A Sensor-Based Online Diagnostic Approach for Physical Systems


This technology addresses the problem of designing and implementing online monitoring and diagnosis systems for complex hybrid systems by focusing on faults that occur in plant components and contains models where faults are represented by changes in system parameters. Hybrid systems are common in the avionics, spacecraft, automotive and robotics domains where behavior is characterized by continuous plant dynamics and discrete supervisory control. These hybrid systems require a tool that analyzes and seamlessly integrates multiple system models on a discrete and continuous basis. As a result, tasks like monitoring, fault diagnosis and control require model selection and switching to be performed online as system behavior evolves.


This technology focuses on faults that occur in plant components. Research- ers have developed parameterized models where faults are represented by changes in system parameters. This model-based approach to hybrid diagnosis involves three primary tasks: (1) A unified topological modeling methodology based on hybrid bond graphs (HBGs), from which models (state space and temporal causal graph) geared toward the different steps of the diagnosis task are derived. The modeling work is supported by the FACT tool-suite. (2) Observers based on hybrid automata and Extended Kalman Filters for tracking system behavior online, and robust fault detectors that quickly and reliably identify discrepancies in observed system behavior. The innovation in this work is that models are generated online from HBGs when mode changes occur. (3) Fault isolation and identification schemes to determine the faulty component or parameter and the magnitude of the change. The innovation in this work is the combined qualitative and quantitative reasoning schemes for isolating parameter deviations across model changes using the TCG, and estimation techniques for computing fault magnitudes.

Gabor KarsaiSherif AbdelwahedJanos Sztipanovits
Licensing manager: 
Peter Rousos

Featured Video

Vanderbilt Patent Activity

View Vanderbilt University Patents

CTTC on Twitter