Contents

List of Tables ............................................................................................................................................... 2 List of Figures .............................................................................................................................................. 2 Chapter 1 Introduction ............................................................................................................................ 3 1.1 1.2 1.3 Need for Modeling of Soot Particles ............................................................................................ 4 Objective ...................................................................................................................................... 5 Organization of Report ................................................................................................................. 5

Chapter 2 Chemical Modeling of Combustion and Soot Growth ........................................................... 6 2.1 Soot Formation............................................................................................................................. 6

2.1.1 Chemical Mechanisms............................................................................................................... 6 2.1.2 Soot Formation and Growth ..................................................................................................... 8 2.1.3 Discussion of a soot growth model ......................................................................................... 10 2.2 Modeling a Chemical Mechanism .............................................................................................. 12

2.2.1 Solving Methodologies ............................................................................................................ 12 2.2.2 Solver Schemes ....................................................................................................................... 13 2.2.3 Stiff Systems - Difference between an Implicit and an Explicit Scheme .................................. 14 Chapter 3 Current Work and Results .................................................................................................... 16 3.1 Current Work ............................................................................................................................. 16

3.1.1 Mechanism Input .................................................................................................................... 16 3.1.2 Explicit Methods ..................................................................................................................... 16 3.1.3 Implicit Method....................................................................................................................... 17 3.2 3.3 Results ........................................................................................................................................ 20 Conclusions ................................................................................................................................ 23

Chapter 4 Summary and Future Work .................................................................................................. 24 4.1 4.2 Summary .................................................................................................................................... 24 Future Work ............................................................................................................................... 24

REFERENCES ............................................................................................................................................. 24

1

List Of Tables
Table 1 : Indian Emission Standards (4-Wheel Vehicles) ............................................................................ 3 Table 2 : Bharat (CEV) Emission Standards for Diesel Construction Machinery .......................................... 4 Table 3 : Generic reactions in Mechanism 3 ............................................................................................... 8

List of Figures
Figure 1 : Soot particles coalescence model. ............................................................................................ 11 Figure 2 : Instability encountered in integrating a stiff equation using an explicit method. ..................... 15 Figure 3 : Graphs of Mole fractions of species vs Time for different temperatures.................................. 22 Figure 4 : Magnified view of Figure 3.(b) to observe the combustion propagation phase for T=700 K ... 23

2

Chennai. For 2-and 3-wheelers.Chapter 1 Introduction The first Indian emission regulations were idle emission limits for both stationary and nonstationary which became effective in 1989. the National Auto Fuel Policy was announced. the National Auto Fuel Policy introduced certain emission requirements for interstate buses with routes originating or terminating in Delhi or the other 10 cities. Hyderabad. Emission Standards for NON-Road Diesel Engines Emission standards for diesel construction machinery were adopted on 21 September 2006. Chennai NCR*. which envisaged a phased program for introducing Euro 2 . Kanpur and Agra The above standards apply to all new 4-wheel vehicles sold and registered in the respective regions. Table 1: Indian Emission Standards (4-Wheel Vehicles) # Standard India 2000 Bharat Stage II Reference Euro 1 Euro 2 Date 2000 2001 04/2003 04/2005 Region Nationwide NCR*. Pune. 2005 and Stage III standards came in force from April 1. Kolkata.4 emission and fuel regulations by 2010. 11 Cities† Nationwide NCR*. 2010. 11 Cities† Bharat Stage III Euro 3 04/2005 04/2010 Bharat Stage IV Euro 4 04/2010 * National Capital Region (Delhi). which were gradually tightened during the 1990’s. Mumbai. 11 Cities† Nationwide NCR*. Since the year 2000.and three-wheeled vehicles. Surat. # Source http://www. In addition. Ahmedabad. These idle emission regulations were soon replaced by mass emission limits for both gasoline (1991) and diesel (1992) vehicles. Bharat Stage II was applicable from April 1. Kolkata. On October 6. 2003. India started adopting European emission and fuel regulations for four-wheeled light-duty and for heavy-duty vehicles. Secunderabad. The standards are structured into two tiers: 3 .dieselnet. India’s own emission regulations still apply to two.com/standards/in/[17] † Mumbai. The implementation schedule of EU emission standards in India is summarized in Table 1. Bangalore.

5 5.2 9. today’s diesel engines have inevitably adopted Diesel Particulate Filter (DPF) after-treatment systems.0 4. the after-treatment devices 4 . The standards are summarized in Table 2.80 0.dieselnet.60 0.2 - 1. Bharat (CEV) Stage II—These standards are based on the EU Stage I requirements.6 6. 2. Besides.3 - 7.0 6.3 1.0 3.3 1.0 6.0 9. it is generally agreed that the use of DPFs is not an ultimate solution for diesel engines to meet forthcoming.80 0.5 7.3 1.0 5.0 8.70 0.2 9.40 0.com/standards/in/[17] 1.0 5. Bharat (CEV) Stage III—These standards are based on US Tier 2/3 requirements.5 1. One of the reasons is that the regeneration of DPFs sets an upper limit on their conversion efficiency and lifespan.3 1. Table 2: Bharat (CEV) Emission Standards for Diesel Construction Machinery # Engine Power kW Bharat (CEV) Stage II P<8 8 ≤ P < 19 19 ≤ P < 37 37 ≤ P < 75 75 ≤ P < 130 130 ≤ P < 560 P<8 8 ≤ P < 19 19 ≤ P < 37 37 ≤ P < 75 75 ≤ P < 130 130 ≤ P < 560 Date CO HC HC+NOx g/kWh NOx PM* 10/2008 10/2008 10/2007 10/2007 10/2007 10/2007 04/2011 04/2011 04/2011 04/2011 04/2011 04/2011 8.1 Need for Modeling of Soot Particles In order to meet this ambitious PM (Particulate Matter) target. but also cover smaller engines that were not regulated under the EU Stage I.54 0.5 4.6 5.85 0.2 9.20 Bharat (CEV) Stage III (To be implemented) *PM = Particulate Matter # Source http://www.2 9.7 4.5 5.1.2 9.00 0.3 1.85 0. even more stringent PM emissions standards.5 7.5 6.85 0. Although the DPF systems can help to effectively reduce diesel tailpipe PM emissions.30 0.

these optimized combustion modes have demonstrated great potential of being able to simultaneously reduce NOx and PM emissions to levels acceptable for future emissions regulations. most importantly. 1. It also investigates possible soot growth scenarios which can be employed at a later stage. Unlike conventional diesel operation. It analyzes the methods used by commercial solvers for the same. we need to first efficiently model a combustion system. modern diesel technologies take advantage of injection system improvement.add additional cost to new vehicles and.3 Organization of Report Chapter 2 of this report gives an account of the chemical modeling of soot formation. engine geometry modification and even fuel/lubricant oil reformation. 5 . minimize engine-out and/or tailpipe emissions. Chapter 4 contains summary of the current work and recommendations for future work.2 Objective The objective of this stage of the report is to calculate the amount of soot released when diesel is burnt in a constant volume chamber. The report studies various mechanisms which can be used to concisely model the same. For better solutions. To optimize combustion systems. Second part of Chapter 2 studies various methods used to solve a chemical mechanism as an initial value problem. It studies stiffness in a set of equations which is very commonly encountered when solving a reaction mechanism.2. It also analyses a typical soot model and what it comprises of. It applies methods that can be used to solve the mechanism as a system of ordinary differential equations. in turn. and researchers are pursuing methods that are able to optimize combustion systems and. It presents reaction mechanisms which have been investigated and soot models associated with them. Chapter 3 presents the algorithms used for implementing the methodologies examined in Section 2. 1.1. their results and conclusions. they cannot help to improve CO2 emissions and fuel consumption. Changes in design parameters as well as their effects on PM and other emissions can be used to design better systems with lower emissions.

It is characterized by a rapid rise in the apparent heat release rate. and the apparent heat release rate curve changes gradually over a relatively long crank angle period. As a result of this. OH and H2O2 balance reactions 5. C7H8 pyrolysis and oxidation reactions 3. The premixed combustion phase is a result of fast. combustion mechanisms for diesel can have as many as 10000 reactions or even more. In the second phase.Chapter 2 Chemical Modeling of Combustion and Soot Growth 2. The following are the reaction mechanisms considered: 1. when computationally solving a chemical mechanism the number of reactions is reduced drastically so that the time taken to solve it is not too long. mixing between the fuel and the surrounding charge becomes a dominant factor that limits the rate of heat-release-controlling reactions. C7H16 pyrolysis and oxidation reactions 4. a considerable number of soot particles (or PMs) are formed in the mixing-controlled phase.1. However.1 Chemical Mechanisms Diesel is a mixture of many hydrocarbons with significant contributions by hydrocarbons from 10 C atoms to 18 C atoms. Chen et al. Due to the existence of over-rich mixtures in regions surrounded by high-temperature stoichiometric burning. direct-injection (DI) diesel engines employ a single spray injection event and their in-cylinder combustion demonstrates a unique heat-release behavior. The fuel is assumed to be comprised of only C14H28. spontaneous reactions of the vapor-fuel and air mixture. lasting only a few crank angle degrees after ignition. The heat release has two distinct phases: the premixed phase and the mixing-controlled phase. [1] have considered a mechanism with 56 species and 145 reactions that is based on the generation of toulene in the very first step followed by further oxidation reactions. by (C7H8 = toulene) 5𝐶14 𝐻28 + 𝑂2 → 3𝐶7 𝐻8 + 7𝐶7 𝐻16 + 2𝐻2 𝑂 2. It is one of the most reduced reaction systems possible for diesel combustion that includes oxidation and pyrolysis of species generated from breaking down of diesel.1 Soot Formation Conventionally. Soot precursor formation reactions 6 . Also. The basic outline of the mechanism is as follows: 1. 2. Breaking down of Fuel. The reaction mechanisms shown here are reduced mechanisms that try to be as accurate as possible and include as many reaction scenarios as possible. though many of them are oxidized quickly on the high-temperature burning surfaces. with increasing the number of reactions the computational requirements go up. an OH release mechanism and finally soot generation in terms of global C atoms.

Surface growth 6. Duran et al. 𝐶𝑂 + 𝑂𝐻 ↔ 𝐶𝑂2 + 𝐻 3. but the used values of the rate constants are calculated through extensive experimentation to disguise them as single steps. Soot production reactions 2. This mechanism is suitable only for tracking the soot growth. In reality. Precursor species (R) Oxidation 𝑅 → 𝑃 𝑥𝑃 → 𝑃 𝑃 + 𝐶2 𝐻2 → 𝑃 + 𝐻2 𝑃 + 𝑂2 → 𝑃 + 2𝐶𝑂 1 𝑃 + 𝑂𝐻 → 𝑃 + 𝐶𝑂 + 𝐻2 2 𝐶2 𝐻2 + 𝑂2 → 2𝐶𝑂 + 𝐻2 𝑧 1 𝑅 + 𝑂2 → 𝑧𝐶𝑂 + 𝐻2 2 2 where P = CX and R = an arrangement of z C atoms (precursor species) The OH balance mechanism is: 1. Acetylene (C2H2) formation 2 2 2 2. 𝐻 + 𝑂2 ↔ 𝑂 + 𝑂𝐻 2.6. Particle inception 4. [4] have described a model with 10 species and nine major phenomenological steps towards soot generation. [5] proposed a detailed kinetic mechanism consisting of 432 reactions and 120 species. 𝐻2 𝑂 + 𝑀 ↔ 𝐻 + 𝑂𝐻 + 𝑀 6. It is the largest mechanism used in this study. OH-related surface oxidation 8. Precursor species (R) formation 𝐶2 𝐻2 → 𝑅 + 𝐻2 𝑧 3. 𝑂 + 𝐻2 ↔ 𝐻 + 𝑂𝐻 3. Overall the reaction system has about 15 reactions per component of the fuel mixture. The reaction system is: 𝑚 𝑛 𝐶𝑚 𝐻𝑚 +𝑛 → 𝐶2 𝐻2 + 𝐻2 1. Tao et al. 𝑂𝐻 + 𝐻2 ↔ 𝐻2 𝑂 + 𝐻 4. most of the reactions do not occur as they are mentioned and instead have several intermediate steps. The fuel is broken down until the mixture consisting of 7 . Particle coagulation 5. O2-related surface oxidation 7. A step may have many reactions or just one. separately. It deals with aliphatic as well as aromatic hydrocarbons. 𝑂 + 𝐻2 𝑂 ↔ 𝑂𝐻 + 𝑂𝐻 5. Acetylene Oxidation 9.

Mechanism 2 (Tao et al. The reactions for breaking down.2 Soot Formation and Growth • 𝑂2 ↔ ∅1 + 𝐻𝑂2 • 𝐻 ↔ ∅1 + 𝐻2 𝑂2 ↔ ∅• + 𝐻𝑂2 2 𝐻 ↔ ∅• + 𝐻2 2 There are different ways to approach soot formation. Ø2H = Naphthalene Since mechanism 3 (Duran et al. This is significantly different from practical experiences where the fuel takes much more time to break down. Also.) lacks accuracy but when implemented is very quick to provide results. C14H28 & C12H24 respectively. Mechanism 1 (Chen et al.1. R• = Free radical. 2. the fuel is consumed very quickly compared to other reactions in the system. PH = Olefinic hydrocarbon. However it has virtually no intermediates that can be tracked and hence is very unreliable. Ø1H = Benzene. Chen et al. proposes to use the global mass of C atoms in order to validate the reduced Mechanism 1.C17H36. So far. Though such a calculation is 8 . C14H30 & C12H28 is reduced to C17H34. only Mechanism 1 has been completely solved for different temperature cases. either of C 2H5• or CH3•.) is faster than Mechanism 3 but is not very accurate. Table 3: Generic reactions in Mechanism 3 Initiation Olefin Formation Odd number of C atoms (n) Propagation Aliphatic Olefin Decomposition • 𝑅𝐻 + 𝑂2 → 𝑅 • + 𝐻𝑂2 • 𝑅𝐻 + 𝐻𝑂2 → 𝑅• + 𝐻2 𝑂2 • 𝑅• + 𝑂2 → 𝑃𝐻 + 𝐻𝑂2 𝐻 • + 𝑃𝐻 → 𝐶3 𝐻6 + 𝐻 • + 𝑃𝐻 → (𝑛−1) 2 (𝑛−5) 2 • 𝐶2 𝐻4 + 𝐶2 𝐻5 • 𝐶2 𝐻4 + 𝐶𝐻3 (𝑛−4) 2 • 𝐶2 𝐻4 + 𝐶𝐻3 Even number of C atoms (n) 𝐻 • + 𝑃𝐻 → 𝐶3 𝐻6 + 𝐻 • + 𝑃𝐻 → ∅1 𝐻 + ∅1 𝐻 + ∅2 𝐻 + ∅2 𝐻 + (𝑛−2) 2 • 𝐶2 𝐻4 + 𝐶2 𝐻5 Oxidation and Pyrolysis Aromatic Initiation Oxidation and Pyrolysis where RH = Aliphatic linear hydrocarbon. according to the first reaction. But it has the advantage of being extremely detailed and gives us the ability to track a lot of the intermediate species that are produced. oxidation and pyrolysis of aliphatic and aromatic hydrocarbons are shown in Table 3. Then it is completely broken into ethene and a residual radical.) has the most number of reactions it consumes the most time during solving.

it is not very helpful in the actual modeling procedure.[1] utilizes a kinetic mechanism to produce solid carbon particles through a set of global reactions that lead to the “graphitization” of the gaseous precursors as shown: 𝐶6 𝐻2 → 6𝐶(𝑠) + 𝐻2 𝐴2 𝑅5 → 12𝐶(𝑠) + 4𝐻2 where 𝐴2 𝑅5 = Cyclopentapyrene Tao et al. phenantherene. Soot particles are assumed to be dry. Soot Precursor and Soot Formation Different reaction mechanisms provide different reactions leading to the formation of soot or soot precursors.[4] proposes a precursor species R. where 𝑘5 = 1. which corresponds to about 100 carbon atoms. The rate for step 4 is 𝑅4 = 4 𝑘4 𝑁 2 . In the latter case. which is produced from acetylene. the gas mean-free path. [7]. Soot particles typically coagulate or break up when they collide with other soot species.28 nm. and the molecular viscosity of the gas. the soot precursor. the Poly-Aromatic Hydrocarbons (PAHs) in their radical states are called soot precursors as they lead to the formation of radicals which can further grow their ring structure. soot growth occurs in step 4 through coagulation ′ and in step 5 through surface growth. where 𝑘4 is the 1 collision frequency constant. particularly for young soot particles that contain significant amount of hydrogen. this assumption is accepted in the absence of other better alternatives for a phenomenological mechanism as this one. flouranthene.05 ∗ 104 exp(− 3100 T ) is from Leung et al. The incipient soot particles are assumed to have a diameter of 1.5 . Nevertheless. Since the gas mean free path is comparable with the particle size in high-pressure diesel combustion environments. 𝐴𝑠𝑜𝑜𝑡 9 . this constant is taken from Kazakov and Foster [6] and is calculated based on the free-molecular collision frequency. A soot particle typically goes through three phenomenological periods which are: 1.useful to test the accuracy of a particular mechanism. It contains z carbon atoms and is generally of the fullerene type (𝐶60 ). This produces Corranulene according to the reaction: 𝐵𝑒𝑛𝑧𝑜𝑝𝑒𝑟𝑦𝑙𝑒𝑛𝑦𝑙 + 𝐶2 𝐻2 → 𝐶𝑜𝑟𝑎𝑛𝑛𝑢𝑙𝑒𝑛𝑒(𝐶20 𝐻10 ) + 𝐻 2. Soot is formed in the third step as shown in the mechanism previously. The soot model proposed by Chen et al. all of which lead to the production of benzoperylenyl. Strictly speaking. this assumption is incorrect. Soot can be modeled as rings of C (similar to fullerenes) or as rings of aromatic hydrocarbons. Rate for step 5 is calculated as ′ 𝑅5 = 𝑘5 [𝐶2 𝐻2 ](𝐴𝑠𝑜𝑜𝑡 )0. graphite-like particles containing solely carbon atoms. Soot Growth As mentioned in Mechanism 2 previously. The most detailed of the 3 reaction systems (Mechanism 3) has many species such as anthracene.

The reactions causing destruction of soot particles or reduction in chain size generically resemble these: 𝑃𝑀 + 𝐻2 𝑂 → 𝐶𝑂 + 𝐻2 + (𝑃𝑀) 𝑃𝑀 + 𝑂2 → 𝑂 + 𝐶𝑂 + (𝑃𝑀) 𝑃𝑀 + 𝑂𝐻 → 𝐶𝑂 + 𝐻 + (𝑃𝑀) where 𝑃𝑀 = Particulate Matter 2. It starts with the assumption that soot particles have generated from radicals. Though it does not go into the details of modeling. the soot particles after inception follow a series of reactions given by: 𝐶20 𝐻10 + 𝐻 = 𝐶𝑜𝑟 − + 𝐻2 𝐶𝑜𝑟 − + 𝐶2 𝐻2 = 𝐶𝑜𝑟 1 + 𝐻 𝐶𝑜𝑟 1 + 𝐻 = 𝐶𝑜𝑟 1− + 𝐻2 𝐶𝑜𝑟 1− + 𝐶2 𝐻2 = 𝐶𝑜𝑟 2 + 𝐻 𝐶𝑜𝑟 2 + 𝐻 = 𝐶𝑜𝑟 2− + 𝐻2 𝐶𝑜𝑟 2− + 𝐶2 𝐻2 = 𝐶𝑜𝑟 3 + 𝐻 𝐶𝑜𝑟 3 + 𝐻 = 𝐶𝑜𝑟 3− + 𝐻2 𝐶𝑜𝑟 3− + 𝐶2 𝐻2 = 𝐶𝑜𝑟 4 + 𝐻 𝐶𝑜𝑟 4 + 𝐻 = 𝐶𝑜𝑟 4− + 𝐻2 𝐶𝑜𝑟 4− + 𝐶2 𝐻2 = 𝐶30 𝐻10 + 𝐻 where 𝐶𝑜𝑟 𝑛 = 𝐶𝑜𝑟 + n additional rings. it does incorporate surface growth.3 Discussion of a soot growth model A broader approach with a model for soot growth is described in Okuyama et al [8]. Soot Destruction Soot particle termination can occur by either oxidation by collision with an oxidizing agent or by coalescence with a larger particle which would result in one new soot particle. 𝑟 + 𝑑𝑟) exist.represents the total surface area of the soot particles which are assumed to be spherical.5 . 𝐶28 𝐻10 denoted by 𝐶𝑜𝑟 4 and 𝐶30 𝐻10 are the soot particles in this reaction system. it is assumed that first. The calculation of radical density (soot precursors) and subsequent generation (formation step) of soot is not discussed here. 𝐶20 𝐻10 and other subsequent species including 𝐶22 𝐻10 denoted by 𝐶𝑜𝑟 1 . The rate is assumed to be proportional to (𝐴𝑠𝑜𝑜𝑡 )0. Oxidizing agents for soot particles can be any of 𝐻2 𝑂. 10 . Also. At time 𝑡 from the start of soot formation. In Mechanism 3. 𝑁𝑠 𝑟 𝑑𝑟 soot particles per unit volume having radii in the interval (𝑟. 𝐶24 𝐻10 denoted by 𝐶𝑜𝑟 2 . and then. 3.1. the endothermic reactions occur for soot formation. the stoichiometric amounts of fuel and oxygen are consumed in the exothermic reactions to produce the maximum temperature possible. 𝐶26 𝐻10 denoted by 𝐶𝑜𝑟 3 . The radii of these soot particles increase at a mean rate 𝑟𝑎′ 𝑟 by both coalescence and surface growth. 𝑂2 and 𝑂𝐻.

small soot particles having radii in the interval (𝑟. large soot particles having radii in the interval (𝜂. if the centers of soot particles with radius 𝑟 exist in the coalescence region which is the spherical space between radius 𝜂 and radius 𝑙. The figures representing the soot model in two-dimensional form are shown in Figure 1(a)-(c). (a) Soot particles coalescence model. (b) Configuration of soot particles with radii larger than r. 𝜂 + 𝑑𝜂). Source : Okuyama et al. (c) Corrected configuration of soot particles. 𝑟 + 𝑑𝑟) will coalescence. (a) (b) (c) Figure 1. we consider the case of the coalescence of soot particles with radius 𝜂 and those with radius 𝑟 (𝜂 > 𝑟). As shown in Figure 1(a). or due to the generation of inter-particular force with. they become extinct by collision with.The coalescence of soot particles is modeled by the coalescence probability which is statistically evaluated for soot particles based on Tanaka’s model [9].[8] 11 . Consequently.

The final equation for the calculation of rate of change of soot density 𝑁𝑠𝑖 𝑟 𝑑𝑟 with respect to time has three major terms corresponding to: 1. i. For a reaction given by. … 𝑁 − 1. 𝑛 is exponential factor. Growth by coalescence 3. 𝑖 = 0. 3. 𝑑𝑦𝑖 (2. the larger one grows. So for a reaction consisting of 𝑁 reactions. without introducing the effect of pressure the problem is reduced to the simultaneous solving of 𝑁 first order rate equations.1) where 𝐴 is the pre-exponential factor. while the smaller one becomes extinct.1. … 𝑦𝑁−1 . 2. 12 .2) = 𝑓𝑖 𝑡. … 𝑁 − 1 𝑑𝑡 where the functions 𝑓𝑖 on the right-hand side are known. [10]. 𝑖 = 0. 2.The main assumptions for the soot growth model are as follows: 1. Soot particles are spherical. having the general form. Formation and termination by collision with radicals 2. In the case of coalescence between two soot particles of different sizes. 2.2. 𝑦0 .e 𝑦𝑖 . and the calculation of the soot formation and growth process is performed three-dimensionally in a unit volume.2 Modeling a Chemical Mechanism In the present study the focus has so far been on calculating the amount of soot formed in various chemical mechanisms with the same initial conditions for different temperatures as cases and making comparisons between them.1 Solving Methodologies The primary equation for chemical kinetics is the Arrhenius Equation. 𝑅𝑎𝑡𝑒 𝑜𝑓 𝑅𝑒𝑎𝑐𝑡𝑖𝑜𝑛 = 𝐴𝑇 𝑛 𝑒 −𝑅𝑇 ∗ 𝐵 𝐸𝑎 𝑏 𝐶 𝑐 (2. Termination by coalescence with larger particles. bB +cC → dD + eE. 𝐸𝑎 is the activation energy and 𝑇 is the temperature. The surface growth model used is similar to the one mentioned in Tao et al. the rate of reaction[18] is given by. [4]. The surface growth of soot particles is due to the collision of radicals with a soot particle as discussed in Mosbach et al.

𝑦𝑛+1 = 𝑦𝑛 + 𝑕𝑓(𝑡𝑛 . 𝑦 ′ 𝑡 = 𝑓 𝑡. This gives algebraic formulas for the change in the functions when 𝑡 is increased by one “stepsize” 𝛥𝑡. 𝑦𝑛 ) The Euler method is explicit.2.e.2 Solver Schemes For the problem. For very small stepsizes. The methods used in this study are explicit Euler. (2. 𝑦 𝑡0 = 𝑦0 . Methods in which the new time-step values are dependent on only the old time-step values are called explicit methods. this results in a good approximation of the solution. Methods in-between the two are called semi-implicit or semi-explicit methods. 2. The basic idea of any routine for solving the initial value problem always involves rewriting the 𝑑𝑦’s and 𝑑𝑡’s in the equations as finite steps 𝛥𝑦 and 𝛥𝑡. and multiplying the equations by 𝛥𝑡. the 4th Order Runge-Kutta method for this problem is given by the following equations: 1 (2.3) the solver algorithms are based on one of the following: 1. This procedure is known as Euler’s method.4) 13 . the solution 𝑦𝑛+1 is an explicit function of 𝑦𝑖 for 𝑖 ≤ 𝑛. Explicit Euler Scheme Euler method from 𝑡𝑛 to 𝑡𝑛 +1 = 𝑡𝑛 + 𝑕 is. 𝑘1 = 𝑓(𝑡𝑛 . 𝑦𝑛 + 𝑕𝑘1 ) 2 2 (2. Explicit Runge-Kutta Scheme Then.5) (2. Methods in which new time-step values are dependent on the most recently available values are implicit methods. 𝑦 𝑡 .7) (2. 2. i. explicit Runge-Mutta method and implicit Backward Differentiation method.4) 𝑦𝑛+1 = 𝑦𝑛 + 𝑕(𝑘1 + 2𝑘2 + 2𝑘3 + 𝑘4 ) 6 𝑡𝑛+1 = 𝑡𝑛 + 𝑕 where 𝑦𝑛+1 is the 4th order Runge-Kutta approximation of 𝑦(𝑡𝑛+1 ) and.Since this is an initial value problem all 𝑦𝑖 ‘s at 𝑡 = 0 are known. 𝑦𝑛 ) 1 1 𝑘2 = 𝑓(𝑡𝑛 + 𝑕.6) (2.

8) (2. even though accuracy requirements allow a much larger stepsize.10) 𝑦𝑛 = 𝑦 𝑡𝑛+1 − 𝑕 = 𝑦 𝑡𝑛+1 − 𝑕 |𝑡 𝑛 +1 + 𝑂(𝑕2 ) 𝑑𝑡 which upon ignoring higher order terms gives. the possibility of a stiff set of equations arises.3 Stiff Systems . the presence of the 𝑒 −1000𝑥 term would require a stepsize 𝑕<< 1/1000 for the procedure to remain stable. Backward Differentiation Scheme The implicit analogue of the explicit Euler method is the backward Euler method [11. Example consider the following set of equations. 𝑦𝑛 + 𝑕𝑘3 ) (2.Difference between an Implicit and an Explicit Scheme As soon as one deals with more than one first-order differential equation. we have to follow the solution on the shortest length scale to maintain the stability of the integration.2. 𝑢 = 2𝑦 − 𝑧 𝑣 = −𝑦 + 𝑧 we find the solution.9) 3. Consider the single equation. 𝑡𝑛+1 ) (2. 𝑦𝑛+1 = 𝑦𝑛 + 𝑕𝑓(𝑦𝑛+1 . 𝑢 0 = 1 𝑣 0 = 0 By means of the transformation. This is based on the following Taylor series expansion: 𝑑𝑦 (2.1 1 𝑘3 = 𝑓(𝑡𝑛 + 𝑕.18]. 𝑦𝑛 + 𝑕𝑘2 ) 2 2 𝑘4 = 𝑓(𝑡𝑛 + 𝑕.12) 𝑦 ′ = −𝑐𝑦 14 . Stiffness typically occurs in a problem where there are two or more very different scales of the independent variable on which the dependent variables are changing. 𝑢′ = 998𝑢 + 1998𝑣 𝑣 ′ = −998𝑢 − 1999𝑣 with boundary conditions. (2. 𝑢 = 2𝑒 −𝑥 − 𝑒 −1000𝑥 𝑣 = −𝑒 −𝑥 + 𝑒 −1000𝑥 . This occurs even though the 𝑒 −1000𝑥 term is negligible while determining the values of 𝑢 and 𝑣 away from the origin.11) 2. As a result. If we solved this system using the explicit method.

Figure 2. 𝑦𝑛+1 → 0. In this case. Courtesy: Numerical Recipes – The Art of Scientific Computing. fig. 𝑦𝑛 1 + 𝑐𝑕 (2. we get the backward Euler scheme: Or 𝑦𝑛+1 = The method is absolutely stable: Even as 𝑕 →∞. The stability of the integration (shown as the unstable dotted sequence of segments) is determined by the more rapidly varying dashed solution. Here it is supposed that the equation has two solutions.5.13) 15 .e. the solution at late times) for large stepsizes. which is in fact the correct solution of the differential equation. as in that case |𝑦𝑛 | = 1 as 𝑛 →∞ even though the solution of the equation is bounded.The explicit method is called so because the new value 𝑦𝑛+1 is given explicitly in terms of the old value 𝑦𝑛 . The implicit method converges to the true equilibrium solution (i. shown as solid and dashed lines. The method is unstable if 𝑕 > 2/𝑐.1 . even after that solution is effectively zero.. Example of an instability encountered in integrating a stiff equation (schematic). 17. pg. where the right-hand side is evaluated at the new 𝑦 location. 3rd Edition. 932[11] The simplest solution is to resort to implicit differencing.

Chapter 3 Current Work and Results 3. 3. and 𝑅𝑖 is the reaction rate..2.1) Where 𝑐𝑖𝑗 is the coefficient of species 𝑗 in the reaction 𝑖. 𝐸𝑎𝑖 as per the Arrhenius equation.1 Mechanism Input The code has been written to be able to accept any reaction mechanism as input through a text file to enable quick comparisons without any change in the code. It is (𝑐𝑖𝑗 < 0) when 𝑗 is a reactant.2. 3. This was achieved with the use of various string handling functions.C.1. Temperature is assumed to be constant for the time-being. It is calculated using the values of 𝐴𝑖 . the focus has been on modeling various chemical mechanisms by using different methods and making comparisons between them.b. explicit Fourth Order Runge-Kutta methods with subroutines to handle stiffness and implicit Backward Differentiation Methods with a Newton-Rhapson method based solver. the function y i for species i can be calculated as: 𝑛 𝑦𝑗′ = 𝑖=0 𝑐𝑖𝑗 . The methods used are explicit Euler.D are the chemical representations of the reactants and products.1 Current Work So far. 𝑛𝑖 .. 𝑅𝑖 (3. The coefficients of the reactants and products are stored in a matrix where the number of rows is the number of reactions and the number of columns is the number of species.d are the reactant and product coefficients and A. . A is the pre-exponential factor. The format of the input reaction mechanism is as follows: aA+bB=cC+dD A1 n1 Ea1 wW+xX=yY+zZ A2 n2 Ea2 . Hence.2 Explicit Methods Both Explicit Euler and Explicit Runge-Kutta Methods are implemented as mentioned previously in section 2.c.1. 16 . Where a. n is exponential factor and Ea is the activation energy.B.

Hence. then the solver enters a subroutine where it calculates the concentrations of the reactants and products with a time-step t x = t n/(100. then the reaction is assumed to have proceeded to completion and the concentration of that reactant is made equal to 0. 𝐾1 =4 𝐾2 =0 were chosen. and is a multistep method. If this time-step is lesser by a factor more than 100 than the solver’s current time-step. [14-16]. the implicit method that was used was chosen from many available options such as EPSODE. The prediction step is the first step during any time-step.𝑛 = 𝑗 =1 𝛼𝑛.Handling Stiffness in Explicit Method For a typical set of equations which was solved the time-step required to ensure stability using either of the explicit methods discussed above was about 10-41 seconds.𝑛 (0) (0) (3. the solver calculates the time-step for which that reactant would become 0. Accordingly.𝑗 𝑦𝑖. it predicts a value greater than the probable value for the new time-step and uses backward differentiation to find the corrector to be subtracted.0 𝑦𝑖. The basic idea behind this method is the same as any predictor-corrector method. However LSODE is documented to have the best accuracy for large and stiff systems. until it reaches the new time-step. the other reactants and products participating in that reaction are updated to keep the concentrations accurate.𝑗 𝑦𝑖. If a time-step is very large for a particular reaction. Since this was very time-consuming a subroutine was added that introduced fractional time-stepping. When the if condition is triggered.100) with the last time-step as the starting point. and is explained shortly. the general form of this technique is given by: 𝐾1 𝐾2 𝑦𝑖. GCKP84. It uses integration techniques developed by Gear[13]. to calculate the percentage amount of change instead of a sign change. CHEMEQ etc.3 Implicit Method LSODE. 𝐾1 𝑦𝑖. to allow for bigger time-steps.𝑛 are denoted by 𝑦𝑖.3) The predicted values of 𝑦𝑖. This implies there are many steps taken inside a single time-step.𝑛 and 𝑦𝑖.𝑛 . If this is not the case and the factor is less than 100. 3. an if condition kept track of any reactants that would become negative while solving for a particular reaction in the loop (this factor could be made more accurate. but that increased the time consumed).𝑛−𝑗 + 𝑕𝑛 𝑗 =0 𝛽𝑛. Each of these methods has its pros and cons. 17 .1.𝑛−𝑗 + 𝑕𝑛 𝛽𝑛.2) For LSODE. To detect and correct this. and use of a Newton-Rhapson solver.𝑛 = 𝑗 =1 𝛼𝑛.𝑗 𝑦𝑖.𝑛−𝑗 (3.𝑛 and 𝑦𝑖. the concentration of the reactant can become negative.

if we have the 𝑧𝑛−1 th row.𝑛 is calculated from the known value of 𝑦𝑖.𝑛 . 𝑷𝑖𝑘 .𝑛 2 3 4 𝑕𝑛 𝑕𝑛 𝑕𝑛 = 𝑦𝑖.𝑛 + 𝑕𝑛 𝛽𝑛.𝑛 at every step.3) to give [13-16].𝑛 .𝑛 = 𝑖𝑘 (𝑚 ) 𝜕𝐺𝑖 𝜕𝑦𝑘 𝑦 𝑛 (𝑚 ) = 𝛿𝑖𝑘 − 𝑕𝑛 𝛽𝑛.𝑛 − 𝑕𝑛 𝛽𝑛. then the predicted values of 𝑦𝑛 are calculated as 18 .0 𝑦𝑖. 𝑦𝑖.8) and 𝐽𝑖𝑗 are the elements of the Jacobian Matrix. to maintain accuracy.9) To reduce computation work. For this purpose.𝑛 (0) (0) (3. where the ith row contains the following 5 elements.4) This equation needs to be solved for 𝑦𝑖. 𝑧𝑖. 𝑦𝑖.(0) (3.𝑛 = 𝑦𝑖.𝑛 (𝑚 ) (3. Thus.𝑛 = (𝑚 ) 𝑦 𝑛 (𝑚 ) (3.𝑛 . 𝛿𝑖𝑘 =0 =1 𝜕𝑓𝑖 𝜕𝑦𝑘 𝑖 ≠ 𝑘 𝑖 = 𝑘 (3. Since the predicted values will satisfy eq.7) where 𝛿𝑖𝑘 is the Kronecker symbol. 𝑕𝑛 𝑦𝑖. (3. 𝑁 (3.The value of 𝑦𝑖. the Jacobian Matrix and consequently. They are updated only when it as absolutely essential for convergence or when the last update was 20 time-steps back.𝑛 − 𝑁 𝑦𝑖. (0) (0) 𝑦𝑖.𝑛 .𝑛 (𝑚 ) = −𝐺𝑖 𝑦𝑛 (𝑚 ) 𝑖 = 1.𝑛 𝑦𝑘. The array is of size N*5. They can be substituted back into eq.3). 𝑃 . 𝐽𝑖𝑘 .𝑛 is an approximation to (𝜕 𝑞 𝑦𝑖 𝜕𝑡 𝑞 ).0 𝑦𝑖. 𝑦𝑖.0 𝑦𝑖.5) This is done iteratively according to the following method.𝑛 2! 3! 4! [𝑞] Where 𝑦𝑖.6) Where 𝑚 + 1 is the current iteration number and 𝑷 is given by. the Newton-Rhapson method based solver is used. The predicted values are calculated from a Nordsieck array which contains the history of the solution.𝑛 (0) . the matrix 𝑷 are not updated at every step.𝑛 𝑘=1 (𝑚 ) (𝑚 +1) − 𝑦𝑘.0 𝑦𝑖.𝑛 − 𝑕𝑛 𝛽𝑛 .0 𝐽𝑖𝑘. The problem is transformed to finding the zero of 𝐺 in the equation. (3. 𝐺(𝑦𝑛 ) = 𝑦𝑖.𝑛 .𝑛 − 𝑕𝑛 𝛽𝑛.𝑛−1 and the predicted value of 𝑦𝑖.

Also. for each of the species in 𝑦𝑛 are supplied by the user and are stored in an absolute tolerance array.13) where 𝜀𝑟𝑒𝑙 is the local relative tolerance and 𝜀𝑎𝑏𝑠 . 19 . After changing the 𝑕. . 𝑑𝑚 𝑑𝑚+1 = 𝑑𝑚 (3.𝑛 − 𝑦𝑖.𝑧𝑛 = 𝑧𝑛 −1 𝐴(𝑞) where matrix 𝐴(𝑞) is given by 𝐴𝑗𝑘 𝑞 = 𝑗 (0) (3. 𝐸𝑊𝑇𝑖 = 𝜀𝑟𝑒𝑙 𝑦𝑖. 𝑘 = 0. otherwise matrix 𝑷 is updated and the corrector step is repeated. Convergence is said to occur if 𝑁 𝑖=1 𝑚 𝑚 𝑦𝑖. if the corrector step passes the convergence test after 𝑚 (𝑚 > 3) steps then an estimate of a local truncation error is made based on the values of local errors. . The absolute tolerances.14) 𝑑𝑚 = 𝑁 ≤ 𝑐𝐿 The constant 𝑐𝐿 is chosen to make convergence tests consistent with the local error [13]. .𝑛 −1 𝐸𝑊𝑇𝑖 2 (3. a relative tolerance is provided by the user.11) where 𝑘 is the binomial coefficient defined as.15) 𝑑𝑚−1 assuming that the iterates converge linearly. Before the values of 𝑦𝑛 can be accepted they are tested for convergence. 𝑗 𝑗! = 𝑘 𝑘! 𝑗 − 𝑘 ! (3.10) 0 𝑗 𝑘 𝑗 < 𝑘 𝑗 ≥ 𝑘 𝑗. eq.𝑖 th (3.𝑖 is the 𝑖 element of the local absolute error tolerance matrix.4 (3.6 is iterated until convergence.12) Once we have obtained the predicted values. The initial values are the midpoint values between the predicted and the previous time-step values. The values of 𝑦𝑛 are accepted if they pass the following error test. If convergence does not occur in the first iteration then 𝑑𝑚+1 is estimated by. If the corrector iterates do not converge (pass the convergence test) in 3 steps then the timestep 𝑕 is reduced and the step is repeated if matrix 𝑷 is current.1.𝑛−1 + 𝜀𝑎𝑏𝑠 . An error vector is constructed as follows. 3.

Since in the beginning the history array would not have any values inside it.2 Results The solver based on explicit Runge-Kutta algorithm is solved for Mechanism 1. the initial iterations were carried out using the explicit Runge-Kutta method until all the species had nonzero values for the last 4 time-steps in their history arrays. Ø = 1. If the step is not accepted even after reaching 𝑕/10𝐸10. 𝑛𝑜𝑥𝑦𝑔𝑒𝑛 = 0. The mole fractions were 𝑛𝑓𝑢𝑒𝑙 = 0.𝑁 𝑖=1 𝑚 0 𝑦𝑖. H2O.03 The plot for concentrations of species vs. 𝑛𝑤𝑎𝑡𝑒𝑟 = 0.01.16) 𝑁 ≤ 𝑡𝐿 If the error test fails the step is repeated again with an 𝑕 = 𝑕/5 until either the step is passed or the 𝑕 reaches 𝑕 < 𝑕/10𝐸10. 3. 700 K and 1000 K. OH. 𝑛𝑛𝑖𝑡𝑟𝑜𝑔𝑒𝑛 = 0. 700K.𝑛 − 𝑦𝑖. The temperature is assumed to remain unchanged. time on a log scale for the different temperature cases is shown in Figure 3(a)(d).𝑛 𝐸𝑊𝑇𝑖 2 (3.e. then the Nordsieck array is updated with the current 𝑧𝑛 values. The initial fuel to air ratio was the stoichiometric ratio for C14H28. CO.21. 1000K and 2000K. i. 𝑛𝑐𝑎𝑟𝑏𝑜𝑛𝑑𝑖𝑜𝑥𝑖𝑑𝑒 = 0. The species tracked are C(s). H2. while HCO is tracked instead of CO for T = 2000K. (a) Temp = 300 K 20 . CO2 for T=300 K. The initial conditions were for four different temperatures of 300K.7768. O2.0002. If the step is accepted. the situation is considered hopeless and the procedure is terminated.

(b) Temp = 700 K (c) Temp = 1000 K 21 .

(a) Temp = 300 K. 1E-7 seconds for 1000K and at 1E-5 seconds for 700K. the maximum time reached by temperature = 2000K was about 1E-6. Graph of Log Concentrations vs Log Time.01) was consumed in 1E-25 seconds.5E-14 to maintain stability and accuracy. as can be clearly seen from the plots. This was observed while running the simulations. (d) Temp = 2000K. when all of the fuel (𝑛𝑓𝑢𝑒𝑙 = 0. The maximum time reached by the solver was 1E-4 for cases (a)-(c). whereas for case (d) the time-step was taken to be 0. The concentrations are in mole fractions. This occurs at around 0. the region of rapid progression of reactions is a trend visible across all of the other three temperatures. the reaction mechanism assumes the fuel to be broken down very quickly compared to other reactions. (b) Temp = 700 K. And. there is very little or no activity at all and concentrations of key species change only slightly. As a result.5E-10 seconds for 2000K.(d) Temp = 2000 K Figure 3. For temperature 300 K. Since reaching a time of 1E-3 would have taken 10 times the time taken to reach 1E-4. One must bear in mind that even though 22 . the simulations were stopped after about 6 days. (c) Temp = 1000 K. The timescales for (d) are lower than the others because of the highly elevated rate of reactions. As explained earlier on. Hence time-step had to be reduced to accurately monitor the reaction progress.5 seconds. The other reaction systems were solved with time-steps in the range of 10E-13 seconds and were allowed to run for up to a week.

The region with rapid progression of reactions can be attributed to the propagation phase of combustion. the rates of reactions attain their maxima owing to generation of intermediate species. and in agreement with theory. Its growth rate is accelerated in the region with rapid progression of reactions. 23 . it displays its behaviour as an intermediate species. it happens over a longer duration of approximately 1E-4.denoted by solid black lines in Figures 3 & 4) increases steadily for all the four cases and is clearly visible for temperatures of 700 K. the graphs are logarithmic and hence the time-span seems shorter. Figure 4: Magnified view for t=1E-4. 3.this phase appears to be over quickly for T = 700 K.881 seconds. spiking during the rapid reaction region and slowly dying out later on.88106 seconds as shown in Figure 4. We know from the Arrhenius equation that reaction rates are exponentially related to temperatures. Note that in Figure 3.8811 seconds to t = 1E-4. and this fact is mimicked by the solver.3 Conclusions The global soot content (C(s) . Based on these reasons. highlighting the rapid combustion phase for T= 700 K.(d). when in reality. in accord with the other species in the reaction mixture. where. HCO was tracked. 1000 K and 2000K. the solver seems to be qualitatively correct.

the implicit code is being implemented. and a computational code was created to solve them for different temperature cases. The next step will be implementing a soot model by incorporating surface growth and aromatic site modeling. Preliminary results suggest that the code is qualitatively correct.1 Summary Emission of Particulate Matter (PM) is a big concern nowadays with the rising threat of global warming. further extension to the work would include adding an ignition mechanism and employing spatial dependence. After extensive testing of the code for different reaction mechanisms. a solver based on implicit methods. One of the major components of PM are soot particles and their modeling is an important area of research. The code is ready though it needs to be debugged. Since the system has not reached steady state. the focus will be on implementing temperature variation by calculating heat release per reaction and implementing the effect of pressure on uni-molecular reactions. If the results look favourable.2 Future Work Presently. 4. Methods to mathematically model such mechanisms as Ordinary Differential Equations were analyzed. the curves representing concentrations do not reach steady state and hence the reaction system cannot be assumed to have reached steady state at the end. small amounts of hydrogen are observed in the later stages. Chapter 4 Summary and Future Work 4. which may be expected from theory for a mixture with Ø = 1. Water and carbon dioxide can be seen approaching their stoichiometric reaction coefficients on the product side as we reach the end. 24 . Various chemical mechanisms for combustion of diesel were studied. though it requires faster routines to reduce the time consumed while solving.As can be seen. Also. Generation of soot particles from a combustion mixture has been modeled. These shortcomings highlighted the need for a faster solver viz. the trace amount of hydrogen cannot be established as a significant component of the final mixture of products.

97 (1975). A comprehensive modeling study of nheptane oxidation.J. Kubo S. 156 (2009). Fuel. Melenk H. Shuai S.. 3) Curran H. 72–78. Yoshida H. No.. Vol. Foster D. Vol.. 1223–1234 5) Duran A.K. Raj A. Lindstedt R. Vol.. Towards a detailed soot model for internal combustion engines.. A simplified reaction mechanism of soot formation in non-premixed flames. 56 (2004).Trans.E. 87 (1991). Vol.P. Vol.. Westbrook C. Monteagudo J..S. Hydrocarbon ignition: automatic generation of reaction mechanisms and applications to modeling of engine knock... Liu Y. Modelling soot and SOF emissions from a diesel engine. ASME. Vol. Foster D. Koda M. Modeling of soot formation during DI diesel combustion using a multi-step phenomenological model. International Journal of Thermal Sciences. Kim K.M.D. 1156– 1165. 1927–1936 2) Chevalier C.. 9) Tanaka H. Nine-step phenomenological diesel soot model validated over a wide range of engine conditions. Celnik M. 4625–4635.K. 47 (2004). Heat Transfer... Vol. 982463 (1998). Carmona M. SAE Paper. 7) Leung K.. Combustion and Flame. Pitz W. 93–101.M. A soot formation embedded reduced reaction mechanism for diesel fuel. Zhang H. 209–225 6) Kazakov A..... Westbrook C. Vol. Combust Flame. Wang Jianxin. Jones W. 25 .. Modeling of soot particles growth in fuel-rich premixed flame. 289–305. Vol. Reitz R. 8) Okuyama M. Chemosphere. Echigo R. Warnatz J.P. 24 (1992).J.. 149–77 4) Tao F.... Kraft M. Hanamura K. SympCombust.E.. A theoretical study of dropwise condensation.R.. International Journal of Heat and Mass Transfer...J.O. 114 (1998).. 88 (2009). Koganezawa T. Pitz W. 48 (2009). 10) Mosbac S. Combustion and Flame....References 1) Chen W. Gauffuri P.

Vol.A.dieselnet...A.C. 18) www. A.. 14) Radhakrishnan K. Third Edition (2007). 3rd Edition (1985).cpluplus.. No.R.com last accessed on 13/9/10.W. 13) Gear C. Cash J. Vol...H. Efficient Implementation of Fully Implicit Methods for Atmospheric Chemical Kinetics. Brugnano L. R. Stiff Computation. 26 . 87 (1997). Potra F. Carmichael G.11) Press W. 19) www. 15) Sandu A. 17) www. Flannery B. 14 (Mar 1971).wikipedia.. NASA Technical Paper. Vetterling W. 176-180. 12) Aiken R. Teukolsky S. 2372 (1984).com/standards last accessed on 8/8/10.P. 1-9. 129 (1996). Vol. Journal of Computational and Applied Mathematics. A note on the efficient implementation of implicit methods for ODEs. Journal of Computational Physics. The Automatic Integration of Ordinary Differential Equations. Communications of ACM. 101–110. Numerical Recipes -Art of Scientific Computing. 16) Amodio P. Damian V.com last accessed on 5/10/10.

Sign up to vote on this title
UsefulNot useful