Intelligent Rolling Bearing Fault Diagnosis Based on Vibration Signal Transformer Neural Network
作者单位：Department of Automation China University of Petroleum Beijing
会议名称：《第35届中国过程控制会议》
会议日期：1000年
学科分类：12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 08[工学] 081104[工学-模式识别与智能系统] 080203[工学-机械设计及理论] 0802[工学-机械工程] 0835[工学-软件工程] 0811[工学-控制科学与工程] 080201[工学-机械制造及其自动化] 0812[工学-计算机科学与技术（可授工学、理学学位）]
关 键 词：Fault Diagnosis Rolling Bearing Transformer Deep Learning
摘 要：Rolling bearing is one of the most important components of the majority of electrical and power drives. It is critical to accurately diagnose bearing faults to maintain the safe and smooth operation of electrical and power drives. In the past few years,deep-learning-based techniques have gained traction in bearing fault diagnosis, where they have attained competitive performance due to their superior feature extraction capabilities. However, when it comes to extracting features of long-term dependencies,the majority of them are inefficient. To address this issue, we propose a novel method for rolling bearing fault diagnosis called Vibration Signal Transformer Neural Network(VSTNN), which inherits the advantage of the Transformer s encoder with multihead self-attention and residual connection to aggregate vibration information from different time-series segment tokens in feature space, thereby improving long-term dependency modeling performance. Distinguishing from the vanilla Transformer which divides sentence sequences into tokens made up of words, we develop a novel vibration signal tokenization strategy that can yield token embedding sequences by combining multiple subsequence features learned from 1D vibration data, class tokens, and position embeddings. Furthermore, since bearing fault diagnosis is a sequence-to-category problem with only the fault category as an output, VSTNN bypasses the vanilla Transformer s decoder block in favor of a fully connected neural network as the fault classifier. This improves computational efficiency and lowers memory usage, allowing the model to more adapt for rolling bearing fault diagnosis. Extensive experiments are carried out on the bearing dataset collected from Case Western Reserve University to validate the effectiveness of the proposed method. The results reveal that VSTNN outperforms CNN and RNN models in terms of diagnostic performance.