Artificial neural network modeling of heavy-duty diesel engine performance and emissions incorporating fuel injection strategy for ECU calibration applications

The increasing complexity of modern heavy-duty (HD) diesel engines and stringent emissions regulations have created a critical need for fast and accurate predictive tools capable of supporting engine calibration and emissions control strategies. This study presents data-driven models based on artificial neural networks (ANN) for predicting steady-state engine performance and engine-out emissions from a HD diesel engine using experimentally measured data. The dataset obtained from a 2021 Navistar E39 engine, covering a wide range of operating conditions, was utilized for model development and validation. Two ANN-based models were developed: a baseline model using conventional engine operating parameters and a hybrid model incorporating detailed fuel injection parameters, including injection timing and quantity of multiple injection pulses. The hybrid ANN model enables explicit representation of electronic control unit (ECU) calibration. Both models were trained, validated, and tested using structured datasets, achieving high predictive accuracy across all outputs. The baseline ANN model demonstrated coefficient of determination (R 2 ) values exceeding 0.93 for all predictions, while the hybrid model achieved R 2 values above 0.91, with engine-out NO x prediction reaching 0.96. Independent evaluation using modified injection timing conditions that were excluded from the training showed that the hybrid model reproduced the measured engine-out NO x , with absolute errors below 4 %, 7 %, and 2.5 % for the pilot, main, and post injection timing variations, respectively. The developed calibration-aware ANN model provides a fast and reliable tool for predicting engine performance and engine-out emissions under varying calibration strategies, offering significant potential to reduce experimental effort in engine calibration and optimization.