Stress distribution characteristics and intelligent online monitoring methods for multilayer wound pressure vessel

Abstract

A digital twin-driven online stress prediction method is proposed to address the stress monitoring requirements for multi-layer wrapped high-pressure hydrogen storage vessels. This method establishes a phased computational framework (offline/online): During the offline phase, the global stress field is computed using the Finite Element Method (FEM), and a random forest hybrid regression prediction model incorporating the whale optimization algorithm (WOA-RF) is trained to establish the mapping relationship between container load, structural features, node coordinates, and stress. During the online phase, the deviation between measured local stresses and offline-predicted stresses is first calculated. Subsequently, a K-Nearest Neighbors (KNN) algorithm constructs a surrogate model linking load-node coordinates to stress deviation. Ultimately, the KNN model is driven by locally acquired real-time measurement data, utilizing its output stress deviation to perform real-time corrections on WOA-RF prediction results, thereby achieving global twin stress prediction for the monitored vessel. To establish a more accurate finite element model during the offline phase, this paper innovatively derives a method for inverting interlayer preload in multilayer vessels based on measured data. Verification conducted on a multi-layer wrapped high-pressure reactor demonstrated that the proposed stress monitoring method achieved prediction errors ranging from 0.4 % to 10 %. Furthermore, the findings elucidate the random and non-uniform stress distribution characteristics exhibited by multi-layer wrapped vessels under loading, which stem from the complex interlayer preload generated during the manufacturing process.