Hierarchical Clustering Algorithm for Anomaly Detection on Intelligent Production Line

Abstract

The technology of “intelligent manufacturing” has been developing rapidly in recent years. The intelligent production line, as the bearer of “intelligent manufacturing”, has promoted the intelligent development of the manufacturing field. However, in the process of intelligent production line operation and maintenance, the sensing data will change with the change of equipment status. Only by accurately identifying abnormal data and prioritizing its transmission to the monitoring equipment can we quickly sense the status of the equipment and make its maintenance plan. Aiming at the problems of high parameter sensitivity and influence by the shape of data samples when we use traditional clustering algorithms to identify abnormal data, this paper proposes a hierarchical clustering algorithm H-DBSCAN for anomaly detection on intelligent production lines. The algorithm is based on the KNN algorithm to obtain the optimal parameters to reduce the parameter sensitivity. And the multi-density clustering of data is accomplished by using multi-layer density for noise clustering with a reasonable fusion of small clusters of linked weights. The experimental results show that the H-DBSCAN algorithm can reduce the influence of input parameters and sample distribution on clustering results, maximize the accuracy of clustering, meet the needs of intelligent production lines for efficient detection of abnormal data, and achieve full automation of data analysis.