南京师范大学学报（工程技术版）2025，Vol.25Issue(2)：28-42,15.DOI:10.3969/j.issn.1672-1292.2025.02.003
基于语义增强型深度自编码器的零样本故障诊断方法
Semantic-Enhanced Deep Autoencoder for Zero-Shot Fault Diagnosis
摘要
Abstract
In recent years,zero-shot industrial process fault diagnosis methods have garnered attention.Addressing the poor performance of traditional embedded models in zero-shot fault diagnosis,a zero-shot fault diagnosis algorithm based on Semantic-Enhanced Deep Autoencoder(SEDAE)is proposed.This method introduces a projection domain shift elimination technique based on triplet loss constraints in the semantic attribute space,to enhance the model's knowledge transfer capability.The triplet loss constraint is integrated into the deep autoencoder to automatically extract semantic features of seen classes,thus achieving optimal mapping between the semantic knowledge of seen classes and prior semantic information.In this context,the nearest neighbor method is used in the semantic space to determine the category of new samples.Simulation results on the Tennessee-Eastman process(TEP)indicate that the proposed zero-shot fault diagnosis method improves accuracy by 7.11%compared to traditional methods and also achieves satisfactory results in generalized zero-shot fault diagnosis.关键词
故障诊断/零样本学习/语义增强型深度自编码器/三元组损失/投影域偏移Key words
fault diagnosis/zero-shot learning/semantic-enhanced autoencoder/triplet loss/projection domain shift分类
信息技术与安全科学引用本文复制引用
李烁琛,任世锦,魏明生,郝国生..基于语义增强型深度自编码器的零样本故障诊断方法[J].南京师范大学学报（工程技术版）,2025,25(2):28-42,15.基金项目
国家自然科学基金项目(61673196),徐州市科技计划项目(KC23239). （61673196）
