Research And System Development Of Rolling Bearing Life Prediction Method Based On Unbalanced Data

With the rapid development of information technology,mechanical equipment is increasingly developing in the direction of large-scale,precision,intelligence,automation and systematization.However,the working environment of mechanical equipment is often abominable and changeable,and the equipment will gradually retrograde after long-term operation,which greatly increases the possibility of breaking out.Bearing is a key component of rotating machinery equipment,and its reliable operation can improve the safety and efficiency of modern production equipment.Generally speaking,bearings usually are used in different environments,and different types of faults often occur in the course of operation.If effective protection measures are not taken in time,it may lead to equipment failures,resulting in great economic losses and even casualties.Therefore,the prediction of the remaining life of the bearing will be a very important work and has profound practical significance.Driven by big data and artificial intelligence,the research on bearing life prediction and health management technology has been more profound.By focusing on complex interaction between rolling bearings and other parts in the actual industrial production environment,the vibration signals contain noise and useless vibration signals,and the characteristics of the vibration signals are relatively weak in the early fault stage.Based on the situation I expound above,a secondary filtering method based on wavelet transform and whale optimization parameter adaptive VMD is proposed.The experimental results show that the algorithm can effectively overcome the noise interference and mode chaos,and robustness is better,because it provides a complete experimental idea for the denoising process of rolling bearing vibration signals.In order to solve the problem that the fault state data set of rolling bearing can not be long-term and the collected fault state data set is far less than the unbalanced data set collected under normal operation state.This paper proposes to use the Cycle GAN network model to carry out iterative training using the existing fault state data set,and to use the mirror cycle of the network model against style transfer mechanism to improve the simulation ability of the generated fault data samples and generate a large number of high-quality fault data samples to solve the problem of sample imbalance in the data set.In the actual engineering environment,the types of bearings are complex and diverse.In order to solve the problem that the learning cost of the training model from scratch in the new data set is too high,transfer learning is introduced to verify the applicability of the existing network model and shorten the training time of the new network model so as to reduce the training cost and difficulty.Combined with the actual situation that offline transfer learning takes a long time,this chapter proposes a model knowledge transfer method based on knowledge distillation pre-training to shorten the knowledge transfer time of rolling bearing life prediction model.Finally,the residual life prediction model of rolling bearing is developed by using Python language and the vibration data processing algorithm is developed rapidly by using modular programming.The interface design of rolling bearing residual life prediction software system is completed based on C sharp,and the improved Cycle GAN network model is built by using Py Torch framework.Finally,the data set is used to simulate the signal data to verify the feasibility of the neural network algorithm to predict the remaining life of rolling bearings.