摘要
针对压力机轴承故障类型多、故障特征难以提取等问题,提出了一种基于堆叠去噪自编码网络(SDAE)与支持向量机(SVM)的压力机轴承故障诊断方法。该方法首先通过堆叠多层去噪自编码器构成深度网络模型,用无监督的方法训练每一层网络并在原始数据中加入噪声以得到更加稳健的特征表达,然后通过反向传播(BP)神经网络进行有监督的学习,对整个网络参数进行微调优化得到更接近样本分布的网络模型,最后利用SVM分类器对故障进行识别分类。实验中所提出的基于SDAE-SVM的故障诊断方法的故障识别准确率高达98%,表明该方法能有效提高压力机轴承的故障识别精度。
Various types of faults,difficult to extract features of the faults are the common problems for the bearing of press machine. In this work,a method of fault diagnosis for such bearings has been developed based on stacked denoising auto-encoder networks( SDAE) and support vector machine( SVM). In this method,a multi-layer denoising auto-encoder( SDAE) is used to construct a deep network model. Then,each layer of network is trained using unsupervised method,and mixed with noise to obtain a robust feature representation. Supervised learning using back propagation( BP) neural network to optimize the whole network parameters,thus obtaining a new identical network model with similar sample distribution. Finally,SVM classifier assists the model to identify the fault classification. The experimental results show that the proposed SDAE-SVM-based method can effectively improve the accuracy of recognizing the fault of the press bearings,with the value up to 98%.
基金
国家自然科学基金资助项目(71171154)
关键词
轴承
堆叠去噪自编码网络
支持向量机
故障诊断
bearing
stacked denoising auto-encoder networks
support vector machine
fault diagnosis
