Preparation of Nanoparticle-Enriched Fuels and Prediction of Cylinder Pressure Through Machine Learning Models

Precise estimation of cylinder pressure (CP) in internal combustion engines (ICEs) is critical for optimizing engine performance, increasing fuel efficiency, and controlling emissions. This estimation is essential in developing engine design and control strategies under various operating conditions and fuel formulations. Considering the time and cost challenges of traditional experimental data collection methods, machine learning (ML) models have great potential in predicting CPs in ICEs. This research aims to experimentally determine the CPs of different nanoparticle (NP)-enriched fuels in diesel engines and to estimate CPs employing ML models on the obtained data. The study is among the pioneering studies emphasizing the importance of early diagnosis and reducing experimental costs by estimating CPs in ICEs. The study prepared and tested nine different fuel formulations (biodiesel and NP-enriched derivatives). Four distinct ML models, specifically Decision Tree, Gradient Boosting, Random Forest, and Polynomial Regression (PR) were employed to predict CPs, with hyperparameters optimized for each model. Experimental findings show that all ML models exhibit high and competitive performances regarding the \( R^2 \) $R^2$ metric. Significantly, the PR model produced values with higher explanatory power than the others, providing a more reliable prediction performance. The performance of the used ML models was evaluated in detail by considering the regression error metrics. The results show that ML models can effectively predict fuel performance in ICEs and have the potential to detect future system anomalies in advance. The results have potential applications in fuel formulation development and real-time engine monitoring systems.