Experimental analysis, performance prediction, and multi-objective optimization of a heavy-duty diesel engine using RSM-ANN

Abstract Heavy-duty diesel engines are essential for transportation and industrial applications but face challenges related to high fuel consumption and carbon emissions. This study aims to develop a hybrid optimization framework combining Response Surface Methodology (RSM), Artificial Neural Networks (ANN), and Multi-Objective Genetic Algorithm (MOGA) to enhance engine efficiency and sustainability. Experimental tests were conducted on a 7.7 L heavy-duty diesel engine operating with commercial diesel with at speeds ranging from 800 to 2400 rpm. Three input parameters—engine speed, fuel mass flow, and air mass flow—were examined with respect to Brake Thermal Efficiency (BTE), Brake Specific Fuel Consumption (BSFC), and CO 2 emissions. Predictive models were established using RSM and ANN, followed by multi-objective optimization via MOGA. The RSM models exhibited strong predictive performance with R 2 values of 0.95 (BTE), 0.98 (BSFC), and 0.99 (CO 2 ). ANN models achieved regression coefficients of R = 0.99 (training), 0.97 (validation), and 0.98 (testing). Optimization results indicated the most sustainable operating region at mid-speeds (1200–1400 rpm), achieving a maximum BTE of 42.5%, minimum BSFC of 0.187 kg kWh −1 , and CO 2 emissions between 0.0128–0.0185 kg s −1 . The ANN–MOGA Pareto front revealed a narrow optimization surface dominated by fuel consumption, whereas the RSM–MOGA approach exhibited broader trade-offs reflecting variable interactions. The findings show that the RSM, ANN, and MOGA framework helps analyze efficiency and emission trade-offs, providing a valuable tool for advancing sustainable diesel engine operation.