Fault Diagnosis of Motors Under Varying Operating Conditions With Only Phase Current and Limited Samples

Fault diagnosis of brushless direct current motor (BLDCM) is of great importance for ensuring the safety of equipment. However, it is challenging to identify the multiple fault types with only phase current as a monitoring signal, especially when facing varying operating conditions and limited samples. To address the above challenges, this article proposes a multimodal harmonic-spatiotemporal meta-learning method to carry out the BLDCM fault diagnosis. Tacholess order tracking harmonic extraction (TOTHE) and Gramian angular field (GAF) are introduced as feature extraction strategies to extract 1-D harmonic components and 2-D time-evolution information from the phase current, respectively. These improved features are further fused by a dual-branch parallel structure. In addition, a model-agnostic meta-learning (MAML) algorithm is developed to guide the learning and updating process of the model in multiple source domain tasks. This well-trained model just needs to be fine-tuned with extremely limited samples of an unseen target domain task, which can further achieve the high fault diagnosis accuracy. Two types of BLDCMs with five health states under varying working conditions are applied to validate the feature extraction, fast convergence, and domain adaptation ability of the proposed method.