Rolling bearing residual life prediction method based on LSTM and TDNN

Abstract

The invention discloses a rolling bearing residual life prediction method based on LSTM and TDNN, and the method comprises the steps: (1) extracting the time domain features of an original vibration signal of a training bearing, forming a time domain feature set, and carrying out the minimum-maximum normalization processing of each feature; (2) constructing a health factor by using at least two nonlinear functions, then constructing a mapping relation between training features and the health factor and a mapping relation between the health factor and the residual life percentage by using LSTM and TDNN in sequence, and constructing an LSTM-TDNN residual life prediction model; and (3) inputting test bearing vibration data, after feature extraction and normalization processing, obtaining a residual life estimation value according to the LSTM-TDNN prediction model, and giving out comprehensive evaluation of prediction accuracy. The rolling bearing residual life prediction result is more accurate, unification of high precision and strong conservative is achieved, the method can be effectively applied to rolling prediction maintenance tasks, and the maintenance cost is reduced.

Description

technical field [0001] The invention relates to the technical field of residual life prediction of rolling bearings, in particular to a residual life prediction method of rolling bearings based on LSTM and TDNN. Background technique [0002] As the core components of modern industrial equipment, rotating machinery has played an important role in aerospace, water conservancy and hydropower, chemical metallurgy and other fields. As a key part of rotating machinery, rolling bearings affect the operation of the entire mechanical equipment. The failure of the rolling bearing may cause the mechanical equipment to fail to achieve its intended function, which will affect industrial production and bring economic losses, and endanger the life safety of employees. Therefore, it is of great guiding significance to predict the life of rolling bearings. [0003] At present, the remaining life prediction methods for rolling bearings are mainly divided into two categories: analytical model...

AI Technical Summary

Problems solved by technology

[0004] The existing intelligent prediction methods for the remaining life of rolling bearings mostly use convolutional neural networks or recurrent neural networks as the main body to complete data preprocessing and estimate the remaining life, and often have the following problems: First, the existence of sequence unequal lengths and time delays is ignored , due to different working conditions, installation locations and environments, the actual service time of bearings is different, and there is a problem of unequal sequence length when training the prediction model. In addition, when using historical operating information to estimate the remaining life at the current moment, the time delay is difficult to quantify; Second, the performance index that reflects the health status of the bearing, that is, the acquisition process of the health factor is complex and the diversity is ignored, and the portability of the method is poor; Lack of evaluation, in practice, high precision and strong conservative unity should be achieved as much as possible

Images