A Novel Fault Diagnosis Method Via Data Dynamic Property Inspired Density-Based-Clustering

State recognition is the crucial stage in the fault diagnosis, where clustering technique usually has abilities to provide effective support, especially under the condition of lacking prior knowledge. In this paper, a dynamic property of data points is proposed from the perspective of movement. Based on this favorable property we develop an effective density-based algorithm by defining a cluster with new connection rule. The property sustains our algorithm on extracting clusters with widely variation in density from datasets, and gain more procedural conciseness in our algorithm by eliding the need of density threshold. The capability of our method is validated by experimental evaluation on both synthetic benchmark datasets and real-world rolling bearing vibration data, which shows certain advantages of our method over conventional and recent algorithms.