Study On Fusion Of Multi Model Intelligent Diagnosis Method On Gear Early Faults Based On Evidence Theory

Gear transmission is a kind of widely used fundamental parts, it is alse said that gear transmission is mechanical system which is most widely used, and gear is alse a key infrastructure components which is damaged easiest,whose running state has running condition has directly affects to the safe operation of the machine or equipment. Therefore, it is most important how to detect the fault of geared system as early as possible to make the repair sehedule reasonably and economically to avoid the occurrence of major accidents and cause great economic loss is of great significance.Gear meshing vibration signals more or less reflect system states(normal or fault), gear faults demonstrate certain regularities during gear transmission as well. Therefore, such signals offer major guidance when monitoring large, key machine units operation or fault diagnosis during machinery management and maintenance.Due to the gear vibration response and environmental noise interference, the early weak signals of the gear fault is often submerged. The fault signals monitored often demonstrate complex non-linear and non-stationary characteristics. It is difficult to achieve an accurate diagnosis with traditional signal processing method based on stationary signal hypothesis. Therefore, effective de-noising, de-noising pretreatments and non-stationary signal processing techniques can make a big difference for the fault diagnosising of machinery equipment.Wavelet analysis and empirical Hilbert transform decomposition are two kinds of time-frequency modes recent developed to handle non-stationary signals. Common de-noising methods, such as wavelet threshold de-noising, morphology filter, singular value decomposition technique, combined with the 2 aforementioned time-frequency methods, are widely used in signal detection, mechanical fault diagnosis.Along with the trend towards high speed, high power, high reliability, large scale(micro), intelligent, integrated equipment, the traditional equipment fault diagnosis and single intelligent diagnosis method cannot fully meet the challenge of the increasing complexity of equipment states. The diagnosis integrating a variety of intelligent diagnosis methods and technologies has now become a research field of rigorous debate. Drawing on the dynamics of gear system, fault formation mechanism, wavelet analysis and threshold de-noising, small spectrum singular value de-noising redistribution, Hilbert Huang transform theory, D-S evidence theory, genetic algorithm, BP neural network and fuzzy optimization, this research put forward an integrated model based on D-S evidence theory for intelligent gear transmission system fault diagnosis. Result of Ananlying the typical fault data of gear verified the effectiveness of the proposed method, which offered a valid basis for the fault diagnosis of gear transmission system.In this paper, the research are carried out in the following way.[1] This paper set up a gear transmission model which take into account of friction, time-varying stiffness, backlash with eccentric gear friction as well as gear clearance. This study also analyzed the gear dynamics and its the spectrum characteristics under the afore-mentioned condition[2] This article proposed a redistribution spectra with wavelet scale SVD noise reduction method based on TBP parameters optimized by Shannon entropy. The analysis of simulated signal indicates that this method has a clear time-frequency convergence in spectra compared with redistribution wavelet scale spectrum and wavelet scales spectruman; the research found several clear pulse components indicating effective improvement of the readability of the time frequency distribution. As a result, the research concluded that such method can identify early fault weak signals in strong noise background.[3] This research proposed a fault diagnosis methods de-noising and diagnosis method based on Integrating wavelet threshold de-noising and the EMD fractal fusion. The research also listed concrete steps during this method, and have applied this method to the fault diagnosis of gear tooth surface wear, broken gear vibration signal; using mean square root value instead of correlation dimension value to diagnose the wear and fracture of the gear tooth surface correctly, and have achieved ideal results.[4] The research came up with an integrated multi-model intelligent gear fault diagnosis method following D-S evidence theory. After testing, this research proved that the proposed multi-model can utilize the merits of different single intelligent models, significantly heightened the differentiation compared with traditional single intelligent model. Such integrated model can also bypass judgmental errors from single intelligent models, and still reach a correct final diagnosis, and have demonstrated better fault tolerance, better faulty correction.