Zero-shot learning for compound fault diagnosis of bearings

Due to the concurrency and coupling of various types of faults, and the number of possible fault modes grows exponentially, thereby compound fault diagnosis is a difficult problem in bearing fault diagnosis. The existing deep learning models can extract fault features when there are a large number of labeled compound fault samples. In industrial scenarios, collecting and labeling sufficient compound fault samples are unpractical. Using the model trained on single fault samples to identify unknown compound faults is challenging and innovative. To address this problem, we propose a Zero-shot Learning Compound Fault Diagnosis Model of bearings (ZLCFDM). We design an encoding method to express the semantics of single faults and compound faults according to the fault characteristics. A convolutional neural network is developed to extract the time–frequency features of the compound fault signal. Then we embed the semantic feature of the fault into the visual space of the fault data. The Euclidean distance is used to measure the distance between the signal features and the semantic features of the compound faults to identify the categories of unknown compound faults. To validate the proposed method, we conduct experiments on a self-built testbed. The results demonstrate that the accuracy of identifying compound fault reached 77.73% when the model was trained without any compound fault samples.