Research On Fault Detection And Diagnosis Of A Cracked Rotor System

Rotating machinery is widely used in modern industrial production environment.As the core part of rotating machinery,the health of rotor plays a decisive role in the normal operation of the whole equipment.Therefore,it is of great significance to detect and diagnose rotor faults timely and accurately.Crack fault is the most common fault in all kinds of rotor faults,so it is also of great significance to study the rotor crack fault.Under the background of mechanical equipment fault detection and diagnosis technology becoming increasingly intelligent and automatic,the traditional fault detection algorithm based on signal processing is time-consuming,high demand for expert knowledge and low generalization,which can not meet the requirements of "industrial big data".Therefore,combining with the theory of deep learning,this paper studied a complete set of fault detection and diagnosis algorithm of rotor system,and made targeted analysis on the crack fault of rotor.The details are as follows.According to the one-dimensional characteristics of rotor vibration frequency domain signal,combined with one-dimensional convolution structure,a rotor fault detection algorithm based on 1D-GANomaly is studied.In the training stage,the algorithm only needs the normal frequency domain signal of the rotor.In the test stage,the potential feature difference of the unknown state samples is encoded twice,and the difference is used as the index of fault detection,so as to identify the rotor fault frequency domain signal.Then,the effectiveness of the algorithm is verified on the simulated rotor fault dataset and the CWRU rolling bearing fault dataset.A rotor system fault diagnosis algorithm called CWT-MS-CNN is studied based on wavelet time-frequency diagram and multi-scale convolution.The algorithm transforms the rotor vibration signal into two-dimensional time-frequency image by continuous wavelet transform as the input of the algorithm.The fault features in time-frequency diagram are extracted by convolution kernel of different sizes to complete fault diagnosis.The accuracy of the algorithm is verified on the rotor fault data set.Then,Aiming at the problem of MS-CNN diagnosis rate reduction caused by unbalanced datasets,a data expansion strategy based on auxiliary classifier generative adversarial network ACGAN is studied,which uses ACGAN to generate wavelet time-frequency images of fault categories with a small number of samples to make the dataset rebalance.The performance changes of MS-CNN before and after ACGAN expansion are compared and analyzed,which shows that the strategy can significantly improve the accuracy of MS-CNN in the case of unbalanced dataset.The rotor model is established by combining modal test and BP neural network.The influence of crack location and depth on the rotor natural frequency is analyzed by finite element method.By calculating the SIF of rotor surface crack,the variation law of SIF value at the midpoint and both ends of crack front with crack depth and length is revealed.