Most structural damage occurs in welds, and transient stress analysis corresponding to the welding procedure is performed to investigate the cause. To accurately reproduce the stress generated in the welds, it is necessary to accurately reproduce the transient heat input in the welding procedure. Thus, traditionally, inverse analysis on the transient heat input is conducted by focusing on the molten shape of the actual components; i.e., try and error work performing transient temperature distribution analysis assuming heat input until the molten shape is reproduced. Goldak's double-ellipsoidal heat source model was used by most of the researchers. In this case, heat source parameters from experience are assumed, and transient temperature distribution analysis is repeated by changing the parameters until the obtained region above the melting point matches the target molten shape. Recently, machine learning methods have been become effective to improve inverse analysis in structural integrity issues to straight-forward analysis. In this work, a model using the deep learning has been developed that can directly determine the parameters of the heat source model from the welding records and the molten shape, which can omit the inverse analysis of the heat source parameters, which has been a bottleneck in the past. This deep learning model can reduce the time required determining parameters used in FEA for failure analysis of welded components.
Read full abstract