Enhancing Data-driven Vibration-based Machinery Fault Diagnosis Generalization Under Varied Conditions by Removing Domain-Specific Information Utilizing Sparse Representation

This paper introduces a novel approach to machinery fault diagnosis, addressing the challenge of domain generalization in diverse industrial environments. Traditional methods often struggle with domain shift and the scarcity of balanced, la- beled datasets, limiting their effectiveness across varied oper- ational conditions. The proposed method leverages the abun- dance of healthy machinery signals as a reference for extract- ing domain-specific information. By doing so, it removes the domain-related variances from the observation signals, focus- ing on the intrinsic characteristics of faults. The methodol- ogy is validated with a case study, demonstrating enhanced diagnosis accuracy and generalization capabilities in unseen domains. This research contributes to the field of vibration- based intelligent fault diagnosis by providing a robust solu- tion to a long-standing problem in machine condition moni- toring.