Federated domain generalization for condition monitoring in ultrasonic metal welding

Abstract

Ultrasonic metal welding (UMW) is a key joining technology with widespread industrial applications. Condition monitoring (CM) capabilities are critically needed in UMW applications because process anomalies, such as tool degradation and workpiece surface contamination, significantly deteriorate the joining quality. Recently, machine learning models emerged as a promising tool for CM in many manufacturing applications. Yet, many existing models lack the generalizability or adaptability and cannot be directly applied to new manufacturing process configurations (i.e., domains). Although several domain generalization techniques have been proposed, their successful deployment often requires substantial training data, which can be expensive and time-consuming to collect in a single factory. Such issues may be potentially alleviated by pooling data across factories, but data sharing raises critical data privacy concerns that have prohibited data sharing for collaborative model training in the industry. To address these challenges, this paper presents a Federated Domain Generalization for Condition Monitoring (FDG-CM) framework that provides domain generalization capabilities in distributed learning while ensuring data privacy. By effectively learning a unified representation from the feature space, FDG-CM can adapt CM models for new clients (factories) with different process configurations. To demonstrate the effectiveness of FDG-CM, we investigate two distinct UMW CM tasks, including tool condition monitoring and workpiece surface condition classification. Compared with state-of-the-art federated learning algorithms, FDG-CM achieves a 5.35%–8.08% improvement in CM accuracy. FDG-CM is also shown to achieve excellent performance in challenging scenarios involving unbalanced data distributions and limited participating clients. Furthermore, by implementing the FDG-CM method on an edge–cloud architecture, we show that this method is both viable and efficient in practice. The FDG-CM framework is readily extensible to other manufacturing applications.