加噪样本扩展深度稀疏自编码神经网络的滚动轴承寿命阶段识别

发布时间：2019-07-28 13:29

【摘要】：针对滚动轴承寿命特征提取与寿命阶段智能识别问题,提出加噪样本扩展深度稀疏自编码神经网络的滚动轴承寿命阶段识别方法。稀疏自编码具有非监督自动学习数据内部结构特征的能力,但属浅层网络,特征提取能力有限且不具备分类能力。因此,将多个稀疏自编码堆栈并添加分类层构建出集寿命特征自动提取与识别功能于一体的深度稀疏自编码神经网络,通过无监督逐层自学习与有监督微调,完成寿命特征的自动提取与表达,并实现寿命阶段智能识别。同时,为解决寿命样本量不足导致的网络过拟合,对原训练样本进行加噪扩展来训练网络,以抑制网络过拟合并提高网络的鲁棒性。通过工程应用,证明了所提方法的可行性和有效性。
[Abstract]:Aiming at the problem of life feature extraction and intelligent recognition of rolling bearing life stage, a rolling bearing life stage recognition method based on noisy sample extended depth sparse self-coding neural network is proposed. Sparse self-coding has the ability to learn the internal structural features of data automatically without supervision, but it belongs to shallow network, and the ability of feature extraction is limited and does not have the ability of classification. Therefore, a deep sparse self-coding neural network which integrates the functions of automatic extraction and recognition of life features is constructed by adding multiple sparse self-coding stacks and adding classification layers. through unsupervised layer-by-layer self-learning and supervised fine-tuning, the automatic extraction and expression of life features is completed, and the intelligent recognition of life stage is realized. At the same time, in order to solve the problem of network overfitting caused by insufficient lifetime sample size, the original training samples are extended by adding noise to train the network, in order to restrain the network overfitting and improve the robustness of the network. The feasibility and effectiveness of the proposed method are proved by engineering application.
【作者单位】：重庆交通大学机电与车辆工程学院;重庆大学机械传动国家重点实验室;四川大学空天科学与工程学院;
【基金】：机械传动国家重点实验室开放基金资助项目(SKLMT-KFKT-201710) 国家自然科学基金资助项目(51305471,51775065) 中国博士后科学基金资助项目(2014M560719)
【分类号】：TH133.33

【相似文献】