Anomaly Detection for Monte-Carlo Simulation Data of Missile Control System

Abstract

In this paper, we propose an anomaly detection method for missile simulation data utilizing a cluster-ing-based algorithm. In the proposed method, to assess the performance of the control system, the pa-rameters related to the missile’s guidance and control are selected from simulation data for the auto-mated detection of anomalous cases. Then, the selected data is sampled at fixed intervals to match the analysis scope and reject unnecessary information. The feature points for each parameter are cho-sen in a way that emphasizes abnormal patterns in time series data. Finally, the feature vector of each data sample is formed by combining all feature points. In data analysis, DBSCAN, an unsupervised clustering algorithm, is applied to detect abnormal simulation data in the feature space. In this work, the proposed method was tested with a 6DOF surface-to-surface guided-missile simulation dataset. The results indicate that the proposed method can be used to automatically identify simulation data that reflects abnormal behavior in the control history, which implies the method’s potential to assist the process of finding critical simulation data without predefined criteria or training data.