基于WDCNN-DLSTM的滚动轴承故障诊断方法

滚动轴承是机械设备中的核心部件，其运行状态对设备的运转有重要影响。深度学习作为滚动轴承故障诊断的重要方法越来越受到重视。由于传统的故障诊断方法没有充分利用数据时序性，提出了一种将第一层为宽卷积核的深度卷积神经网络(deep convolutional neural networks with wide first-layer kernels, WDCNN)和深度长短时记忆网络(deep long short-term memory networks, DLSTM)相融合的模型(WDCNN-DLSTM)。WDCNN将传统的CNN第一层卷积核尺寸加宽，提高了模型对一维振动信号中的空间特征信息的提取能力；DLSTM将多个LSTM模块进行堆叠，提高了模型对一维振动信号中时序信息的提取能力。WDCNN-DLSTM将二者通过连接层融合，优势互补，提高了模型的判别能力。通过实验结果表明，相较于一些其他模型，所提出的方法具有更高的精确度。在变负载的情况下，也仍然实现了更好的分类效果。

Fault Diagnosis Method for Rolling Bearings Based on WDCNN-DLSTM

Rolling bearing is the core component of mechanical equipment, whose operating state has an important impact on the operation of the equipment. As a crucial method for fault diagnosis of rolling bearings, deep learning has been paid more and more attention. Since the time sequence of data has not been fully used by the traditional fault diagnosis methods, a model (WDCNN-DLSTM) is proposed by combining the deep convolutional neural networks with wide first-layer kernels (WDCNN) with deep long short-term memory networks (DLSTM). The size of the first-layer kernel of the traditional CNN is widened in WDCNN, which strengthens the model''s ability to extract spatial feature information from one-dimensional vibration signals. Multiple LSTM modules are stacked in DLSTM to enhance the model''s ability of extracting time sequence information from one-dimensional vibration signals. WDCNN-DLSTM is generated by combing the two networks through the connection layer to complement their advantages, and as such the discriminant ability of the model is improved. The experimental results show that the proposed method obtains higher accuracy compared with some other models. In the case of variable loads, better classification effect is still achieved.