A Statistical Anomaly-Based Algorithm for On-line Fault Detection in Complex Software Critical Systems

The next generation of software systems in Large-scale Complex Critical Infrastructures (LCCIs) requires efficient runtime management and reconfiguration strategies, and the ability to take decisions on the basis of current and past behavior of the system. In this paper we propose an anomalybased approach for the detection of online faults, which is able to (i) cope with highly variable and non-stationary environment and to (ii) work without any initial training phase. The novel algorithm is based on Statistical Predictor and Safety Margin (SPS), which was initially developed to estimate the uncertainty in time synchronization mechanisms. The SPS anomaly detection algorithm has been experimented on a case study from the Air Traffic Management (ATM) domain. Results have been compared with an algorithm, which adopts static thresholds, in the same scenarios [5]. Experimental results show limitations of static thresholds in highly variable scenarios, and the ability of SPS to fulfill the expectations