Intelligent Identification Of Rotor Rub-impact Fault Based On Deep Learning

With the continuous improvement of industrialization and the advancement of modern technology,the important role of large rotating machinery in many fields is becoming increasingly prominent.The rotor system is an important component of rotating machinery,and modern mechanical equipment requires a tight mechanical structure,which continuously reduces the gap between the rotor and stator.As a result,rotor rub-impact has become a very common fault.The occurrence of rotor rub-impact faults seriously affects the stability and efficiency of mechanical work,and even leads to shaft fracture,leading to major accidents and causing huge losses.Therefore,it is necessary to propose a diagnostic method for rotor rubimpact faults.This thesis focuses on the time-domain acoustic emission signals of rotor rubimpact faults under different working conditions,and combines algorithms to denoise them.After establishing a dataset,a neural network model is constructed to intelligently identify the acoustic emission signals of rotor rub-impact faults,achieving good recognition results.This thesis introduces the application of acoustic emission detection technology in rotor rub-impact fault detection,introduces the principle of acoustic emission detection technology and the characteristics of acoustic emission signals,and elaborates on the analysis and recognition methods of acoustic emission signals.Construct a simulation experiment for acoustic emission detection of rotor rub-impact fault,collect acoustic emission signals under different working conditions,and analyze the time-domain and frequency-domain characteristics of the signals.In order to solve the problem of a large amount of noise in the acoustic emission signal of rotor rub-impact fault,this thesis selects the wavelet threshold denoising method and proposes an improved threshold function to address the problems of continuity and constant error in traditional soft and hard threshold functions.The processing results of simulation signals and acoustic emission signals of rotor rub-impact indicate that this algorithm can effectively improve signal-to-noise ratio and reduce root mean square error.In response to the difficulties in feature extraction and reliance on expert diagnostic experience in traditional diagnosis methods for rotor rub-impact faults,this thesis uses a onedimensional Convolutional Neural Networks to preliminarily achieve the recognition.Based on this,a rotor rub-impact fault recognition method based on Convolutional Neural Networks,Long Short-Term Memory neural network,and Attention Mechanism is proposed.This model takes the amplitude sequence of acoustic emission signals from rotor rub-impact faults as input,adaptively extracts features from the sequence through Convolutional Neural Networks,and then extracts temporal features through Long Short-Term Memory neural networks.Combining Attention Mechanism,it effectively weights the hidden states,thereby improving the recognition performance of the model.Through experimental verification,compared to traditional neural network models,this model has a higher recognition accuracy for rotor rubimpact faults.