2025 International Conference on Sensing, Measurement & Data Analytics in the era of Artificial Intelligence (ICSMD 2025)

In real-world scenarios, mechanical equipment operating under complex and dynamic conditions frequently encounters unexpected unknown fault types. These unseen fault categories continuously emerge as operational conditions evolve, rendering conventional transfer diagnosis methods relying on known fault types inadequate for such dynamic diagnostic challenges. To address this issue, this paper proposes a prototype-based continual domain adaptation network (PCDAN) model for incremental fault-type transfer diagnosis. First, a deep decoupled prototype network is designed to align known fault categories between the target domain and the source domain while enabling the identification and diagnosis of emerging unknown fault types under evolving operational conditions. Secondly, the model autonomously updates its diagnostic knowledge by synergizing prototype-based knowledge replay and parameter transfer mechanisms for newly encountered operational scenarios with emerging fault types, which can accomplish target domain fault diagnosis under incremental fault type scenarios. Finally, the effectiveness and practicality of the proposed method are verified through bearing-fault experiment on typical mechanical components, which demonstrates the significant diagnostic performance of the proposed model by comparing with other methods

incremental domain adaptation,continuously emerging unknown fault types,fault diagnosis,variable operating condition