Research On Fault Diagnosis Method Of Mechanical Equipment Based On Nonlinear Filtering In The Background Of Complex Noise

Accurate and timely identification of faults during the operation of the equipment is of great significance for ensuring the safe operation of the mechanical system and reducing or avoiding major catastrophic accidents.Modern large-scale complex mechanical equipment has the characteristics of large scale,high integration,complicated operation mechanism,and parameter and structure uncertainties.At the same time,the modeling errors,noise disturbances,and high complexity of the algorithms in the diagnosis make the fault diagnosis of large and complex equipment difficult.Improving the accuracy and real-time of fault diagnosis of complex mechanical equipment is a critical problem that needs to be solved urgently in the field of fault diagnosis.This thesis focuses on the fault diagnosis technology of nonlinear filtering,and particularly studies the key techniques such as performance optimization of filtering algorithm in complex noise background,and real-time of online diagnosis.The main innovative research results of this thesis are as follows:1.The particle filter algorithm with unknown time-varying noise and noise correlation is studied.By analyzing the influence of unknown time-varying characteristics of noise in complex systems on system accuracy,the particle filter algorithm of unknown time-varying noise is studied.The accuracy of state estimation in complex noise environment is improved by real-time estimation and correction of noise statistical characteristics.Aiming at the phenomenon of particle degradation while considering the time-varying characteristics of noise,the resampling optimization strategy is studied,and a combination of partial resampling and mutation operations is introduced to improve the sample quality,and increase the sampling efficiency.By analyzing the correlation between system noise and measurement noise and the influence on filtering performance,the particle filtering method of correlated noise is studied.Based on the correlation noise,this study deduces the specific mathematical expressions of the optimal proposal distribution function under noiserelated conditions according to the criterion of minimization of variance of importance weight conditions,and establishes a particle filter algorithm based on the correlation noise to improve the accuracy of state estimation in complex noise background.2.The intelligent optimization CRPF algorithm under complex noise conditions is studied.Aiming at the unique advantages of CRPF algorithm in solving the problem of high nonlinearity and unknown statistical characteristics of noise,the characteristics and existing problems of CRPF algorithm are deeply analyzed,namely the problem of particle deficiency and the state estimation under the background of strong disturbance and time-varying noise.When the system noise and measurement noise are relatively large,resampling based on the weight of the class will result in loss of sample diversity.Aiming at this problem,before the resampling step,the intelligent optimization strategy of partial cross mutation is introduced,and the optimized particle set is resampled according to the probability mass function,which increases the particle diversity and expands the posterior distribution area of the particle.Aiming at the problem that the adjustment ability of the state transition density covariance is worse under the background of strong noise,an adaptive adjustment strategy of state transition density covariance is designed to increase the adaptability of the algorithm to strong noise and improve the accuracy of state estimation under complex noise conditions.3.The adaptive CRPF fault diagnosis method under strong noise background is studied.Aiming at the problem of low fault diagnosis accuracy caused by strong noise interference in nonlinear non-Gaussian system in a real working environment,the fault diagnosis method based on residual based on the proposed IOCRPF algorithm is studied.The fault diagnosis system based on IOCRPF algorithm is constructed to obtain an accurate residual signal.At the same time,according to the change of noise and uncertainty of the system,the adaptive threshold of the residual discriminant function is studied and designed on the basis of statistical methods,which introduces the sliding window to find the interval mean value instead of the mean and variance based on the parameter confidence interval adaptive threshold,and under the premise of ensuring the accuracy of fault diagnosis,the calculation time is reduced,and the accuracy of fault decision is improved.4.Aim at the real-time problem of fault diagnosis,the parallel accelerated CRPF algorithm is studied under the CUDA framework.In the background of complex noise such as unknown and time-varying,this thesis proposes an IOCRPF algorithm based on noise adaptive filtering,which has achieved good results in algorithm accuracy,but at the same time greatly increases the complexity and computational time of the algorithm,and can not meet the online real-time requirements for diagnosis.Aiming at this problem,the parallel structure of CRPF under CUDA architecture is studied,and the parallel structure is optimized by the idea of block parallel for the problem that parallel resampling cannot be performed due to data correlation.Furthermore,in order to reduce the global particle performance degradation caused by block resampling,the particles with low probability mass in each block are optimized by global high-quality particles,and a CRPF algorithm for optimizing block parallel acceleration is constructed.Based on this parallel algorithm,a fault diagnosis method based on residuals is designed,and the real-time performance of the algorithm is significantly improved,meanwhile,the accuracy of fault diagnosis is ensured.5.In the CUDA architecture,the multi-model fault diagnosis method based on multi-GPU parallel acceleration is studied.The process and characteristics of multi-model fault diagnosis based on parallel CRPF algorithm are analyzed.Aiming at the shortcomings of the single-GPU parallel algorithm in multi-model and multi-fault diagnosis process,a multi-GPU parallel fault diagnosis method based on multi-GPU parallel CRPF is proposed.The CRPF algorithm for block-parallel acceleration builds a two-layer parallel model between the GPU and multiple GPUs,which greatly improves the parallelization of the program and improves the real-time performance of multi-fault diagnosis.For the fault of DFIG current sensor,the fault detection and isolation strategy based on residual is designed,and experimental analysis of the accuracy and real-time of fault diagnosis in CUDA environment is launched,which proves that this method can not only improve the accuracy of diagnosis,but also has superior acceleration performance.This thesis launches exploration and research for the accuracy and real-time fault diagnosis of mechanical equipment under complicated noise environments.Some innovative research ideas are proposed for the critical problems of particle filter algorithm performance optimization,precision improvement,and online diagnostic operation efficiency.It further solves the real-time problem brought by the particle filter algorithm in order to improve the accuracy in practical applications,and is of great significance to the innovative research of nonlinear filtering algorithms in the field of fault diagnosis.