The rapid growth of control cyber-physical systems (control-CPS) complexity necessitates the use of fault localization tools to locate cyber subsystems' bugs. Many mainstream fault localization tools need large number of labeled ("correct" or "faulty") execution traces as their inputs. To prepare the traces, for most control-CPSs (particularly those large or without complete/accurate models), assertions designed by domain experts are used. This approach is best-effort: often heavily subjective and ad-hoc. To make the approach more principled and systematic, we exploit the state-of-the-art hybrid systems modeling and stability theories to propose a new approach. Empirical evaluations upon commercial-product-grade large control-CPS platform show that our proposed approach achieves significant improvements (42% ~ 300% improvement in accuracy, recall, and latency medians) over the existing approach.
|Date of Award||2017|
- The Hong Kong Polytechnic University
|Supervisor||Qixin Wang (Chief supervisor)|