针对滚动轴承损伤程度的特征自学习提取与智能诊断问题,提出栈式稀疏加噪自编码深度神经网络的滚动轴承损伤程度诊断方法.滚动轴承损伤特征受到工况、环境噪声等干扰,浅层自编码网络对损伤特征的自学习、提取能力不足.为此,论文将稀疏项限制和加噪编码融入自编码网络,同时将自编码网络堆栈并添加分类层,构建出栈式稀疏加噪自编码深度神经网络,进行轴承损伤特征非监督自动提取与损伤程度智能诊断.稀疏项限制和深度神经网络的构建提高了特征学习能力,加噪编码的融入改善了网络的鲁棒性.所构建深度神经网络通过多层无监督逐层自学习和有监督微调,完成损伤特征自动提取与表达,并实现了损伤程度智能诊断.不同工况下轴承损伤程度诊断的实验验证证明了所提方法的可行性和有效性.%Aiming at the self-taught learning of fault severity features and the intelligent diagnosis for rolling bearings,a fault severity diagnosis method based on a stacked sparse denoising auto-encoder was proposed.The fault severity feature of rolling bearings is easy to be disturbed by the operating conditions and noises,and shallow networks are usually lack of enough ability in the self-taught learning and fault feature extraction.Therefore,a sparsity penalty term and a denoising encoder were the fused into the auto-encoder.Moreover,the auto-encoder network was stacked and a classification layer was added to construct the stacked sparse denoising auto-encoder deep neural network and to achieve the unsupervised feature extraction and intelligent diagnosis for rolling bearings.The ability of feature learning was improved by the sparsity penalty term and stacked auto-encoder,and the robustness of network was improved by the denoising encoder.The fault feature was automatically extracted and expressed to realize intelligent diagnosis,through training the layers individually without supervision and fine tuning with supervision.The feasibility and validity of the present method were verified by its application in diagnosing the fault severity of rolling bearings under different operation conditions.
展开▼
