Acoustic Emission-Based Cross-Domain Process Health Monitoring for Additive Manufacturing

The process health monitoring of wire arc additive manufacturing (WAAM) is significant for product quality. Most existing additive manufacturing process monitoring is based on image data, such as temperature and spatters. However, these monitoring methods do not reflect status information promptly. Moreover, the issue of limited cross-domain diagnosis generalization ability is faced by traditional neural networks for health state discrimination. To address these issues, this work puts forward a bi-classifier and orthogonal constraints jointly guided domain adaptation method based on acoustic emission (AE) signal for WAAM health monitoring. Specifically, we first build a min-max optimization strategy using bi-classifier discrepancy loss to achieve feature adaptation of different domains. Meanwhile, the orthogonal loss increases the dispersion of interclass features and the aggregation of intraclass features. Moreover, a nonlocal module is attached for obtaining the remote dependence relationship between the sample pixels of AE signal. Finally, based on the AE signals from the WAAM process, the performance of the method is evaluated, and the comprehensive results prove its effectiveness and superiority.