基于CNN与BLS的滚动轴承故障诊断方法

Fault Diagnosis Method of Rolling Bearing Based on CNN and BLS

针对传统滚动轴承故障诊断方法训练时间长和效率低的问题,提出一种基于卷积神经网络(convolutional neural networks,简称CNN)和宽度学习系统(broad learning system,简称BLS)的故障诊断方法,实现了端到端的快速准确模式识别。首先,建立CNN与BLS结合的宽度卷积学习系统(broad convolutional learning system,简称BCLS),利用CNN提取信号特征和BLS进行分类,获得系统输出;其次,通过残差学习增加BLS层数,形成堆叠宽度卷积学习系统(stacked broad convolutional learning system,简称SBCLS),优化预测输出与真实标签的误差,对轴承故障模式进行识别;最后,通过试验将所提方法与3种BLS方法的预测结果进行了比较验证。结果表明,与几种常见故障诊断方法相比,所提方法诊断效果更佳,具有更高的准确率和训练效率,在边缘端的智能故障诊断中具有较好的应用前景。

To address the issue of long training time and low efficiency in traditional rolling bearing fault diagnosis methods,a fault diagnosis method based on convolutional neural networks(CNN)and broad learning system(BLS)is proposed to realize fast and accurate end-to-end pattern recognition.A broad convolutional learning system(BCLS)is established by combining CNN and BLS,using CNN to extract signal features and BLS for classification to generate system output.BLS layers are integrated through residual learning to form a stacked broad convolutional learning system(SBCLS),which optimize the error between predicted outputs and real labels,thereby recognizing bearing fault patterns.Control experiments are set up to verify the proposed method.A comparative test with three BLS methods indicate that the proposed method offers superior diagnostic performance.In addition,when compared to several common fault diagnosis methods,the proposed method demonstrates higher accuracy and training efficiency,showing promise for intelligent fault diagnosis at the edge.