【作者】 敦泊森； 王奉涛； 邓刚； 刘晓飞；

【Author】 DUN Bo-sen;WANG Feng-tao;DENG Gang;LIU Xiao-fei;Institute of Vibration Engineering,Dalian University of Technology;

【机构】 大连理工大学振动工程研究所；

【摘要】 提取滚动轴承有效的故障特征参数是轴承故障诊断重要的组成部分,针对支持向量机(Support Vector Machine,SVM)难以表征被测信号与轴承健康状况之间复杂的映射关系,以及高维数据下特征选取困难的问题,提出一种结合稀疏自动编码器(Sparse Auto-Encoder,SAE)与SVM的方法。首先,提取振动信号的时域、频域和时频域特征构成高维特征向量;其次,采用深度稀疏自动编码器对特征向量进行学习,并提取深度特征,输入到SVM中进行训练;最后,采用有标签数据对结构进行微调,优化训练模型。为评估方法有效性,采用实验室数据进行测试,并与传统SVM方法进行比较,结果显示该方法具有更好准确性和稳定性。更多还原

【Abstract】 Extracting effective rolling bearing fault feature parameters is an important part of the bearing fault diagnosis. For the problem of Support Vector Machine not capable enough to represent the complex mapping relation between signal and bearing, and the difficulty of high-dimensional data feature selection. A novel method of combined Sparse Auto-Encoder and SVM is proposed. Firstly, the vibration signal of time domain, frequency domain and time-frequency domain features are extracted to constitute a high-dimensional feature vector; Secondly, using the Deep Sparse Auto-Encoder to study the deep feature, and enter into the SVM training; Finally, using the label data for fine-tuning, and optimize training model structure. To assess the validity of the method, the laboratory test data are adopt to comparing the proposed method with the traditional SVM, the results show that this method has better accuracy and stability.更多还原