An Attribute Description Transfer Based Model for Zero-Shot Intelligent Diagnosis of High-Voltage Circuit Breakers

In this article, a novel attribute description transfer based model for zero-shot intelligent diagnosis (ADT-ZSID) is proposed to address the challenges in diagnosing high-voltage circuit breakers (HVCBs), particularly the common incidence of new, unknown faults in practice. ADT-ZSID is designed to perform fault diagnosis without the need for training on unknown fault samples. First, universal fault attribute descriptions are defined by analyzing multisource signal features, considering the specific traits and fault patterns of HVCBs. Next, attribute classifiers are constructed using only easily accessible fault data. These classifiers implement cross-domain mapping between the feature space and the label space, establishing relationships among different fault patterns. Finally, the attribute vector of target faults is predicted by the classifiers, coupled with similarity evaluation, to achieve the zero-shot fault diagnosis task. Experimental results demonstrate the effectiveness of ADT-ZSID in diagnosing target faults without requiring their training samples.