Machine learning-driven combustion pressure reconstruction and prediction for free-piston linear generators using kinematic parameters

The free-piston linear generator (FPLG) emerges as a groundbreaking linear power generation technology, offering superior energy conversion efficiency and promising sustainable energy solutions. Accurate reconstruction of in-cylinder pressure without dedicated sensors is crucial for detecting abnormal combustion and improving on-board tests. The interaction between the piston's linear motion and combustion dynamics allows for thermodynamic state estimation using kinematic parameters. This paper proposes a combustion pressure reconstruction method utilizing machine learning. By applying mode decomposition theory, the system's operating frequency characteristics are used to extract velocity. The piston's displacement and extracted velocity serve as inputs for precise in-cylinder pressure reconstruction. Validation results show outstanding performance, with 5-fold cross-validation yielding average MSE, MAE, and R2 values of 0.0581, 0.1210, and 0.9998, respectively. This approach can also enable accurate point prediction (average relative error <3 %) and reliable interval estimation of critical combustion parameters over continuous operating cycles. This breakthrough establishes a non-intrusive diagnostic framework for FPLG system, offering significant potential for real-time monitoring and control optimization.