Bpnn Model Based Ai for the Estimation of Soot Data from Flame Spectral Emissions in H2/N2 Diluted Ethylene Laminar Diffusion  Flames

Abstract

Advanced combustion devices and alternative fuels like hydrogen require a estimation of impact of such changes on regulated emissions like soot. Huge data is then required to be collected and post-processed both in academic and industrial setups. To reduce cost and time, a Back Propagation Neural Network (BPNN) model has been validated and optimized for the simultaneous prediction of soot volume fraction, temperature and particle sizes from flame luminosity measurements. Thecompatibility, robustness and reliability of neural network models to fuel composition and flow rate variations was analyzed by both varying fuel flow-rate and fuel stream dilution in axis-symmetric diffusion flame burning in a coflow of air. BPNN model predicted results were contrasted against those obtained through image processing and deconvolution technique. A detailed analysis based on an extensive experimental study of the performance of BPNN model revealed indifference to N2/H2 dilution and fuel flow-rate variations. Furthermore, the original BPNN model is optimized  to improve learning rate and reduce computation cost, by applying a Bayesian algorithm which continuously monitors and minimizes an error function. It was concluded that a training set size of three was sufficient to predict soot field parameters with a fair prediction accuracy. These results have direct implication for the application of proposed technique in practical combustion equipment where operating  parameter, e.g. Air-fuel ratio and EGR, show both gradual and instantaneous  variations, affecting performance and emissions. These results clearly demonstrate the robustness of this technique which offers opportunities for real-time, in situ, monitoring of practical combustion environments. Real-time data of sooting characteristics of flames can, therefore, be used toboth monitor and control emissions by integrating a response strategy in the system control.

Keywords: Soot Spectral Emissions, Diffusion Flames, Neural Networks, Machine learning, Bayesian Optimization