Research On Deep Learning Identification Algorithm Of Subhealth

In the industrial production process, a fault will occurr due to the complexity of the work environment and production process, but the fault not occur instantaneously, from a normal to a fault condition is a cumulative process, it is important to the industrial production and the smooth progress of economic and benefit maximization if we can predict the fault in advance and in timely regulation. Therefore, the research of fault occurrence and prediction becomes a focus in the field of fault diagnosis, and caused the attention of many experts and scholars.After reading a large number of deep learning and methods of fault diagnosis, the traditional fault diagnosis feature extraction difficult diagnosis result is not satisfactory, so we propose a method of fault diagnosis based on improved deep learning. In this paper, the stacked denoising autoencoder which can be filtered the noise of a large number of mechanical vibration signals is used for deep learning structure to extract the characteristics of the noise. The depth of the data can be extracted automatically by the unsupervised pre-train method, which greatly simplifies the process of traditional manual extraction. In this paper, a method is proposed to improve the stacked denoising autoencoder to get ride of the the gradient go away with deeper layers. To avoid the deep network expression ability than shallow network faults, combine the shallow and deep nodes is proposed and finally the use of support vector machine to extract the depth feature for classification.In this paper, for machinery and equipment from the "normal" state to the "failure" state classification limitations, introduce the reliability theory of Weibull Distribution to division of the subhealth state of rolling bearing.Due to the same bearing under the same operating lifetime distribution is not uniform, so it’s not only need to the overall prediction of bearing, but also need to test the online status of bearing, the deep learning of the method is proposed to divison a bearing different period state division method, after smoothing of time series, it is used in rolling bearing life prediction. In view of the different running state of rolling bearings, this paper put forward a based on statistical probability online prediction method, the rolling bearing condition time series prediction can be effective to the bearing of the service life prediction.Through the experimental data of this method verify from the comparison of the experimental results show that, improved stacked denoising autoencoder for rolling bearing fault diagnosis has good performance and different rolling bearings under different working conditions of the subhealth state diagnosis also achieved good results.