Research On The Elevator Fault Diagnosis Method Based On Data Fusion

The elevator’s composition is very complex. While bring convenience topeople’s daily life, the elevator will inevitably appear all kinds of faults. There aremany cases have taken place that passengers lost their lives because of elevator faultsin the past two years. Frequent elevator safety accidents catch the elevator safetyregulators, industry associations and manufacturer’s deep thinking. People also putforward higher requirements for the safety of using the elevator. Therefore, it is verymeaningful to do the research in how to diagnosis the fault of the elevator effectively.In the paper, the method of elevator fault diagnosis is studied based on elevatorjerk fault. Firstly, according to the features of the elevator jerk fault, a BP neuralnetwork classifier of fault diagnosis was designed, and the performance of theclassifier was compared under different training functions. On the BP neural networkis easy to fall into local minimum value when be trained, the genetic algorithm wasused to optimize the initial training weights and thresholds of BP neural network. Theperformance of optimized GA-BP neural network classifier was tested. Secondly, theexisting feature parameters of elevator jerk fault were optimized from the perspectiveof optimizing feature parameters of elevator jerk fault. The method of clusteringanalysis was first used to confirm that redundant features were existed in the currentfeature parameters of elevator jerk fault. Then the main feature parameters wereconfirmed from the perspective of new message feature vector and the variation ofthe feature parameters, and the redundant features were removed. Finally, a singlesample feature data was used to conduct a diagnosis may appear wrong judgment dueto noise interference and other factors in the process of practical application. Themethod of using multiple sampling feature data to conduct a diagnosis was proposedin this paper. Through the principal component analysis method, a GA-BP classifierwas designed at the case of multiple sampling feature data to conduct a diagnosis andthe number of its inputs was simplified.In the paper, all of the proposed optimization methods were validated by usingMATLAB software based on data samples of elevator jerk fault. Results show that theselected optimized feature parameters can be accurately realize elevator jerk fault diagnosis, and the performance of optimized elevator jerk fault diagnosis classifierwas also improved. At the same time, adopt the method of multiple sampling data toconduct a diagnosis, can make the elevator jerk fault diagnosis results more reliable.