Detecting deformation mechanisms of metals from acoustic emission signals through knowledge-driven unsupervised learning

Timely detection of deformation mechanisms in metallic structural materials is essential for early-warning alerts on potential damages and fractures. Acoustic emission (AE) technologies are commonly used for this purpose due to their non-destructive nature. However, traditional methods often struggle with distinguishing AE signals associated with multiple co-existing deformation mechanisms. To address this challenge, we propose a knowledge-driven unsupervised learning approach. The novel method leverages a family of gradient-driven supervised base learners and integrates them with a knowledge-infused aggregate loss function, effectively transforming the approach into an unsupervised learning framework. Compared to existing methods, our approach excels in identifying co-existing deformation mechanisms associated with AE signals. Experiments on porous 316L stainless steel during tensile process show that the avalanche statistics of the identified dislocation and crack AE signals align closely with classical statistical methods and fracture theory. By integrating with the avalanche theory, our proposed approach can continuously monitor material deformation mechanisms in real-time and provide dynamic early failure warnings. Additionally, the framework demonstrates strong transferability in recognizing multiple co-existing deformation mechanisms in new materials, leveraging its unsupervised learning capability.