| Rolling bearings are a very important part of all rotating machines.Their function is to support the machine and allow the shaft to rotate relative to the fixed structure.Failure of the bearing may further lead to machine failure and affect safe production.Therefore,they are considered to be key components in industrial applications,during operation,the bearing is subjected to heavy and dynamic loads generated by the machine tool and transmitted through the components of the rolling bearings.The state of rolling bearing is very important in high-volume systems.It is necessary to monitor its operating status in real-time and identify any defects in the bearing in time.Avoid increased downtime,production time and catastrophic failure of the machine.Acoustic measurements,temperature monitoring,wear debris analysis and vibration measurements are common for fault analysis of rolling bearings.Vibration monitoring is a widely used and economical monitoring technique for identifying,locating and distinguishing defects in rolling bearings.The main content is divided into the following sections:1.Aiming at the nonlinear and non-stationary of the original vibration signal of rolling bearing,a data preprocessing method based on mixed feature extraction is proposed in this paper.This method is mainly divided into four parts:extract the time-domain characteristics of the original vibration data;extract the frequency domain characteristics of original vibration data;based on the defects of traditional wavelet transform and EMD decomposition,combined with the theories of empirical wavelet transform and fuzzy entropy,a new time-frequency domain feature extraction method was proposed,which could well characterize the time complexity and energy characteristics of bearing signals under different working conditions;A method of rolling bearing mixed feature extraction is proposed to obtain a mixed feature vector,which can accurately reflect the characteristics of the bearing in different states and better adapt to the state monitoring and fault diagnosis models.2.Aiming at the problem of rolling bearing condition monitoring,based on the mixed feature extraction method,the EWT-MSPCA is proposed to improve the traditional multi-scale principal component analysis method by combining the empirical wavelet transform.The real vibration data collected by the experimental table are used for simulation analysis,and the monitoring experimental simulation results were completely consistent with the actual running state of the bearing,which fully verified the effectiveness and correctness of the monitoring algorithm proposed in this paper.Besides,by changing the diameter of bearing faults to simulate different degrees of faults and increasing comparative experiments,the results show that the proposed algorithm can accurately monitor various working conditions of bearing.3.Aiming at the problem of rolling bearing fault diagnosis,a rolling bearing fault diagnosis model based on a mixed feature extraction method and PNN algorithm is proposed.The real vibration data collected by the experimental table are used for simulation analysis,and compared with the experimental results of BP,AdaBoost,SVM,Random forests,ELM and non-mixed feature-PNN,the results showed that the proposed fault diagnosis model was better than other algorithms in accuracy and time. |
