Feature Augmented Label-Distribution-Aware Margin Loss for Fault Diagnosis Under Class Imbalance
Peng Han, Baoping Tang, Chongyang Zhang, Peng Yi Zhu
Chongqing University
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
Re-weighting methods can effectively reduce the diagnosis bias of deep learning models under class-imbalanced data. However, data distribution information is neglected in most reweighted fault diagnosis methods, leading to the inadequate representation ability of deep learning models. To address this problem, a novel loss function called Feature Augmented Label-Distribution-Aware Margin Loss (FALDAMLoss) is proposed. At the feature level, FALDAMLoss algorithm implicitly augments the feature set, which enables the model to learn more features from training samples. And at the decision level, further corrects the biased decision boundary to enhance the decision ability of the model, which promotes the model to adapt to the imbalanced distribution of samples. The deferred re-weighting (DRW) is implemented to enable the model to learn better generalizable representations. The experiments verify that the proposed FALDAMLoss algorithm is more effective than other advanced loss functions.
逐年被引趋势
暂无年度引用数据
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
工程Machine Fault Diagnosis Techniques
Imbalanced Data Classification Techniques · Fault Detection and Control Systems