Research On Rolling Bearing Fault Diagnosis Method Based On Blind Source Deconvolution And Architecture Search

With the advancement of science and technology,the degree of automation of industrial production has gradually deepened,which puts forward higher requirements for the reliability of mechanical equipment,so fault diagnosis technology has achieved more attention,in recent years.Rolling bearings are the core parts of mechanical rotary motion,and it is of great significance to monitor and diagnose the health status of rolling bearings.Traditional fault diagnosis technology often requires manual extraction of features,which has disadvantages such as low detection efficiency,high application environment requirements,and a high probability of misjudgment.When using machine learning algorithms for diagnosis,it is difficult to remove the influence of signal noise and affect the classification results.And the algorithm model needs to be adjusted according to the actual situation.This paper proposes an adaptive maximum second-order cyclostationarity deconvolution algorithm and a rolling bearing fault diagnosis algorithm that improves the differentiable neural architecture search algorithm,so that the pre-processing operation and classification process of the bearing vibration signal can be adaptively constructed,and a fault test bench is built to verify the effectiveness of the algorithm.Finally,a rolling bearing fault diagnosis system is developed based on the algorithm.The main research contents are as follows:1.In this paper,according to the characteristics of the blind source deconvolution method with few parameters and high efficiency,the maximum second-order cyclostationarity blind source deconvolution algorithm is selected for signal denoising,and the algorithm is self-adapted to the bearing state.Firstly,the simulation signal of rolling bearing is constructed,and the influence of main parameters on algorithm performance is analyzed.Then,according to the characteristics of the low coupling effect of the main parameters of the algorithm on the performance,a stepwise optimization strategy is proposed.In order to realize the adaptive selection of the algorithm parameters,this paper use the spindle rotation frequency of the bearing running as the initial cycle frequency,the filter length is optimized based on fuzzy entropy and whale optimization algorithm,and according to the characteristics of the vibration signal of the faulty bearing,the comprehensive impact index is designed for the step size optimization of the cycle frequency.Finally,the open data set is used as the research object to carry out in-depth experiments on various kinds of bearing faults,and the effectiveness of the algorithm is verified by comparing similar methods.2.In order to solve the problems of low efficiency of artificial diagnosis and manual adjustment of the intelligent classification model,this paper introduces differentiable architecture search to construct a classification network adaptively,and proposes a multioptimization differentiable architecture search based on feature aggregation.Firstly,according to the characteristics of the rolling bearing signal,the time-frequency characteristics and highorder spectrum of the signal were fused to generate images to improve the sensitivity of the model to the impact characteristics.A multi-optimization strategy is proposed to solve the problem of non-synchronous optimization of structure and parameters.According to the multiple optimization strategies,the infrequently used operation list is generated,and the new network is trained after the reduced operation space.The original model is used for knowledge distillation of the training model after the reduced operation space,which can accelerate the convergence of the model and reduce the size of the model.Combined with adaptive blind source deconvolution algorithm,the self-adaptive construction of the highprecision bearing fault classification model is realized.A bearing fault vibration signal acquisition platform is built to verify the effectiveness of the algorithm,and the proposed method is compared with other common methods to verify the performance of the algorithm.3.According to actual production needs,a rolling bearing diagnosis system has been developed.First,design the overall structure of the software,database structure and information interaction logic according to the needs of the use function,and then select the acquisition driver support to realize the management function of the sensor management and the signal acquisition function of the software,and complete the information management logic inside the software and user interaction interface,and finally embed a variety of signal post-processing algorithms into the software.According to the characteristics of the algorithm,based on technologies such as C language reconstruction and process communication,multilanguage mixed programming is realized.The method proposed in this paper is embedded in this software for better processing effects.The software supports industrial use and enables real-time monitoring of bearing status.