Model-based Fault Classification for Automotive Software
summary
Model-based algorithm replays failing CAN traces symbolically to localize the exact divergence, computes minimal subsets of messages that suffice to cause the fault using weakest preconditions and a Hoare-logic style proof, then groups failures by equality of satisfied timing atoms so small timing shifts are allowed but 50ms guards are respected. Implementation uses symbolic execution and Hoare logic, handles clock variables, is prototyped in C# against CAN bus Unified Diagnostic Services, and clusters thousands of failures into short templates (often two events).
AI-written summary · about
chapters (13)
- 0:00Fault classification in automotive software
- 2:51Manual trace analysis challenges
- 4:58Formal model requirements
- 7:49Automata with memory and clocks
- 11:18Fault localization via symbolic execution
- 14:54Reducing irrelevant fault traces
- 17:40Fault classification with time wiggling
- 21:10Equivalence via satisfied guards
- 23:16Implementation results and feedback
- 25:11Future challenges
- 28:15Value mismatches in classification
- 29:41Testing failures insights
- 30:50Automotive specs limitations
related talks
22mFormal Runtime Error Detection During Development in the Automotive Industry
19mCAFault: Enhance Fault Injection Technique in Practical Distributed Systems via...
11mCausal testing: understanding the root causes of defects
27mLet's Bring System Dynamics Back to CS!
15mPAFL: Enhancing Fault Localizers by Leveraging Project-Specific Fault Patterns
16mA Search-Based and Fault-Tolerant Approach to Concurrent Model Synchronisation
21mBounded-Exhaustive Subspace Diversification for SMT Solver Testing
43mOmg, how do I write software that isn't a ticking timebomb!?