The plastic strain accumulation results of the multi-layered wrapped pressure vessel liner during long-term service are an important basis for its safety performance evaluation. However, the complex welds distributed on the liner bring challenges to the calculation of plastic cumulative strain. To this end, a novel hybrid deep learning framework is proposed for the efficient and precise prediction of ratcheting behavior in the liner welds of multilayered pressure vessels. By employing a BiLSTM network to extract bidirectional temporal dependencies from the strain history and incorporating a Multi-Head Attention (MHA) mechanism for adaptive feature weighting, the proposed method effectively addresses the difficulty of modeling cumulative effects in long-sequence ratcheting data. Firstly, asymmetric cyclic loading experiments were conducted on various welded joints and base metals to reveal the evolutionary laws of ratcheting behavior and construct a training dataset. The results show that the ratcheting strain evolution of different structural specimens shows the typical ‘two-stage’ characteristics, and the ratcheting strain accumulation on the base metal is significantly higher than that of the weld structure specimen. The proposed deep learning model can not only accurately capture the ‘two-stage’ evolution of ratcheting strain, but also directly use the base metal data to accurately predict the ratcheting strain accumulation in different weld parts, avoiding the complex parameter calibration process of the traditional constitutive model. It provides technical support for the integrity assessment and online monitoring of the complex weld structure of multi-layered pressure vessels.
Li et al. (Thu,) studied this question.