Domain Generalized Open-Set Fault Detection and Diagnosis for Belt Conveyor Systems With Prototype Learning

Abstract

Belt conveyor systems are essential across various industries but are prone to faults due to their distinctive design and challenging operational environments. Various approaches have been explored for fault detection and diagnosis (FDD) in belt conveyor systems, but research specifically addressing the unique features of these systems and providing a comprehensive analysis remains limited. In this paper, we propose a deep learning-based two-stage FDD approach specifically tailored to address the challenges associated with belt conveyor systems. Challenges in applying deep learning-based FDD include managing the vast number of features collected from large-scale systems, the scarcity of diverse fault data, and distribution shifts between training and testing environments. To address these issues, we developed a FDD approach that integrates open-set recognition and domain generalization, leveraging multivariate sensor data. Our two-stage design effectively mitigates the limitations posed by insufficient faulty samples while enhancing overall FDD performance. To properly assess the proposed method, we utilized datasets collected from a testbed that simulates the actual operational conditions of belt conveyor systems. The evaluation results demonstrate that the proposed FDD method exhibits robust performance in real-world applications on belt conveyor systems.