ANN metamodel assisted Particle Swarm Optimization of the performance-emission trade-off characteristics of a single cylinder CRDI engine under CNG dual-fuel operation

The encouraging potential of a CRDI coupled CNG application in providing superior PM-NO_x-BSFC trade-off footprint as compared to the BDO of a single cylinder diesel engine, provided a strong motivation to undertake an in-depth exploration of the parametric design space to investigate the possibility of obtaining superior emission-performance trade-off characteristics as compared to the experimental endeavour. To obviate the infeasibility of a full factorial based experimental approach to the intended exploration, an offline metamodel assisted optimization study within the same experimental range of the input/control variables was considered as a viable strategy. FIP, CES and engine load were chosen as the control variables for the optimization study. An Adaptive Merit Function (AMF) was constituted as the objective function to be optimized encapsulating the simultaneous necessity to meet the EPA Tier 4 PM-NHC emission mandates and lower BSFC expectations of the consumer. An ANN based metamodel was developed to correlate the objective function with the chosen control variables while Latin-Hypercube was chosen as the DoE scheduler to provide the initial population sample for the optimization iteration sequence. PSO was chosen as the optimization algorithm due to its inherent simplicity and efficiency in yielding an ensured convergence of the objective function at significantly less computational cost. Validation of the obtained optimal solutions at the experimental load steps of 25%, 50%, 75% and full load yielded 9.9%,16.4%, 10.5%, 2.7% higher AMF than that yielded by the best CRDI-CNG operation at the corresponding load steps respectively during actual experimentation. The inherent design of the AMF objective function ensured that all such optimal trade-off values simultaneously honoured the expectations of the EPA Tier 4 PM and NHC mandates.