Diagnosis of Safety Incidents for Cyber-Physical Systems: A UAV Example

Abstract

As capabilities of cyber-physical systems (CPS) increase, the interaction of software and physical components becomes more complicated. When a CPS encounters an incident, the increased complexity makes diagnosis a challenging task for traditional diagnostic approaches. To overcome this problem, we split the diagnostic procedure into three steps, namely: (1) type causality, (2) detection and (3) actual causality analyses. We then utilize various technologies to automate each step. Fault trees are extracted from the four variable model of a CPS. This results in modular and human-readable fault trees. Moreover, CPS logs are mapped to the instances of the fault tree nodes using time series analysis techniques. Through examples of unmanned aerial vehicles (UAV), we demonstrate that our framework can diagnose a wide range of scenarios including software, sensor, and actuator failures.