Precision Fatigue Life Prediction for Cast Iron Engine Cylinder Blocks Using Machine Learning

Accurate assessment of engine block fatigue life ( N f ) is critical for enhancing transportation system reliability, particularly through precise modeling of the fatigue crack propagation (FCP) rate (d a /d N ) in the upper main bearing wall. Existing d a /d N models, however, exhibit limited N f prediction accuracy due to neglecting structural and loading effects. This study first employs a component equivalence method to experimentally derive true d a /d N curves for diverse upper main bearing wall configurations, with standard specimens as controls. Three machine learning (ML) models are then trained on these datasets, with the random forest algorithm selected for its superior accuracy to establish the final d a /d N model. Incorporating crack closure effects via the virtual crack closure technique, simulated N f predictions are generated. Fatigue tests on engine blocks validate the model's predictions, demonstrating a 13.8% improvement in accuracy over conventional approaches, attributable to the explicit integration of structural and loading parameters. The research method presented in this paper provides a more effective tool for the accurate establishment of the d a /d N model for actual engineering components and the precise prediction of fatigue life, thereby laying a solid foundation for improving component reliability.