Abstract
Embedded systems increasingly rely on complex interactions between discrete control logic and continuous physical processes. This paper introduces a framework that bridges the gap between these domains by integrating two formal verification tools: Uppaal TRON for timed discrete behavior and PHAVer for continuous dynamics. By employing a parallel architecture, the framework enables real-time monitoring and testing of hybrid systems, ensuring correctness across both precise timing constraints and evolving physical signals.
Introduction
Embedded systems are rarely isolated; they operate within a physical environment defined by continuous variables such as temperature, pressure, or velocity. Traditional testing frameworks often struggle to handle this duality. Timed Automata (TA) are effective for modeling discrete events and clock constraints but lack the capacity for continuous dynamics. Conversely, Hybrid Automata (HA) excel at differential equations but suffer from state-space explosion when faced with complex discrete logic.
Existing solutions often force a compromise—either approximating physics with discrete steps or using deterministic simulations that ignore real-world noise and non-determinism. This research addresses the challenge of verifying systems where sophisticated discrete logic and continuous trajectories must be validated simultaneously.
Methodology: Parallel Tool Architecture
The proposed framework utilizes a "parallel tool" architecture to leverage the specialized strengths of existing verification engines.
- Uppaal TRON: Serves as the primary test engine, managing discrete I/O and real-time constraints.
- PHAVer: Functions as a specialized monitor, performing reachability analysis on Linear Hybrid Automata (LHA) to ensure continuous signals remain within safe, mathematically defined bounds.
A Hybrid Adapter software layer synchronizes these tools, translating physical sensor data into symbolic bounds for Uppaal and feeding timed event scripts into PHAVer. This setup allows for the verification of conformance relations such as rtioco (relativized timed input/output conformance) and hioco (hybrid input/output conformance).
Modeling Patterns and Innovations
A core innovation of this work is the "Divide and Conquer" modeling approach. Instead of a single monolithic model, requirements are split: a TA model handles the "when" (timing/logic), while an LHA model handles the "how much" (dynamics).
Key features include:
- Symbolic Over-Approximation: Rather than checking for exact values, PHAVer computes a "region" of valid values. This "symbolic cloud" naturally accounts for sensor noise and measurement imprecision.
- Non-Deterministic Testing: The framework supports "fuzzy" specifications (e.g., an event occurring within a 7-14 second window), verifying system behavior across the entire interval rather than a single point.
Case Study: Danfoss EKC204A
The framework was evaluated using the Danfoss EKC204A electronic cooling controller. The study focused on the controller's temperature estimation algorithm, which employs a PID-like filter to stabilize sensor data.
During testing, the framework successfully tracked the temperature region as it evolved. When new temperature data was injected, PHAVer recomputed the future "valid region," confirming that the controller's internal estimate converged correctly within the expected 15-second window. This demonstrated the framework's ability to handle real-world industrial components in real-time.
Results and Limitations
The integration proved successful, providing a sound testing pipeline where failures reported by the framework definitively indicate violations in the Implementation Under Test (IUT).
However, performance remains a challenge. The high computational cost of PHAVer’s state set updates currently limits its role to a monitor rather than a real-time stimulator. Occasional lags occur during complex polyhedral computations for reachability analysis.
Conclusion and Future Work
This research demonstrates that parallelizing discrete and continuous verification tools is a viable path for testing complex embedded systems. Future work will focus on optimizing the PHAVer engine and implementing new monitoring operators based on time-discretization to reduce lag. Ultimately, the goal is to enable PHAVer to function as a trajectory generator for more advanced "what-if" testing scenarios.
Summary updated