Research On Bearing Fault Diagonals Based On Dimension Transformation And Deep Learning

Rolling bearings,as one of the most important components in rotating machinery,play a crucial role in ensuring the smooth and efficient operation of mechanical systems.It has been observed that approximately 30% of failures in rotating machinery are related to bearing damage.In this paper,based on deep learning theory,we proposed a mechanical fault diagnosis algorithm that combined deep forests with neural networks to address issues such as the need for expert experience and the strong imbalance in fault signals.The specific contributions of this work are as follows:(1)To address the nonlinear nature of vibration signals in rolling bearings and the requirement of expert experience in traditional machine learning diagnosis algorithms,a fault diagnosis algorithm based on Convolutional Cascade Forests(CDF)was proposed.First,the one-dimensional vibration signals were normalized and converted into grayscale images;then,a multi-width convolutional neural network(CNN)was employed to perform end-to-end feature extraction on the images;finally,the features were analyzed and classified using cascade forests.Compared to multi-scale scanning methods,the multi-width CNN demonstrated superior representation learning capabilities with lower time complexity.Experimental results showed that CDF achieved high accuracy and exhibited good classification performance under different operating conditions and noise environments.(2)To improve the accuracy of bearing fault diagnosis under imbalanced samples in practical operating conditions,a Conditional Generative Adversarial Network(CGAN)and an improved Deep Forest(DF)fault diagnosis algorithm were proposed,referred to as CGAN-IDF.First,CGAN was used to generate augmented samples from real fault samples;then,the cosine similarity(CS)and Pearson correlation coefficient(PCC)were employed to evaluate the similarity between the generated samples and real samples,and highly correlated generated samples were saved during training;finally,the generated and real samples were mixed and input into the improved DF for training.Experimental results demonstrated that this method achieved high accuracy and good generalization performance in small-sample scenarios.(3)Building a rolling bearing fault analysis platform was developed using Tkinter.This platform includes data slicing and recombination,model training,and visualization operations,providing a more intuitive data display.