Fatigue life prediction of aluminum-steel magnetic pulse crimped joints based on point cloud measurement and gradient boosting regression trees

The fatigue life of magnetic pulse crimping (MPC) joints is crucial for the safe fatigue design of connection structures. Traditional fatigue life prediction methods primarily rely on loading condition analysis and fail to fully account for the impact of manufacturing variations (such as raw material dimensions, process parameters, and joint deformations), which presents challenges for accurate fatigue life prediction. To address this issue, this paper proposes a fatigue life prediction method for MPC joints that combines point cloud measurement and machine learning (ML) models. Random sample consensus (RANSAC) and point cloud segmentation are used to extract the joint deformation contour precisely. Compared to metallographic analysis, this method achieved non-destructive extraction of joint deformation features. Based on this, an integrated dataset covering the entire process from raw materials to fatigue testing is established. Five machine learning models are trained and tested, with results showing that the gradient boosting regression trees (GBRT) model performs the best. The visualization of a single decision tree in the GBRT model is analyzed, providing a transparent decision-making process. A comparison is made between the GBRT model and the traditional Basquin model. In the GBRT model, 100% of the training set and 90% of the testing set fall within the 1.5 times error band, while only 45% of the training set and 60% of the testing set in the Basquin model fall within this range. Additionally, the GBRT model achieves a higher coefficient of determination (R²) on the dataset compared to the Basquin model.