Research On Fault Diagnosis Method Reducer Coal Conveyor Information Fusion

Reducer plays an important role as a drive device on coal conveyor. It is easily broken with ahigh frequence, making a bad effect on delaying long time to the production. Therefore, it is ofgreat significance to make a reaserch on reducer fault diagnosis in early failure period. There areso many kinds of failure to the reducer, and the main diagnosis is by the analysis on the powerspectral, frequency spectrum and vibratation signals, which get from many single sensorsinstalled on the reducer. The whole acquiring process is hard and have a low rate on faultdiagnosis. Based on lots of references on fault diagnosis of reducers and multi-sensorsinformation fusion, the paper introduces multi-sensors information fusion technology to explorefault diagnosis on reducers. The paper has done the following work, as show below.(1) In fault diagnosis of mechanical failure on the reducer, such as aixs unbalance, bearingfailure, gear failure and so on, the index of kurtosis, margin, wave shape and peak in time-domainare selected to judge the running status owing to the characteristic of eigenvalue, while thespectrum center, spectral variance and harmonic factor in frequency-domain are also selected.According to the characteristic of multi-sensors fusion, the diagnosis model is established on thefusion of feature-level and decision-level, selecting the method on information fusion of neuralnetworks and evidence theory.(2) Due to the flaw in classic fusion formula on D-S evidence theory, the paper applies anew synthetic formula. The formula takes three evidence bodies from three eigenvectors, whichis abstracted from three vibration sensors, to judge if the reducer is normal or have fault on gearor bearing. Comparing with the classic formula, the synthetic formula gets a better result onfusion, which shows the synthetic formula is available.(3) As regard to the specific failure on the reducer, The paper make feature fusion firstly,then use the decision judgement method to fault diagnosis. According to the simulation result ofBP and RBF network, The RBF network is selected to the fusion on feature-level. It takes threeevidence bodies from the results abstracted from three single sensors using RBF network, to fusewith the improved evidence fusion formula. The way is used to distinguish the failure from thebearing, including the fault on internal ring, external ring, rolling element and bearing cage. Thefusion results show that fault diagnosis, based on the neural network and evidence theory,improve the effectiveness and accuracy on the result of diagnosis.