Bearing Health Assessment And Prediction Research Based On Convolutional Neural Network And Grey Model

The speed of science and technology has gone far beyond the imagination of our human beings.The era of “big data” has come to the fore.Every day,hundreds of millions of data was generated.These data contain a lot of information.Using this information reasonably can help people understand more clearly and understand the intrinsic links in the data.The rolling bearings are one of the key components in many mechanical equipment.The rolling bearings are usually working under complex working conditions,high strength and high load working conditions.If it breaks down,other parts will have a chain reaction,which will eventually lead to a complete collapse of the mechanical equipment,and the loss and damage are extremely serious.Using the data-driven method to deal with the rolling bearings vibration signal gradually replaces the purely researched signal processing knowledge and principles.The artificial intelligence algorithm is used to reasonably extract the characteristics of the vibration signal for the rolling bearings fault recognition and life prediction.In recent years,the development of artificial intelligence and deep learning has been blown out.Therefore,this paper proposes a bearing health assessment and prediction research based on convolutional neural networks and grey model.The main points of this paper are as follows:(1)In the fault recognition of rolling bearing,in order to avoid the characteristics of human interference extraction,the convolutional neural network is used to directly diagnose and intelligently identify the vibration signal of the rolling bearing to realize the end-to-end process.Aiming at the problem that the network convergent speed is slow and the effect is not good due to too many parameters of the common convolutional neural network,the Convolutional Neural Networks of Average Information(AICNN)is proposed.The layer is replaced by the mean pooling layer,which not only can maximize the signal characteristics extracted by the previous convolutional layer and the pooling layer,but also reduce the number of network parameters.Finally,the reliability of the algorithm is verified by the case data of Case Western Reserve University.Even in the small sample training set,high recognition accuracy can be obtained.In the case of increased training set in the later stage,the accuracy rate is continuously improved and the test results are stable.Under the complex vibration signal of multiple working conditions,the factors that can be excluded from the working conditions can achieve a high recognition rate and verify that the algorithm is still universal in the case of noise interference.(2)In the extraction of the characteristic curve of rolling bearing state,the previous health index HI is constructed with a large number of features such as feature extraction,feature screening and feature fusion,and human interference.The structure of the convolutional neural network,which is often used for intelligent recognition and image classification,is improved.Forming a Regression Convolutional Neural Network(RCNN),the Softmax classification layer of the convolutional neural network is directly removed,and then the three-layer fully-connected layer is used to construct the training bearing model.The test bearing is obtained through the trained model to obtain the state characteristic curve,and the state characteristic curve is used to quantitatively evaluate the method in different training models.The most representative state characteristic curve is extracted.This method can effectively characterize the monotonicity and sensitivity of the state characteristic curve,and finally extract the state characteristic curve of the rolling bearing full life.(3)In the prediction of rolling bearing life,it is proposed to use the Affinity Propagation(AP)algorithm to perform unsupervised clustering on the state characteristic curves of all tested bearings.State characteristic curves with the same degraded morphology are grouped into one class without setting the number of categories in advance.The gray model is often used in the prediction of industrial,agricultural and transportation fields.For the first time,the Gray Forecasting Model with Full Order Time Power terms(FOTP-GM)is applied to the life prediction of rolling bearings.Under the premise of setting the failure threshold reasonably,all the test bearings are followed by themselves.The prediction of life expectancy shows that the method has a certain predictive effect,and the effect has a certain improvement compared with other researchers.