Multi-input multi-output machine learning predictive model for engine performance and stability, emissions, combustion and ignition characteristics of diesel-biodiesel-gasoline blends

In this research, a multi-input multi-output artificial neural network (MIMO-ANN) is developed, in which 14 features associated with the engine performance and stability, emissions, combustion and ignition characteristics of diesel-biodiesel-gasoline mixtures are meant to be modelled by a diverse combination of engine/combustion parameters. The selected targets comprise brake specific fuel consumption (BSFC), brake thermal efficiency (BTE), combustion efficiency, coefficient of variance (COV), NOx, CO2, CO and HC emissions, exhaust temperature (Texh), in-cylinder pressure (Pcyl), maximum pressure rise rate (MPRR), heat release rate (HRR), combustion duration (CD) and ignition delay (ID). The inputs variables entail the load, speed, compression ratio, gasoline, biodiesel and diesel ratios, crank angle (CA), injection temperature (Tinj), injection pressure (Pinj), brake mean effective pressure (BMEP) and start of injection (SOI). Sensitivity analysis and outlier detection are applied in order to eliminate less-effective inputs/data points. The prepared data sets are then used to train and test the ANN model, in conjunction with benchmarking the model outcomes using coefficient of determination (R2), average absolute relative deviation (AARD) and relative mean squared errors (RMSE). The R2 ranged within 0.9804-0.9998, which is close to unity, proving that the proposed network is accurately capable of predicting the intended combustion characteristics.

Automated summary

In this research, a MIMO-ANN is developed to model the characteristics of diesel-biodiesel-gasoline mixtures and predict engine performance and stability, emissions, combustion and ignition characteristics. By performing sensitivity analysis and outlier detection on the less-effective inputs/data points, accurate predictions are obtained.