Automatic Detection of Feature Interactions Using Symbolic Analysis and Evolutionary Computation

Ensuring acceptable and safe behavior is paramount for high-assurance systems. However, independently-developed features often exhibit overlapping, yet conflicting behavior termed feature interactions. This paper introduces Phorcys, a design-time approach for detecting unwanted failures caused by n-way feature interactions at the requirements level using both symbolic analysis and evolutionary computation. Unlike previous n-way feature interaction detection approaches that look for each unique unwanted interactions, Phorcys analyzes each feature for its ability to cause unwanted behavior, including failures. By using a combination of symbolic analysis and evolutionary computation, Phorcys is able to identify multiple counterexamples, thus providing more guidance for mitigation (e.g., revising specifications, adding constraints, etc.). To the best of the authors' knowledge, Phorcys is the only technique to detect failures caused by n-way feature interactions using a combination of symbolic analysis and evolutionary computation. We illustrate our approach by applying Phorcys to an industry-based automotive braking system comprising multiple subsystems.