Fault Diagnosis Of Planetary Reducer Based On Iterative Hilbert Transform And Support Vector Machine

As an important part of the mechanical transmission, planetary reducer is widelyused. If faults occur on the reducer, the production safety and economic benefit will besubjected to severely negative impact. According to the urgent need of reliability ofequipments, aiming to the key problems of planetary reducer fault diagnosis, especiallythe problems of big noise, non-stationary and multi-components feature of vibrationsignal because of the complicated transmission path and interaction，the study of featurecomponents extracting method and fault mode recognition method is launched. All inall, the study in this thesis has the vital scientific significance and the project use value.Considering the effects of the time-varying meshing stiffness, the influence of thegear backlash, and error excitation, a nonlinear dynamic model of planetary reducerwith multi-clearance, variable parameter and crinkle coupling is established to study thedynamic characteristics of planetary reducer.The signal model and frequency spectrum model of planetary reducer in normal,distributed fault and local fault conditions are analyzed in this paper. Compared withgeneral reducer, planetary reducer consists of more parts and runs more sophisticatedly.The ‘passing effect’ caused by the revolution of the planet gear leads to the vibrationsignals collected by sensors fixed on the reducer containing more signal components.This signal obtained is multi-component AM-FM signal.An improved IHT method is put forward in this paper to extract the features ofmulti-component Am-Fm vibration signal. The original IHT method exists thedisadvantages of End Effect and the design of the filter relies on the priori knowledge.Analyzing the causes of the End Effect in the method of IHT, a method based on thesymmetric extension of signal to eliminate the End Effect is proposed. Meanwhile, anadaptive zero-phase filter design method is proposed to replace the fixed filter in theoriginal method. Simulation testing shows that the new method in this paper has ahigher accuracy in the demodulation and efficiently inhibits the end effect comparedwith the original method.The sample entropy can reflect the complexity of the signal’s internal structure andmulti-class classification SVM can classify the small sample effectively. This paperdesigned an experiment of planetary reducer with different faults, and used theimproved IHT method, original IHT method and EMD method to exact the feature signal components. The contrasts show that improved IHT method has the best effect.Then the sample entropy of components exacted and speed information build the featurevector and SVM method is used to recognize the fault pattern. The results show that theimproved IHT and SVM method can be used to diagnose the fault of planetary reducer.