Research On Prediction Method Of Rolling Bearing RUL On Graph Structure

In recent years,while mechanical equipment is developing in the direction of automation and intelligence,its reliability is also receiving more and more attention.As a typical type of mechanical equipment,rotating machinery plays an important role in various fields.Rolling bearings are the core components of rotating mechanical equipment.Accurate remaining useful life prediction of rolling bearings can provide valuable reference for predictive maintenance of the equipment,which is conducive to eliminating the hidden danger of equipment failure,improving operational reliability,increasing the length of equipment use,and improving production efficiency and economic benefits.Therefore,it is of great practical significance to predict the remaining useful life of rolling bearings by means of intelligent technology.In this thesis,the rolling bearing is considered as the research object,the vibration signal is analyzed and modeled by data-driven method to realize the remaining useful life prediction of rolling bearings.The main research contents of the thesis are as follows:(1)In order to solve the problem that the standard sparrow search algorithm has the tendency to fall into local optimum and weak global search ability,an improved sparrow search algorithm(ISSA)which incorporates elite reverse learning with adaptive differential perturbation is studied.Firstly,the elite reverse learning mechanism is introduced to construct a reversed population by elite individuals in the current population,and the optimal individual is selected from the new population composed of the current population and the reversed population to enter the next iteration,so as to avoid the stagnation of optimization.Secondly,an adaptive differential perturbation strategy is introduced to balance the accuracy and speed of the algorithm by dynamically adjusting the perturbation factor so that the individual location updates can be adapted to the current stage.Finally,by comparing the standard test functions with other optimization algorithms,the overall performance of ISSA is verified,and the effectiveness and superiority of the proposed method is demonstrated..(2)Aiming at the problem that it is difficult to obtain accurate degradation trends from rolling bearing vibration signals due to non-smoothness and non-linearity,a method based on structural entropy(SE)for remaining useful life prediction of rolling bearings is studied based on a data-driven approach.Firstly,the non-stationary and non-linear vibration signals are processed by continuous wavelet transformation to obtain the time-frequency image containing the running state of the bearing.Secondly,the simple linear iterative clustering algorithm is used to segment the time-frequency image into super-pixels to obtain the structure of graphs.Then the community discovery algorithm and Shannon entropy are utilized to obtain the structure entropy of the graphs.Subsequently,the ISSA is used to optimize the input batch size of long short-term memory(LSTM),the number of hidden layer neurons,and the learning rate,avoiding the blindness and randomness of using empirical methods to determine network hyperparameters.Finally,health indicators describing bearing degradation are obtained,and the remaining useful life prediction of rolling bearings is achieved by using Gaussian process regression.The model performance is verified by PHM and XJTU-SY datasets,experimental results show that the SE-ISSA-LSTM model can predict the remaining useful life of rolling bearings with high accuracy.(3)The large volume and high dimensionality of vibration signal data make it difficult to build an accurate remaining useful life prediction model,and the manual design of features requires complicated steps.To address this problem,a graph convolutional networks(GCN)-based method for remaining useful life prediction of rolling bearings is presented in this thesis.First of all,the set empirical mode decomposition is performed on the data samples to obtain the intrinsic mode function(IMF)dataset of the vibration signal,and the Pearson correlation coefficient between the IMF components of each order is utilized to obtain the graph structure.Secondly,the graph structure is input to GCN to extract the spatial information of the data,and the output of GCN is used as the input of LSTM network to extract the time information.The network hyperparameters are determined by the ISSA proposed before.Then,health indicators describing bearing degradation are obtained,and the remaining useful life prediction of rolling bearings is achieved by using Gaussian process regression.Finally,the model is validated by PHM dataset and XJTU-SY dataset.The validity,generalization and robustness of the model are verified by comparing experiments and noise enhancement experiments.At the same time,the ablation experiments are designed to verify the performance improvement of the model by deep learning network.The experiment results show that the GCN-ISSA-LSTM model can effectively estimate bearing degradation,and the prediction results are close to the true values,which has certain advantages in practical applications.