Uncertainty-Aware Behavior Modeling and Quantitative Safety Evaluation for Automatic Flight Control Systems

Automatic flight control systems (AFCS) are safety-critical systems tightly integrating computation, networking and physical processes. However, the uncertainty resulting from evolving dynamics in cyberspace and the physical world can affect the reliability of decision-making in the controller, threatening the system's safety. How to accurately capture the uncertainty, effectively control the aircraft and improve safety has become an unavoidable challenge for the software industry. To this end, we define an uncertainty-aware modeling language (UAML), which supports modeling the AFCS's dynamic behavior and environmental uncertainty using formal specifications. We use a machine learning-based method to predict the risk levels in operating environments as the representation of uncertainty from the physical world. The prediction result is transferred to UAML as the parameters. On this basis, we present a framework for quantitative safety evaluation using statistical model checking based on UPPAAL-SMC to help AFCS make reliable decisions at runtime. We illustrate our approach by modeling and analyzing a realistic example, and the experimental result demonstrates the effectiveness of our approach.