Equipment Health Condition Monitoring Technology Using Adaptive Mean Shift Clustering

With the development of science and technology, machinery and equipment becomemore complex,and the level of automation becomes higher.As the role and impact ofmachinery and equipment in modern industrial production are growing, the costs relatingthereto it are getting higher and higher.Any malfunction or failure in the operation of themachine will cause significant economic losses and may even result in casualties. Therefore,there should be a timely manner to monitor the malfunction of the equipment to make it worksafely and economically. This paper, which is using the rolling bearing as the research object,studies the equipment health condition monitoring techniques based on Adaptive Mean Shiftclustering.Signal preprocessing method such as time domain indexes, frequency domain indexesand wavelet packet transform has been studied. 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. In addition,aftervibration signal was decomposed by wavelet packet transform, different types of damageexhibit different characteristics of energy distribution. Accordingly, extracting the vibrationsignal with time domain indicator, frequency domain indicator or wavelet packet indicator canreduce the dimension of the vibration signal and describe the damage condition of differenttypes effectively.The method of health monitoring based on energy entropy has been studied. 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.The theory of Mean Shift clustering is discussed firstly.It is proved by experiments thatthe selection of kernel function, kernel radius and threshold will influence the performance ofMean Shift clustering. The kernel function affects the accuracy and the numbers of iterationof the cluster. For each kernel function, there is a reasonable range of kernel radius for theclustering algorithm.The clustering performance become worse when kernel radius is beyondthe range.Moreover reducing the threshold is helpful to improve the efficiency and accuracyof the algorithm.The principles of Adaptive Mean Shift algorithm (AMS) has been discussed. Theselection of kernel function, initial kernel radius and the number of iterations can influence the performance of AMS. The choice of initial kernel radius has a great impact on theclustering when using Gaussian kernel function.While using Epanechnikov kernelfunction,the choice of initial kernel radius has less impact on clustering.But compared withthe former it has a lower classification accuracy rate.In addition, increasing the numbers ofiteration can improve the performance of clustering. Compared with the Mean Shift, the AMShas a better clustering performance.A structural health monitoring method based on Adaptive Mean Shift centroid has beenproposed.This method which uses the centroid of the health state as a reference,identify theoccurrence of damage and the degree of injury with the offset between the centroid of a stateand the reference. The experimental work shows that the farther the centroid away from thereference centroid, the more serious structural damaged. Therefore,offset of the centroid caneffectively assess the health condition of the structure.