An Artificial Neural Network for the Low-Cost Prediction of Soot Emissions

Abstract

Soot formation in combustion systems is a growing concern due to its adverse environmental and health effects. It is considered to be a tremendously complicated phenomenon which includes multiphase flow, thermodynamics, heat transfer, chemical kinetics, and particle dynamics. Although various numerical approaches have been developed for the detailed modeling of soot evolution, most industrial device simulations neglect or rudimentarily approximate soot formation due to its high computational cost. Developing accurate, easy to use, and computationally inexpensive numerical techniques to predict or estimate soot concentrations is a major objective of the combustion industry. In the present study, a supervised Artificial Neural Network (ANN) technique is applied to predict the soot concentration fields in ethylene/air laminar diffusion flames accurately with a low computational cost. To gather validated data, eight different flames with various equivalence ratios, inlet velocities, and burner geometries are modeled using the CoFlame code (a computational fluid dynamics (CFD) parallel combustion and soot model) and the Lagrangian histories of soot-containing fluid parcels are computed and stored. Then, an ANN model is developed and optimized using the Levenberg-Marquardt approach. Two different scenarios are introduced to validate the network performance; testing the prediction capabilities of the network for the same eight flames that are used to train the network, and for two new flames that are not within the training data set. It is shown that for both of these cases the ANN is able to predict the overall soot concentration field very well with a relatively low integrated error.

An Artificial Neural Network for the Prediction of Penetration Height of Aerated Elliptical Liquid Jets in Gaseous Crossflows

“…Each section is usually referred to as a layer. In general, each network has one input, one output and can have multiple hidden layers depending on the desired accuracy and available computational power [13], [18]. In each layer, several nodes, or neurons (shown as circles in Figure 4) are included.…”

Exploring soot inception rate with stochastic modelling and machine learning

“…A number of studies have successfully applied ML algorithms to predict soot output in engines and burners [29,30,31], to assist with experimental soot measurement procedures [32,33], and to classify soot in TEM images [34]. A pair of recent studies by Dworkin and co-workers [35,36] used artificial neural networks (ANN), a class of ML algorithms, to predict the soot volume fraction (f v ) in laminar coflow diffusion flames. In [35], the authors paired the Lagrangian histories of key combustion variables (temperature, mixture fraction, and the concentrations of O 2 , CO, CO 2 , H 2 , OH, and C 2 H 2 ), the inputs, with f v , the target, in a multilayer perceptron ANN.…”

“…A pair of recent studies by Dworkin and co-workers [35,36] used artificial neural networks (ANN), a class of ML algorithms, to predict the soot volume fraction (f v ) in laminar coflow diffusion flames. In [35], the authors paired the Lagrangian histories of key combustion variables (temperature, mixture fraction, and the concentrations of O 2 , CO, CO 2 , H 2 , OH, and C 2 H 2 ), the inputs, with f v , the target, in a multilayer perceptron ANN. The ANN, trained with CFD data from the combustion simulation code CoFlame [25], was shown to predict the peak f v in the flame to within 20% of the simulated values and the spatially-integrated f v to within 5%.…”

A Long Short-Term Memory Neural Network for the Low-Cost Prediction of Soot Concentration in a Time-Dependent Flame

“…The cumbersome nature of high-fidelity models has pushed many industries to use semi-empirical soot models which, although faster, may sacrifice a good deal of accuracy compared to their high-fidelity counterparts (relative errors of 1-2 orders of magnitude in soot volume fraction prediction [10][11][12][13]). To circumvent the clash between accuracy and cost in traditional soot modelling methods, a novel computational concept called the "soot estimator" has been recently introduced [14][15][16][17]. In the soot estimator concept, only a validated combustion model is needed, and the soot characteristics are estimated without solving the soot-related terms and equations.…”

“…would attempt to relate the soot volume fraction in the flame to selected gas parameters, namely: the histories of temperature, mixture fraction, and H 2 concentration, respectively. The relationship between the soot characteristics and the gas-related parameters (i.e., the function f in Equation ( 3)) is central to the soot estimator concept and was a major focus in previous developments [14][15][16][17].…”