| The advanced equipment of modern industry keeps developing towards complex,high-speed, efficient, lightweight, large-scale direction, with the development ofscience and technology. Coupling and convergence between the device componentsare very close, and when a component fails, the entire device, or even the entireproduction line will be affected, which will bring huge economic losses, and evencause unnecessary casualties. But, many of the components defects are not fatal fornormal operation of the equipment, or it will be understood well, if we describe it asthis: once some certain components fails, the equipment is in the state between normaland failed state, called “sub-healthy†state. If the equipment in “sub-healthy†stateruns as normally without maintenance, that will lead huge economic losses, evenunexpected serious damage. Thus, the diagnostic of the “sub-healthy†state is theurgent problems of mechanical equipment fault diagnosis and maintenance.This paper focuses on the “sub-healthy†state diagnosis of rolling bearing, andthe main works are as follows: Firstly，we characterized the bearing failure andanalyzed the applied fields of current intelligent fault diagnosis technology. On thisbasis, we learned more about the “sub-healthy†state of bearings, which is about howto determine the “sub-healthy†state based on the remaining useful life of bearings.Second, feature reduction, as a necessary step, affects significantly on the patternrecognition problem. Taking the reality, that it is hard to find out the best combinationof features, into account, this paper used the improved artificial immune algorithm toselect the optimal combined features, which is applied to the rolling bearing data sets.And it showed its advantages on feature selection, compared with other algorithms.Finally, a “sub-healthy†recognition model is established based on the support vectormachine whose parameters are optimized by an improved ant colony optimizationalgorithm. SVM is good for pattern recognition of small sample, compared with otherrecognition algorithms. But, the performance of SVM is influenced significantly by itsparameters. We used the improved ant colony algorithm for parameters search, inorder to avoid the problem that large time consumption and difficult to find theoptimal solution.Bearing data collected from PRONOSTIA is applied to verify the method inmatlab platform. Experimental results show its better performance in diagnostic and generalization ability, compared with genetic algorithm and cross-validation method. |
