ML-Assisted Prediction of In-Cylinder Pressures of Spark Ignition Engines

In-cylinder pressure is a critical parameter for assessing the combustion process and per-formance of spark ignition internal combustion engines. However, obtaining measured in-cylinder pressure data across the full operating range, particularly at various spark advance angles (SA), is costly and technically demanding, limiting its widespread application in engineering practice. This study proposes two artificial neural network (ANN) based in-cylinder pressure reconstruction methods utilizing in-cylinder pressure and heat release rate data from a three-cylinder motorcycle gasoline engine under varying spark advance angle conditions, aiming to achieve cost-effective, high-precision pressure prediction through machine learning technology. Both pressure reconstruction methods employ crank angle and spark advance angle as input features. Method 1 (ANN-P) directly predicts the in-cylinder pressure curve, achieving a coefficient of determination R² exceeding 0.99 on both training and validation sets with a root mean square error (RMSE) below 0.13 bar, accurately reproducing the pressure evolution throughout the compression, combustion, and expansion processes while achieving high-precision prediction of indicated mean effective pressure (IMEP). Method 2 (ANN-HRR) adopts an indirect strategy of "predicting heat release rate followed by integration to reconstruct pressure." This method first derives the apparent heat release rate (HRR) from measured in-cylinder pressure data, trains a machine learning algorithm with HRR data to predict the target operating condition's HRR curve, then reconstructs in-cylinder pressure through integral inversion based on a single-zone thermodynamic model, thereby avoiding error amplification caused by differential operations. This approach demonstrates superior performance in predicting combustion characteristic points (CA10, CA50). The results demonstrate that both methods accurately capture the influence of spark timing on combustion phasing and peak pressure. Method 1 achieves high accuracy in predicted pressure curves but exhibits lower accuracy in capturing combustion characteristics; Method 2 effectively compensates for the limitations of Method 1 in characterizing combustion features through heat release rate curve prediction. This study provides an economical and efficient technical approach for gasoline engine combustion diagnosis, performance calibration, and control optimization.