Research On Prediction Method Of Remaining Useful Life Of Rolling Bearing Based On Deep Learning

In industrial production,as one of the core parts of mechanical equipment,rolling bearings are widely used in mining industry,automobile industry,aerospace industry,petroleum and chemical industry and other fields.Their health status directly affects the normal operation of mechanical equipment.Once bearing failure occurs,it will cause industrial production suspension and economic losses in light,and cause casualties in heavy.However,bearings often work in extremely harsh environments such as high speed,heavy load and variable working conditions,which are easy to be damaged.Therefore,it is of great significance to accurately predict the remaining service life of rolling bearings for improving the reliability of mechanical equipment,ensuring the safety of personnel and making reasonable maintenance plans.In recent years,with the continuous development of science and technology,sensors and computer hardware are updated and iterated rapidly.Therefore,monitoring data and computer computing power in the industry have experienced explosive growth,while deep learning relies on the existing computing power and has been widely used in various fields and achieved important results with its powerful data processing ability.Based on this,this thesis adopts the deep learning method,mainly aiming at the long service life of wind turbine bearings and shield machine bearings,inconvenient disassembly and maintenance of bearings,and high reliability requirements.Focusing on the construction of feature extraction model of bearing original vibration data,life prediction model,network structure and parameter optimization,the remaining service life prediction of rolling bearing was studied deeply,and based on this,the deep learning based rolling bearing life prediction software was designed and developed.The details are as follows:1)The input data of traditional remaining useful life prediction model depends on the output of the previous moment,which leads to a long time of model training.A prediction method of remaining useful of rolling bearings based on self-attention mechanism convolutional neural network and bidirectional simple recurrent unit was proposed.Firstly,the original vibration signals are processed using a self-attention mechanism and convolutional neural network to extract key features,obtaining distributed feature representations for both training and prediction samples.Then,a fixed time window sliding method is employed to assign lifespan labels to the extracted windowed feature sequences,resulting in a lifespan feature matrix that captures temporal information.Finally,the lifespan feature matrix and its corresponding labels are fed into a bidirectional simple recurrent unit network for training and lifespan prediction.Experimental data show that the proposed method achieves better results in training time and prediction accuracy.2)The prediction model of the remaining useful life of the proposed rolling bearing has many super-parameters,complex network structure and over-reliance on manual experience,which leads to difficulty in parameter adjustment and poor prediction accuracy.The optimization of bearing life prediction network super-parameters based on mixed sine and cosine algorithm and Sparrow algorithm of Levy flight is proposed.Firstly,the ISSA is used to optimize the parameters and structure of the feature extraction model,which is a self-attention mechanism convolutional neural network.Then,the optimized feature extraction model is employed to extract key features from the original vibration signals.Secondly,a fixed time window sliding method is applied to the extracted features to obtain a lifespan feature matrix that captures temporal information.This lifespan feature matrix is then used as input data to further optimize the structure and parameters of the prediction model using ISSA.Finally,the BSRU is utilized to efficiently train the model by parallel processing the data.The effectiveness and feasibility of the proposed method are verified by experiments.3)According to the requirements for life prediction of rolling bearings in industry,and combined with the proposed algorithm,C# and Python are used to jointly design and develop a deep learning-based residual service life prediction system of rolling bearings that integrates data management,time-frequency analysis,model training and other functions.The software system mainly includes user information management,database management,data analysis(time domain characteristics,time domain diagram,frequency domain diagram),life prediction and other functions,which can easily realize the data collected for deep learning training,and easily complete the remaining useful life prediction of rolling bearings.Finally,summarize the work done and look forward to the next research direction.