华中科技大学学报（自然科学版）2025，Vol.53Issue(4)：38-44,84,8.DOI:10.13245/j.hust.250274
基于Proto-DANN的电机变工况迁移诊断方法
Motor transfer diagnosis method for varying operating conditions based on Proto-DANN
摘要
Abstract
Current transfer diagnosis methods often rely on diverse manually labeled data but struggle to adapt transfer strategies to varying motor operating conditions,and this limitation reduces their ability to extract domain-invariant fault features effectively when labeled data is limited,leading to diminished diagnostic performance under changing conditions.To address these challenges,a Prototype Similarity Domain Adaptation Neural Network(Proto-DANN)was proposed.In this approach,labeled data from a specific condition served as the source domain,while unlabeled data from other conditions acted as the target domain.The Prototype Network aligned feature distributions between domains using distance-based similarity metrics.The meta-training process included internal supervised training of the source network and external unsupervised training of the target network,facilitated by a virtual label backpropagation algorithm.Through alternating internal and external training,Proto-DANN minimized feature distribution discrepancies,enabling accurate identification of motor faults under various operating conditions.The results show that the proposed method achieves outstanding diagnostic performance and exhibits robust generalization capabilities,accurately detecting unlabeled faults in motors under various operating conditions.关键词
异步电机/变工况/原型相似/无标签故障/域自适应/迁移诊断Key words
asynchronous motor/variable operating conditions/prototype similarity/unlabeled fault/domain adaptation/transfer diagnosis分类
动力与电气工程引用本文复制引用
姜苗,向阳,盛晨兴..基于Proto-DANN的电机变工况迁移诊断方法[J].华中科技大学学报（自然科学版）,2025,53(4):38-44,84,8.基金项目
国家自然科学基金资助项目(52241102) （52241102）
国家工信部专项资助项目(20201g0079). （20201g0079）
