Machine learning–based method for analyzing stress distribution in a ship

Ship structural health monitoring (SHM) systems are essential for ensuring operational safety. Monitoring stress distribution is a critical function of such systems. Analysis of stress data collected by pressure sensors can enhance understanding of SHM performance and facilitate the prevention of structural failures. In this study, we propose a machine learning-based computational method to analyze the stress distribution of a crane ship. The method first employed extreme gradient boosting (XGBoost) to evaluate relationships among stress data from multiple pressure sensors, and then used multilayer perceptron (MLP) to construct regression models. For each pressure sensor, an optimal MLP regression model was established to infer its stress values from those of related sensors. These models demonstrated strong fitting performance on both training and independent test datasets. Furthermore, the method identified key pressure sensors whose measurements were particularly important for recovering stresses at most monitoring points. These findings improve interpretability of the method and provide insights into the mechanisms of the monitoring system. The robustness and rationality of the approach were further examined through ablation tests, confirming its effectiveness for ship SHM applications.