摘要详情

ID / 提交时间

202 / 2025-06-14 00:06:44

标题

Deep Reinforcement Learning-Guided Adaptive Feature Extraction with Transformer Classification for Variable Speed Bearing Fault Diagnosis

关键字

Bearing fault diagnosis, variable speed machinery, Vold-Kalman filter, AMFEN, Self-Attention mechanism, Transformer

主题及专题

4、System Monitoring, Fault Diagnosis and Prediction

状态

全文待审

作者

Liren Pan / Zhenjiang Technician College Jiangsu Province

Leng Xue / Zhenjiang Technician College Jiangsu Province

Jialin Li / Zhenjiang Technician College Jiangsu Province

Di Zheng / Zhenjiang Technician College Jiangsu Province

Weilun Guo / Zhenjiang Technician College Jiangsu Province

Yongwei Su / Zhenjiang Technician College Jiangsu Province

摘要

The inherent non-stationarity of vibration signals from variable-speed rotating machinery poses significant challenges for traditional bearing fault diagnosis methods. This paper proposes a novel framework that synergistically integrates Deep Deterministic Policy Gradient (DDPG) optimization with an Adaptive Multi-scale Feature Extraction Network (AMFEN), self-attention mechanism and Transformer-based classification to address these challenges. The proposed methodology employs DDPG to adaptively optimize Vold-Kalman filter parameters for robust signal decomposition, eliminating the need for manual parameter tuning. Subsequently, 5 parallel AMFEN strategies extract complementary features from multiple domains, capturing comprehensive fault characteristics. A self-attention mechanism then intelligently selects the most discriminative features, reducing the dimensionality from 320 to 48 while preserving diagnostic information. Finally, a Transformer model performs fault classification, leveraging its superior capability in capturing feature dependencies and temporal patterns. Experimental validation on BGCF dataest demonstrates the framework's superior performance, achieving 97% diagnostic accuracy under variable speed conditions.