As an essential feature of smart cyber-physical systems (CPSs), self-healing behaviors play a major role in maintaining the normality of CPSs in the presence of faults and uncertainties. It is important to test whether self-healing behaviors can correctly heal faults under uncertainties to ensure their reliability. However, the autonomy of self-healing behaviors and impact of uncertainties make it challenging to conduct such testing. To this end, we devise a fragility-oriented testing approach, which is comprised of two novel algorithms: fragility-oriented testing (FOT) and uncertainty policy optimization (UPO). The two algorithms utilize the fragility, obtained from test executions, to learn the optimal policies for invoking operations and introducing uncertainties, respectively, to effectively detect faults. We evaluated their performance by comparing them against a coverage-oriented testing (COT) algorithm and a random uncertainty generation method (R). The evaluation results showed that the fault detection ability of FOT+UPO was significantly higher than the ones of FOT+R, COT+UPO, and COT+R, in 73 out of 81 cases. In the 73 cases, FOT+UPO detected more than 70% of faults, while the others detected 17% of faults, at the most.

Cyber-physical systems, Model execution, Reinforcement learning, Self-healing, Uncertainty
Software Quality Journal
Department of Systems and Computer Engineering

Ma, T. (Tao), Ali, S. (Shaukat), Yue, T. (Tao), & Elaasar, M. (2019). Testing self-healing cyber-physical systems under uncertainty: a fragility-oriented approach. Software Quality Journal. doi:10.1007/s11219-018-9437-3