Equipment Health Condition Monitoring Technologies Using Affinity Propagation Clustering Algorithm

With the development of science and technology, machinery and equipment becomemore complex, any malfunction or failure in the operation of the machine will causesignificant economic losses. The promotion and application of equipment conditionmonitoring and fault diagnosis technology can not only protect the safe and reliable operationof the equipment to prevent catastrophic accidents. This paper, which is using the rollingbearing as the research object, studies the equipment health condition monitoring techniquesbased on Affinity Propagation clustering.This paper studies signal preprocessing method such as time domain indexes, frequencydomain indexes and wavelet packet transform. The experiments show that: when the damageof rolling bearings appeared, both of time domain indexes and frequency domain indexes willchange, and they are different in different types and degrees of the damage. Accordingly,extracting the vibration signal with time domain indicator, frequency domain indicator orwavelet packet indicator can reduce the dimension of the vibration signal and can effectivelydescribe the health status of the equipment.In order to describe the equipment health status, the method of health monitoring basedon energy entropy has been studied. First do wavelet packet decomposition of equipmentvibration signal to obtain relative energy of each band, and then get the entropy. Theexperiment result indicates that wavelet packet energy entropy can identify the condition anddegree of damage effectively. It can be used to monitor the changes of the health status ofrolling bearings.In view of the general clustering algorithms need to specify the category numberbeforehand or clustering convergence problems, this paper studies the AP algorithm. Degreeof preference and convergence coefficient have great influence to the clustering results, whenpreference is identical, the convergence coefficient increases, number of iteration also willincrease, if convergence coefficient is too small, convergence may appear oscillation even;under the condition of the same convergence coefficient, degree of preference will affect theresult of cluster analysis, when the degree of preference value is bigger, it can produce moreclassifications, and when the degree of preference value is smaller, there will be owed toclassification. In contrast, degree of preference values has bigger influence on the clusteringresults.In view of the AP algorithm clustering results are still affected by degree of preference value, this paper studies semi-supervised AP clustering algorithm. Semi-supervised clusteringalgorithm does not need to set AP reference value in advance, embedded in the evaluationcriteria of the iterative process effectiveness index, which can supervise and guide theclustering process, producing the optimal clustering results. And compared with thetraditional AP algorithm, semi-supervised AP clustering algorithm is more suitable for thecondition monitoring of the structure.