Study On Assessment Method Of Bearing Performance Degradation Based On Transfer Clustring

With the continuous development of technology,the requirements for modern complex machinery and equipment are more stringent.Due to long-term continuous work under high load and high speed,the key component in the rotating machinery and equipment-bearings,the failure rate is very high.The performance of the bearing can gradually pass through a series of different degradation until complete failure.The service life of most bearings is not long.Once the operating time exceeds the service life,its operating accuracy will drop sharply,which will cause the mechanical equipment to malfunction.Therefore,in order to ensure the safe,stable,and efficient operation of mechanical equipment,this paper conducts a bearing performance degradation assessment study.This method can predict the future operating trend of mechanical equipment,thereby helping equipment to be maintained in a timely manner.So it is very important to achieve intelligent maintenance.In actual production,the bearing is compared with other mechanical parts,and the life is more discrete.Even for bearings produced by the same technology and process,the operating life is different.So when a new type of part is used and its life needs to be predicted,the monitoring data available is very rare.And the cost of obtaining a large amount of monitoring data is high and the Time to get data is long.In this case,the accuracy of the performance evaluation and prediction results is difficult to guarantee.Actually,with the advent of the Big Data era,more and more monitoring data are accumulated in the process of rotating machinery health monitoring.However,due to the special nature of bearing performance degradation,these historical data are related but not similar.So they cannot be used directly.In consideration of this background,this paper proposes an adaptive transfer spectrul clustering algorithm(ATSC-MDK)based on the manifold distance kernel,which can effectively improve the problems of insufficient data or knowledge in performance degradation evaluation.This paper takes rolling bearings as the research object,first pre-processes its vibration signals,then synthesizes multi-domain characteristic indicators,and screens out feature sets that are sensitive to performance state transitions and can accurately describe the trend of degradation,and finally uses the local linear embedding(LLE)method to reduce Dimension,structural degradation feature space;Secondly,according to the characteristics of bearing data,this paper proposes an adaptive transfer spectrul clustering algorithm(ATSC-MDK)based on the manifold distance kernel.Due to the bearing samples often show a non-Gaussian distribution,when the neighborhood information and spatial distribution of the samples are not fully considered,clustering problems often occur.Therefore,the manifold distance and shared neighbor method are used to adjust the similarity matrix of spectral clustering to improve the accuracy of clustering.This paper introduces transfer learning and uses source domain knowledge to guide target domain clustering to improve the lack of data in performance degradation evaluation.Finally,the performance degradation assessment of rolling bearings is carried out,the cluster center of normal training samples is obtained by using the ATSC-MDK algorithm.The method described has been validated on the standard IEEE PHM 2012 bearing full life data set.