledledtothe
to
lowinterior
low oiloil of theascarousel,
velocity,
velocity, as
in in
the the original
original case.
case.
the still relatively large cross-sectional area of the tank led to low oil velocity, as in the original case.

(a)(a) (b)(b)

Figure 14. Computed velocity maps at frontal face areas of virtual boxes. (a) Original case and (b)
Carousel with baffles installed through its perimeter to block the oil flow. Inlet oil velocity was 0.5 m/s.
Metals 2017, 7, 190 23 of 29

Figure 14. Computed velocity maps at frontal face areas of virtual boxes. (a) Original case and (b)
Carousel with baffles installed through its perimeter to block the oil flow. Inlet oil velocity was 0.5
Metals
m/s. 2017, 7, 190 23 of 29

4.3.6. Effect of Tank Depth

4.3.6. Effect of Tank Depth
A decrease in the tank depth was aimed to reduce the tank cross-sectional area for the oil to flow,
A decrease in the tank depth was aimed to reduce the tank cross-sectional area for the oil to
increasing its velocity. Figure 15a,b show the computed norm velocity maps for the original case and
flow, increasing its velocity. Figure 15a,b show the computed norm velocity maps for the original case
for the tank with a shortened depth. A comparison between the velocities shows that they are
and for the tank with a shortened depth. A comparison between the velocities shows that they are
essentially equivalent systems. This means that the sole reduction of the tank depth is insufficient to
essentially equivalent systems. This means that the sole reduction of the tank depth is insufficient to
modify the oil velocity pattern in the virtual box faces. The oil flows mostly through the interior of
modify the oil velocity pattern in the virtual box faces. The oil flows mostly through the interior of
the carousel.
the carousel.

(a) (b)
Figure 15. Computed
Figure velocity
15. Computed maps
velocity mapsat frontal face
at frontal areas
face ofofvirtual
areas virtualboxes.
boxes.(a)
(a)Original
Original case and (b)
case and (b)Tank
Tankwith
withshortened
shorteneddepth.
depth.Inlet
Inletoil
oilvelocity
velocitywaswas0.5
0.5m/s.
m/s.

4.3.7. Effect of Nozzle Direction

4.3.7. Effect of Nozzle Direction
Figure 16a,b show computer digital representations of the original feeder and the modified one,
Figure 16a,b show computer digital representations of the original feeder and the modified one,
respectively. In both figures, the computed velocity vector distribution is seen at the nozzle outlets.
respectively. In both figures, the computed velocity vector distribution is seen at the nozzle outlets.
Figure 16c,d show the computed velocity norm maps at the frontal face of the virtual boxes when
Figure 16c,d show the computed velocity norm maps at the frontal face of the virtual boxes when
using a feeder as in Figure 16a,b, respectively. Notice that in Figure 16d, the second box receives oil
using a feeder as in Figure 16a,b, respectively. Notice that in Figure 16d, the second box receives oil at
at a relatively higher velocity compared with the second box in Figure 16c. This is the result of
a relatively higher velocity compared with the second box in Figure 16c. This is the result of relocating
relocating the second row of nozzles. In contrast, the first box receives oil at a lower velocity
the second row of nozzles. In contrast, the first box receives oil at a lower velocity compared with
compared with the first box in Figure 16c. This solution may be suitable for our purpose because it
the first box in Figure 16c. This solution may be suitable for our purpose because it improves the
improves the oil velocity in the second box, in which the measured cooling rate decreased, as shown
oil velocity in the second box, in which the measured cooling rate decreased, as shown in Figure 9a.
in Figure 9a. Figure 17a,b show the computed velocity maps for the cases in Figure 16c,d, respectively,
Figure 17a,b show the computed velocity maps for the cases in Figure 16c,d, respectively, but in a 3D
but in a 3D view, to show velocity maps in additional faces. Again, it is seen that oil velocity in box 1
view, to show velocity maps in additional faces. Again, it is seen that oil velocity in box 1 is larger for
is larger for the original system but in box 2 the situation is opposite. Figure 17c,d show the computed
the original system but in box 2 the situation is opposite. Figure 17c,d show the computed average
average wall shear on each of the 4 faces of both leaf springs when they are in box 1, Figure 17c, and
wall shear on each of the 4 faces of both leaf springs when they are in box 1, Figure 17c, and in box 2,
in box 2, Figure 17d. It is clear from these plots that wall shear stress is larger for the original case in
Figure 17d. It is clear from these plots that wall shear stress is larger for the original case in box 1 but
the 90◦ nozzle rows case dominates in box 2. The inferior (face 2) and superior (face 3) leaf surfaces
represent the main heat removal areas. The change in nozzle configuration represents, for box 1,
Metals 2017, 7, 190 24 of 29
Metals 2017, 7, 190 24 of 29
box 1 but the 90° nozzle rows case dominates in box 2. The inferior (face 2) and superior (face 3) leaf
surfaces represent the main heat removal areas. The change in nozzle configuration represents, for
a decrease in shear stress in these areas from 0.4–0.9 Pa to 0.4–0.5 Pa. However, it represents for box 2
box 1, a decrease in shear stress in these areas from 0.4–0.9 Pa to 0.4–0.5 Pa. However, it represents
an
forincrease
box 2 aninincrease
shear stress fromstress
in shear 0.08–0.11
fromPa0.08–0.11
to 0.10–0.43
Pa toPa.0.10–0.43
To find outPa.ifTothis increase
find out if is high
this enough
increase is
to compensate
high enough to thecompensate
decrease in thethe shear stress
decrease ininthe
boxshear
1, it isstress
necessary
in boxto 1,
have a relationship
it is necessary tobetween
have a
heat flux andbetween
relationship wall shear stress.
heat flux and wall shear stress.
Before
Before stepping forwardtoto
stepping forward heat transfer
heat analysis,
transfer it isitinteresting
analysis, to notice
is interesting that Figures
to notice 14–17 14–17
that Figures show
that
showoilthat
velocity drops sharply
oil velocity after boxafter
drops sharply 1. This
boxmeans
1. Thisthat most that
means of themost
momentum and turbulence
of the momentum and
kinetic energy of the feeding jets are dissipated in a relatively short distance
turbulence kinetic energy of the feeding jets are dissipated in a relatively short distance from the nozzles outlet.
from the
Only box 1 receives oil at a moderate velocity. These results offer an explanation
nozzles outlet. Only box 1 receives oil at a moderate velocity. These results offer an explanation onon the resulting high
cooling rates that
the resulting highwere measured
cooling in leaf
rates that sections
were during
measured in the
leaffirst seconds,
sections see Figure
during 9a.seconds,
the first In contrast,
see
the measured cooling rates corresponding to positions of boxes 2 to 5 were very small.
Figure 9a. In contrast, the measured cooling rates corresponding to positions of boxes 2 to 5 were This is consistent
with
very asmall.
predicted very
This is low oil velocity
consistent flowing through
with a predicted very lowthese boxes. flowing through these boxes.
oil velocity

(a) (b)

(c) (d)
Figure 16.
Figure 16. Computed
Computed oiloil velocity
velocity distribution
distribution at
at nozzle
nozzle exit
exit for:
for: (a)
(a) Original
Original case,
case, (b)
(b) nozzle
nozzle rows
rows at
at
90°
90 between them.
◦ between them. Computed velocity maps
Computed velocity maps at
at frontal
frontal face
face areas
areas of
of virtual
virtual boxes
boxes for
for (c)
(c) Original case,
Original case,
(d) nozzle rows at 90°.
◦
(d) nozzle rows at 90 .
Metals 2017, 7, 190 25 of 29
Metals 2017, 7, 190 25 of 29

(a) (b)

(c) (d)
Figure
Figure 17.
17.Computed
Computedvelocity
velocitymaps
mapsatat face areas
face areasbut in in
but perspective
perspectiveview for:for:
view (a) Original casecase
(a) Original andand
(b)
nozzle rowsrows
(b) nozzle at 90°.
at Inlet
90 . oil
◦ velocity
Inlet was 0.5
oil velocity m/s.
was 0.5Comparison of average
m/s. Comparison of wall shear
average stress
wall in (c)
shear box in
stress 1
and (d) box
(c) box 2 computed
1 and in each leafinsurface
(d) box 2 computed face,surface
each leaf 1 = front, 2 =1inferior,
face, = front, 32==superior
inferior,and
3 = 4superior
= back. and
4 = back.
4.4. Isothermal Fluid Flow and Heat Flux
4.4. Isothermal Fluid Flow and Heat
The Reynolds-Colburn Flux Equation (4), is our starting point to obtain a relationship
analogy,
between
The the heat flux and the
Reynolds-Colburn wall shear
analogy, stress,
Equation and
(4), it can
is our be rewritten
starting point toas:
obtain a relationship between
qw be rewritten
the heat flux and the wall shear stress, and it can τ as:
= w, (15)
Cp Tf - Tw vf
qw τw
This equation is the basis to postulate theCpfollowing = , (15)
( Tf − T w )empirical
vf equation:
q = q1w τnw , (16)
This equation is the basis to postulate thewfollowing empirical equation:
In this power law equation, q1w is numerically equal to the heat flux when the wall shear stress
n
is equal to 1, and n is a real number; both were q1w τw
qw = determined, (16)
from the laboratory results that were
shown in Figure 8a. In this figure, the heat flux depends on the oil impact velocity, and the
In this power law equation, q1w is numerically equal to the heat flux when the wall shear stress is
corresponding curves are fairly parallel each other. These velocities led to wall shear stresses that
equal to 1, and n is a real number; both were determined from the laboratory results that were shown
were estimated from previous studies of jet impacting a horizontal planar surface. The early work of
in Figure 8a. In this figure, the heat flux depends on the oil impact velocity, and the corresponding
Bradshaw and Love [32] reports measurements of velocity magnitude and direction, static pressure,
curves are fairly parallel each other. These velocities led to wall shear stresses that were estimated
and skin friction in a circular turbulent air jet impinging normally on a flat, isothermal, surface. They
from previous studies of jet impacting a horizontal planar surface. The early work of Bradshaw and
found that the maximum value of skin friction shear stress occurs at a radius approximately equal to
Love [32] reports measurements of velocity magnitude and direction, static pressure, and skin friction
that of the jet and is about 0.006 times the jet maximum dynamic pressure. Nowadays, these results
in a circular turbulent air jet impinging normally on a flat, isothermal, surface. They found that the
have been confirmed and accepted even for liquids impinging over flat surfaces. The authors
reported that the skin friction changes from zero at the axis of the jet (r = 0) to its maximum value at
the radius of the jet (r ≈ R) in a quasi-linear form. Therefore, to determine the wall shear stress for our
Metals 2017, 7, 190 26 of 29

maximum value of skin friction shear stress occurs at a radius approximately equal to that of the jet and
is about 0.006 times the jet maximum dynamic pressure. Nowadays, these results have been confirmed
and accepted even for liquids impinging over flat surfaces. The authors reported that the skin friction
changes from zero at the axis of the jet (r = 0) to its maximum value at the radius of the jet (r ≈ R)
in a quasi-linear form. Therefore, to determine the wall shear stress for our laboratory experiments,
we considered a value half the maximum shear stress, which corresponds to an arithmetic average
shear stress. The gravity effect was considered in the analysis by estimating de impact jet velocity with
Equation (2).
Table 9 shows the estimated average wall shear stress for each experiment. The table also shows
the estimated ratio of the heat flux divided by the corresponding heat flux at shear stress equal to
one. This ratio was determined from Figure 8a by considering the heat fluxes at 250 ◦ C. Notice that at
this temperature, one phase convection is the heat transfer mechanism. However, we are applying
Equation (16) to the whole boiling curve. Finally, the table shows the computed value for n using
Equation (16). We present this coefficient as,

n = 0.06, for τ ≤ 0.2 Pa; and n = 0.46 (average value), for τ ≥ 1 Pa, (17)

where n should be interpolated for a shear stress value between 0.2 Pa and 1 Pa. Furthermore, if we
consider that the effect of the initial steel temperature, TI , is to displace the heat flux curve as it was
shown in Figure 8c, we use the following expression for q1w :

q1w = f ( TI − Tw ), (18)

This equation means that q1w is a function of the difference between initial and wall temperatures
rather than a function of only Tw . Of course, the same will be true for qw . Equation (16) together with
Equations (17) and (18) represent a relationship between the computed isothermal wall shear stress and
the expected heat flux. Application of this equation to the production tank requires that the laboratory
heat flux curve should also be corrected for the Taylor wavelength according with Equation (13).

Table 9. Estimated average wall shear stress for the laboratory experiments using Bradshaw and Love
results [32]. The ratio qw /q1w was obtained from Figure 8a.

Impact Velocity, VI (m/s) Average Wall Shear Stress, τmax /2 (Pa) qw /q1w n
0.4 0.2 0.903 0.06
0.9 1.0 1 -
1.3 2.1 1.452 0.503
2.0 4.96 1.935 0.412

Based on the computed shear stress, Equation (16) predicts that heat flux decreases in leaves of
box 1 around 15% when changing the feeder from the original case to the proposed 90◦ rows case.
However, the improvement in heat flux obtained in the leaves in box 2 is around 10%. This means
that the expected overall cooling rates should be very similar for both cases. Another different design
and/or operating solution should be found to improve the rate of cooling of the leaf springs.

5. Summary and Conclusions

An efficient fluid-dynamic analysis has been developed and applied to the redesign of
an industrial oil-quenching tank for automotive leaf springs. The analysis is based on a proposed
empirical power-law equation to relate the isothermal wall shear stress, computed numerically on the
surface of the leaf springs, with the heat flux removed during their quenching. Laboratory experiments,
carried out on a modified Jominy End-Quenched rig, allowed us to determine the parameters of
an empirical relationship between the measured heat flux and the estimated isothermal average shear
Metals 2017, 7, 190 27 of 29

stress on the surface impacted by the oil jet. This empirical equation is based on the Reynolds-Colburn
analogy and includes correction for the effect of the initial temperature of the steel. The result should be
also corrected for the Taylor wavelength, that accounts for the difference in surface area size between
the laboratory and plant specimens.
The numerical model predicts efficiently the isothermal velocity field in the tank, by solving
the momentum and continuity equations under steady-state conditions and turbulent flow in 3D.
Computer efficiency is based on “cascade” calculations in three subdomains: the oil feeder, the tank,
and 5 virtual boxes that contain the leaf springs held by the appropriate tools and immersed in
the tank. Therefore, the oil velocity distribution at the nozzle outlet of each subdomain was used
as the inlet boundary condition for the next subdomain. Calculations were carried out in each
subdomain one at a time and compared with the corresponding results when using the full domain
in a single run. In both methods, meshes with similar average element size and skewness were used.
In addition, the corresponding fluid flow results fairly agree. It was shown that “cascade” calculations
save a considerable amount of computer memory and processing time compared with full-domain
calculations, without a considerable loss of accuracy.
The computed norm velocity maps on the faces of the virtual boxes are convenient indicators to
explore a potential benefit or drawback when implementing new design options. Calculating the flow
field inside the boxes reveals details of the shear stress distribution on the leaf spring surfaces.
In the present work, the measured rate of cooling at several locations on the surface of a leaf
spring showed that there are significate differences between these rates, particularly during the period
at position 1 of the carousel. This is attributed to a non-uniform oil velocity distribution near the
leaf spring. The rate of cooling decreases from positions 2 to 5. Therefore, oil velocity should be
improved in these positions. The fluid-dynamic analysis of the original quenching system and most of
its modifications showed little improvement of the oil velocity in the virtual boxes that contain the
leaf springs. A plain increment in the oil flowrate increases the fluid velocity only in virtual box 1 but
remains with little change in the other boxes. A reduction of the cross-sectional area of the flow path
by either introducing baffles to block the flow through the center of the carousel or reducing the tank
depth was insufficient to improve the liquid velocity at the other boxes. The better case was the feeder
with nozzle rows 90◦ from each other. In this case, the oil velocity in the second carousel position
increased with respect to its value in the original case. However, a detailed heat transfer analysis
using the proposed empirical equation showed that no major benefit could be obtained from this
solution. More possible solutions can be conveniently simulated and evaluated using the proposed
“cascade” calculations.

Acknowledgments: The authors are thankful for funding from Consejo Nacional de Ciencia y Tecnología
(CONACyT) for PhD scholarship of MJBR and Grant # 183161; Subsecretaría de Economía, Programa de Apoyo
para la Mejora Tecnológica de la Industria de Alta Tecnología (PROIAT), and Rassini Suspensiones Piedras Negras,
Coahuila. We also thank Grupo SSC for kind and prompt technical support about ANSYS FluentTM .
Author Contributions: Francisco A. Acosta-González and Marco A. González-Melo conceived and designed the
laboratory experiments while Francisco A. García-Pastor and Eddy Alfaro-López conceived and designed the plant
experiments. Marco A. González-Melo and Manuel de J. Barrena-Rodríguez performed the laboratory experiments.
Francisco A. Acosta-González and Marco A. González-Melo analyzed the data, Manuel de J. Barrena-Rodríguez
conceived, developed and run the mathematical model for fluid flow and analyzed its relationship with heat flow
measurements. Francisco A. Acosta-González wrote the paper.
Conflicts of Interest: The authors declare that there is no conflict of interest regarding the publication of this paper.

References
1. Canale, L.D.C.F.; Totten, G.E. Quenching technology: A selected overview of the current state-of-the-art.
Mater. Res. 2005, 8, 461–467. [CrossRef]
2. Totten, G.E. Quench Process Sensors. In ASM Handbook; Dossett, J., Totten, G.E., Eds.; ASM International:
Materials Park, OH, USA, 2013; pp. 192–197.
3. Gür, C.H.; Şimşir, C. Simulation of quenching: A review. Mater. Perform. Charact. 2012, 1, 1–37. [CrossRef]
Metals 2017, 7, 190 28 of 29

4. Totten, G.E.; Bates, C.E.; Clinton, N.A. Handbook of Quenchants and Quenching Technology; ASM International:
Materials Park, OH, USA, 1993; pp. 339–411.
5. Garwood, D.R.; Lucas, J.D.; Wallis, R.A.; Ward, J. Modeling of the flow distribution in an oil quench tank.
J. Mater. Eng. Perform. 1992, 1, 781–787. [CrossRef]
6. Halva, J.; Volný, J. Modeling the flow in a quench bath. Hut. Listy 1993, 48, 30–34.
7. Bogh, N. Quench Tank Agitation Design Using Flow Modeling. In Proceedings of the International Heat
Treating Conference: Equipment and Processes, Schaumburg, IL, USA, 18–20 April 1994; Totten, G.E.,
Wallis, R.A., Eds.; ASM International: Materials Park, OH, USA, 1994.; pp. 51–54.
8. Kernazhitskiy, S.L. 2003. Numerical Modeling of a Flow in a Quench Tank. Master’s Thesis, Portland State
University, Portland, OR, USA, 2003.
9. Kumar, A.; Metwally, H.; Paingankar, S.; MacKenzie, D.S. Evaluation of Flow Uniformity around Automotive
Pinion Gears during Quenching. In Proceedings of the 5th International Conference on Quenching and
Control Distortion at 2007 European Conference on Heat Treatment, Berlin, Germany, 25–27 April 2007;
International Federation for Heat Treatment and Surface Engineering: Berlin, Germany, 2007.; pp. 1–8.
10. Romão-Bineli, A.R.; Rocha-Barbosa, M.I.; Luiz-Jardini, A.; Maciel-Filho, R. Simulation to Analyze Two
Models of Agitation System in Quench Process. In Proceedings of the 20th European Symposium on
Computer Aided Process Engineering, Naples, Italy, 6–9 June 2010; Pierucci, S., Ferraris, G.B., Eds.;
Elsevier B.V.: Naples, Italy, 2010; pp. 1–6.
11. Krause, F.; Schüttenberg, S.; Fritsching, U. Modelling and simulation of flow boiling heat transfer. Int. J.
Numer. Methods Heat Fluid Flow 2010, 20, 312–331. [CrossRef]
12. Srinivasan, V.; Moon, K.-M.; Greif, D.; Wang, D.; Kim, M.-H. Numerical simulation of immersion quench
cooling process using an Eulerian multi-fluid approach. Appl. Therm. Eng. 2010, 30, 499–509. [CrossRef]
13. Gao, W.-M.; Fabijanic, D.; Hilditch, T.; Kong, L.-X. Integrated fluid-thermal-structural numerical analysis for
the quenching of metallic components. J. Shanghai Jiaotong Univ. Sci. 2011, 16, 137–140. [CrossRef]
14. Ricci, M.G.; Santin, G.; Cascariglia, F.; Scimiterna, E.; Parrabbi, A.; Carpinelli, A. Numerical analysis applied
to thermal and fluid-dynamic quenching simulation of large high quality forgings. La Metall. Italiana 2011, 9,
3–10.
15. El Kosseifi, N. 2012. Numerical Simulation of Boiling for Industrial Quenching Processes. Ph.D. Thesis,
Paris National School of Mines, Paris, France, 2012.
16. Ko, D.-H.; Ko, D.-C.; Lim, H.-J.; Kim, B.-M. Application of QFA coupled with CFD analysis to predict the
hardness of T6 heat treated A16061 cylinder. J. Mech. Sci. Technol. 2013, 27, 2839–2844. [CrossRef]
17. Kopun, R.; Škerget, L.; Hriberšek, M.; Zhang, D.; Edelbauer, W. Numerical investigations of quenching
cooling processes for different cast aluminum parts. J. Mech. Eng. 2014, 60, 571–580. [CrossRef]
18. Passarella, D.N.; L-Cancelos, R.; Vieitez, I.; Varas, F.; Martín, E.B. Thermo-fluid-dynamics quenching model:
Effect on material properties. Blutcher Mech. Eng. 2012, 1, 3625–3644.
19. Yang, X.-W.; Zhu, J.-C.; He, D.; Lai, Z.-H.; Nong, Z.-S.; Liu, Y. Optimum design of flow distribution in
quenching tank for heat treatment of A357 aluminum alloy large complicated thin-wall workpieces by CFD
simulation and ANN approach. Trans. Nonferr. Met. Soc. China 2013, 23, 1442–1451. [CrossRef]
20. Banka, A.L.; Ferguson, B.L.; MacKenzie, D.S. Evaluation of flow fields and orientation effects around ring
geometries during quenching. J. Mater. Eng. Perform. 2013, 22, 1816–1825. [CrossRef]
21. López-García, R.D.; García-Pastor, F.A.; Castro-Román, M.J.; Alfaro-López, E.; Acosta-González, F.A. Effect
of immersion routes on the quenching distortion of a long steel component using a finite element model.
Trans. Indian Inst. Met. 2016, 69, 1645–1656. [CrossRef]
22. Standard Test Methods for Determining Hardenability of Steel; ASTM International: West Conshohocken, PA,
USA, 2007; pp. 18–26.
23. The Reynolds Analogy. Available online: http://web.mit.edu/16.unified/www/FALL/thermodynamics/
notes/node122.html (accessed on 20 March 2017).
24. Ansys Fluent Theory Guide, Release 15.0; ANSYS, Inc.: Canonsburg, PA, USA, 2013; pp. 39–55.
25. Shih, T.-H.; Liou, W.W.; Shabbir, A.; Yang, Z.; Zhu, J. A new k-ε eddy-viscosity model for high reynolds
number turbulent flows. Comput. Fluids 1995, 24, 227–238. [CrossRef]
26. Kim, S.-E.; Choudhury, D.; Patel, B. Computations of Complex Turbulent Flows Using the Commercial Code
ANSYS Fluent. In Proceedings of the ICASE/LaRC/AFOSR Symposium on Modeling Complex Turbulent
Flows, Hampton, VA, USA, 11–13 August 1997.
Metals 2017, 7, 190 29 of 29

27. Ansys Fluent User’s Guide. Release 12.0; ANSYS, Inc.: Canonsburg, PA, USA, 2009; pp. 7-128–7-130.
28. Patankar, S.V. Numerical Heat Transfer and Fluid Flow; CRC Press: Boca Raton, FL, USA, 1980.
29. Beck, J.V. User’s Manual for CONTA: Program for Calculating Surface Heat Fluxes from Transient Temperatures
Inside Solids; Michigan State University: East Lansing, MI, USA, 1983.
30. Liscic, B.; Tensi, H.M.; Canale, L.C.F.; Totten, G.E. Quenching Theory and Technology, 2nd ed.; CRC Press: Boca
Raton, FL, USA, 2011.
31. Lienhard IV, J.H.; Lienhard V, J.H. A Heat Transfer Textbook, 3rd ed.; Phlogiston Press: Cambridge, MA, USA,
2001; pp. 436–445.
32. Bradshaw, P.; Love, E.M. The Normal Impingement of a Circular Air Jet on a Flat Surface; Her Majesty’s Stationary
Office: London, UK, 1959; pp. 1–8.

© 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).