Research On Fault Diagnosis Method Of Rolling Bearings Based On Bayesian Optimization Dratnet

Rolling bearing is crucial for the correct operation of mechanical systems in modern equipment,and its condition is essential for the proper function of mechanical systems..In this paper,IGAF,convolution neural network,Gaussian process and Bayesian optimization are used to study the fault feature extraction and fault diagnosis methods of rolling bearing.Based on the vibration signal features specific to rolling bearing faults,this paper uses IGAF to characterize one-dimensional signals as two-dimensional images,which can extract fault signal characteristics more effectively.For the shortcomings existing in practical applications,a signal characterization method based on IGAF was proposed to improve the length of the signal needed to generate IGAF and improve the speed and characterization ability of IGAF in fault diagnosis,The proposed method’s efficacy was confirmed by simulating faulty signals.This research investigates the use of DenseNet in the context of recognizing images with high accuracy.In view of the difficulty of fault feature extraction and easy interference by noise,DenseNet is improved by using attention mechanism to enhance its ability to focus on bearing fault features adaptively.According to the characteristics of bearing fault vibration signals,the attention mechanism is improved,and The soft threshold algorithm is integrated to preserve fault characteristics while reducing noise and interference.This paper presents a bearing fault diagnosis method using a dense residual attention network(DratNet),utilizing IGAF feature images as inputs for rolling bearing fault diagnosis and the effectiveness of the improved correlation algorithm is demonstrated through ablation experiments.In the process of using DratNet for bearing fault diagnosis,the optimal hyperparameters are obtained by using Bayesian optimization algorithm.In the process of determining the probabilistic proxy model,a Gaussian process based on Marton-square exponential kernel function is proposed,which improves the approximation effect of Gaussian process in curve fitting of data.The flower pollination algorithm(FPA)was used to improve the MCMC,and then the parameters of the kernel function are optimized by MCMC-FPA,which further improves the accuracy of Gaussian process in predicting the distribution of unknown objective functions.A new acquisition function based on an adaptive confidence boundary strategy was proposed to enhance the adaptability of adjustment parameters in the process of determining the collection function.Based on the research content of this chapter,this paper proposes a Bayesian optimization DratNet method for rolling bearing fault diagnosis,and obtained the reliability and effectiveness through experiments.The rolling bearing fault simulation experimental platform was established to collect vibration signals from a faulty bearing and diagnose faults using the proposed method.High accuracy was achieved in diagnosis under varying working conditions.The effectiveness of the proposed fault determination method based on Bayesian optimization DratNet is proved by cross-working condition experiment and anti-noise experiment.