Research On Fault Diagnosis Method Of Roller Bearing In Ring Spinning Frame

In modern textile spinning production,textile enterprises pay more and more attention to the health status monitoring and fault diagnosis of key components of ring spinning frame.At the same time,improving reliability and life prediction technology of key components of textile equipment is one of the development goals of "The 14 th Five-year Development Guidance for Textile Machinery Industry".At present,the health monitoring and maintenance of the roller bearing of the ring spinning machine in the workshop is still by regular inspection and maintenance,through hand touch to sense whether the vibration is too large and whether the temperature is too high;through listening to detect abnormalities.However,there are great limitations in the reliance on the experience to judge health status,such as low diagnosis accuracy,difficulties to inherit experience and knowledge.In order to improve the intelligent maintenance technology of roller bearing in ring spinning frame and reduce the dependence on empirical knowledge,this paper studied the fault diagnosis method of roller bearing in ring spinning frame,proposed a fault diagnosis method based on frequency domain indicator and random forest and signal processing and deep learning respectively.A fault diagnosis system for roller bearing in ring spinning frame is designed and developed in this paper in the last.This research has great significance to the intelligent diagnosis of roller bearing in ring spinning frame.The main research work of this paper is as follows:Firstly,the failure mode and reason of roller bearing in ring spinning frame are analyzed,and the formation mechanism of fault signal and fault diagnosis principle are described.Then the dataset of real and simulated fault bearings were made,which provided data support to verify the method proposed in this paper under serious and minor fault.The fault diagnosis and analysis of roller bearing in ring spinning frame are carried out by using common signal processing method.Firstly,The results showed that the sensitivity of each time domain indicator and frequency domain indicator changes at different rotational speed,which makes it difficult for the time domain indicator and frequency domain indicator to be used for fault detection and diagnosis at different rotational speeds of roller bearing.Secondly,four common bearing fault diagnosis methods,fast Fourier transform(FFT),envelope spectrum,short-time Fourier transform and Hilbert-Huang transform,are used for diagnosis and analysis.The results show that none of the four methods can achieve fault detection and diagnosis of roller bearing in ring spinning frame.To realize roller bearing fault diagnosis,this paper proposes an intelligent fault diagnosis methods based on frequency domain indicators combined with random forest,the accuracy of proposed method on the real fault bearing dataset and simulated fault bearing dataset are 85.05% and91.79% respectively,and its accuracies are better than that of k-nearest neighbor,decision tree,support vector machine and extreme machine learning four machine learning methodsConsidering that there may be information loss of the characteristics extracted by frequency domain indicators,machine learning has limitation on ability to extract deep features.In addition,considering the actual spinning production should set different roller speed according to different spinning task and the roller bearing fault characteristic frequency and amplitude under different speed are variable,thus a roller bearing fault diagnosis method based on FFT and 2D-CNN is proposed to solve across speed problem.The accuracy of this method is 99.70% and 99.64% on the real and simulated fault bearing dataset respectively,which shows superiority over other signal processing methods.Considering the requirements of enterprise data storage cost and rapid diagnosis,an intelligent diagnosis method based on FFT and 1D-CNN is proposed.The optimal sampling frequency can be reduced to 2560 Hz under the requirement of accuracy proposed by the enterprise.Compared to original sampling frequency 20480 Hz,the test time on the real fault bearing dataset is reduced by 56.60%,and the test time on the simulated fault bearing dataset is reduced by 25.28%.The average actual diagnosis time of a single roller bearing is reduced from 10.02 s to 0.49 s for 1D-CNN relative to 2D-CNN,which greatly reduces the diagnosis time.Considering the three fault diagnosis methods proposed in this paper from the perspectives of accuracy,diagnosis time and data storage cost,the fault diagnosis method based on FFT and 1D-CNN is selected as the final fault diagnosis method for roller bearingFinally,a fault diagnosis system for roller bearing of ring spinning frame is developed.The signal processing method and diagnosis method for roller bearing are embedded into the system.The system realizes the functions of health status monitoring and diagnosis result statistics,signal processing and analysis,maintenance information editing and saving,maintenance report generation and so on,which provides broad application prospect for the intelligent fault diagnosis roller bearing in ring spinning frame workshop by passing the system function and performance test.