Research On Early Fault Detection And Diagnosis Of Bearing Based On Optical Fiber Sensor

Rotary machinery is widely used in aviation engines,high-speed railways,wind turbines and other mechanical devices are important structural components,its performance and status will affect whether the system works normally.Rolling bearings are considered the most common cause of faults in rotating machinery,and according to statistics,over 40% of faults in induction motors are mainly caused by rolling bearings.When the bearing fails,the overall vibration level is affected.If the fault is not detected and the correct decision is not made in a timely manner,the consequences of the fault may be catastrophic.Therefore,detecting the early fault characteristic of bearing is great significance for ensuring the safe and reliable operation of mechanical system.When there is a slight fault on the bearing surface,the vibration signal will produce the periodic micro impact.However,they are often overwhelmed by heavy background noise and harmonic interference from coupling components such as gearboxes and shafts,which makes it difficult to detect and diagnose micro fault impact at an early stage.At present,the piezoelectric resonance sensor used for early bearing fault detection has the problem of electromagnetic interference,and optical fiber vibration sensor has the problem of expensive demodulation system.Consider the problems that early fault is difficult to extract,difficult to diagnose and low precision.In this thesis,optical fiber sensor and fusion filter method for early fault detection and diagnosis are studied.In this thesis,the optical fiber sensor based on mechanical second-order system is designed to monitor the rotating bearing,and then pick up the micro impact vibration signal generated by the early bearing fault,and then analyze the vibration signal,and finally realize the early bearing fault diagnosis.The main research contents are as follows:1.Establish the theoretical model of optical fiber sensor for early bearing fault detection.Firstly,based on the bearing fault types and characteristics,the characteristics of early bearing fault signals are analyzed.By analyzing the bearing structure and fault characteristic frequency,the advantages of generalized resonance in early bearing fault extraction are clarified.According to the principle of generalized resonance excitation and weak signal extraction,a theoretical model of optical fiber sensor based on mechanical second-order system and fiber Bragg grating sensor is established for early bearing fault detection.Compared with piezoelectric resonance sensor and optical fiber vibration sensor,it has the advantages of anti-electromagnetic interference,high sensitivity,high cost performance and suitable for engineering applications.2.The sensor for early bearing fault detection is researched and manufactured,and the bearing fault test platform is built.Based on the theoretical model of sensing,the sensor structure is designed,the 3D model is built by Unigraphics NX software,and the modal analysis and natural frequency analysis are carried out by finite element software.The natural frequency of the sensor is tested by impact experiment,and the natural frequency of the sensor is determined to be 6.335 k Hz.Finally,the experimental platform for sensor detection of fault bearings is designed.By testing and analyzing the faults of the inner and outer races of rolling bearings with different crack sizes at different speed.The design of optical fiber sensor with high natural frequency is realized,and the feasibility and effectiveness of the resonant method to extract the early weak impact fault of bearing are verified.3.The method for denoising and early fault diagnosis of bearing vibration signal based on Kalman filter and spectrum analysis is proposed.The method is to build the autoregressive model based on bearing vibration signals,use the final prediction error criterion and recursive least squares method to determine the autoregressive model,and then transform it into the state-space model to achieve the optimal state estimation by using Kalman filter.Finally,the improved autocorrelation envelope power spectrum and autocorrelation envelope maximum entropy spectrum method are used to achieve the early bearing fault diagnosis respectively.The method is sensitive to state changes and improves the filtering performance,in which the SNR is increased by at least 2.23%and the RMS error is reduced by at least 8.72%.In the aspect of fault diagnosis,it is more accurate and intuitive,and the relative error of fault characteristic frequency extraction is less than 1.867%,which effectively realizes the diagnosis of the outer race,inner race and rolling element faults of bearing early pitting corrosion and crack fault.4.Aiming at the problem that the filtering and spectrum analysis methods cannot realize the type and identification of bearing early fault signals,a method for bearing early fault signal identification and diagnosis with unknown input fusion filtering is proposed.In this method,the bearing fault signal is taken as unknown input,and a multi-sensor autoregressive model with unknown input,unknown model parameters,unknown system noise variance and unknown observation noise variance is established based on the bearing vibration signal.Recursive extended least squares algorithm,matrix-weighted fusion estimation algorithm in the sense of linear unbiased minimum variance,correlation function method and weighted average method are used to achieve fusion parameter identification and noise variance estimation.Finally,the optimal state filter and unknown input filter under linear unbiased minimum variance are obtained by model transformation.This method realizes bearing unknown fault identification based on multi-sensor fusion,and the fusion identification accuracy is higher than local accuracy,the fusion parameter identification accuracy is at least 45% higher.The relative errors of the observation noise variance and the system noise variance identification values are below 17.9% and 11.8%,respectively.This method can effectively identify and diagnose early time invariant impact fault,slow time variant impact fault,periodic impact fault and fast time variant impact fault of bearings.