State Tracking and Fault Diagnosis for Dynamic Systems Using Labeled Uncertainty Graph.

Area of Science:

• Cyber-physical systems engineering
• Automated control systems
• Reliability engineering

Background:

• Increasing complexity in cyber-physical systems (CPS) heightens vulnerability to failures.
• Accurate system dynamics tracking and fault diagnosis are critical for CPS reliability.
• Probabilistic models are essential for representing the uncertainties in CPS dynamics.

Purpose of the Study:

• To develop an efficient state estimation method for dynamic systems modeled as concurrent probabilistic automata.
• To enhance fault diagnosis capabilities in complex cyber-physical systems.
• To improve the reliability and safety of autonomous systems.

Main Methods:

• Utilized the Labeled Uncertainty Graph (LUG) method for state tracking and fault diagnosis.
• Employed Monte Carlo techniques to sample belief state probability distributions.
• Introduced an innovative look-ahead technique to address sample impoverishment and recursively generate likely belief states.

Main Results:

• Developed an efficient state estimation algorithm for probabilistic dynamic systems.
• The algorithm incorporates roll-forward (state estimation and fault identification) and roll-backward (trajectory analysis) processes.
• Demonstrated effectiveness on a spacecraft power supply control unit.

Conclusions:

• The proposed method provides efficient state estimation and fault diagnosis for complex probabilistic systems.
• The look-ahead technique effectively mitigates sample impoverishment in Monte Carlo simulations.
• The approach is validated for real-world applications in critical infrastructure like spacecraft.