Neural Network-Based Diesel Engine Emissions Prediction for Variable Injection Timing, Injection Pressure, Compression Ratio and Load Conditions

The present study investigates the use of artificial neural network modelling for prediction of emission parameters of a four stroke single cylinder variable compression ratio diesel engine. ANN model was developed to predict emissions namely CO, NOX and HC. Emission data was collected by conducting experiments by varying compression ratio, Injection time, and injection pressure in four steps and load in five steps. Two training algorithms traingd and trainlm with hidden nodes varying from 3 to 20 in step of one were developed and trained. Best network from 36 networks was selected based on MSE, regression coefficients for training, validation, testing and correlation coefficient for prediction of unseen data. The best model was found to be Levenberg-Marquardt algorithm with 17 neurons and regression coefficients for training, validation and testing are 0.99628, 0.99561, 0.99472 and 0.99577 respectively. The correlation coefficient R for training data is 0.99643 and for unseen data is 0.99322. The regression coefficients for prediction of training sets of CO, NOX and HC are 0.99643, 0.99486 and 0.99601 respectively. The average % error for prediction of CO, NOX and HC are -0.16178, -0.38814 and 0.7459 respectively which are less than 1. It is found that artificial neural networks serve as an excellent tool for prediction of emissions from diesel engine under variable operating and design parameters.