Badmus, Taofeeq Alabi
(2025)
A novel fault detection and diagnostic petri net methodology for dynamic systems.
PhD thesis, University of Nottingham.
Abstract
Faults in dynamic systems can lead to significant consequences, such as performance degradation, safety hazards, or economic losses. Monitoring, detecting, and diagnosing these faults promptly and accurately is essential. Petri Nets provide a graphical and mathematical tool for modelling and analysing dynamic systems but have limitations in handling feedback control loops, nonlinear dynamics, uncertainties, and multiple faults. This thesis addresses these challenges by proposing an integrated Petri Net methodology for fault detection and diagnosis, combining extended Generalised Stochastic Petri Net (xGSPN) for system operation modelling and modified Bayesian Stochastic Petri Net (mBSPN) for fault diagnosis.
The xGSPN-mBSPN methodology operates as an integrated model. The xGSPN models operational and failure behaviour, establishing causal relationships between component failures and system behaviour. The mBSPN integrates Bayesian Network diagnostic features, enabling early fault diagnosis through conditional probabilities and inference sampling algorithms. This approach traces the paths leading to observed system failures during operation. The methodology is validated using a water tank level control system, demonstrating effectiveness in detecting and diagnosing single and multiple faults. The diagnostic results align with those obtained from Hugin software.
Additionally, the thesis extends the scope of the methodology by detailing steps for adapting the approach to more complex systems, exemplified by a three-phase separator in the oil and gas industry. While the water tank system demonstrates the methodology’s application to a simple yet dynamic system, the three-phase separator case study showcases its potential to address the complexities of multi-phase flow systems with intricate dependencies. This adaptation framework provides a solid foundation for future implementation and validation in larger-scale industrial systems across sectors such as oil and gas, power generation, and manufacturing.
The key contributions of this thesis include the proposition of an xGSPN formalism incorporating new features like time-varying conditional places and transitions for realistic and flexible modelling of system dynamics. The mBSPN formalism integrates Bayesian Network features with Petri Net constructs to diagnose faults based on observed deviations. Efficient algorithms for automatically generating input Conditional Probability Tables (iCPTs) and the system’s fault diagnostic model reduce manual effort, ensuring consistency and accuracy of the developed model.
The thesis further contributes to the body of knowledge by developing a bespoke C++ programming code to implement the xGSPN-mBSPN method. Its application to the water tank system validated the methodology’s effectiveness for monitoring, early fault detection, and diagnosis of dynamic systems with feedback control loops. The extended case study of a three-phase separator illustrates the potential of the methodology to address complexities in multi-phase flow systems, laying the groundwork for its application in industries like oil and gas, power generation, and manufacturing.
In summary, this research advances fault detection and diagnosis methodologies by introducing a scalable and robust PN-based approach capable of addressing the complexities of dynamic systems. The xGSPN-mBSPN methodology paves the way for reliable and efficient system monitoring, early fault detection, and accurate diagnosis, with promising applications in diverse real-world scenarios.
Actions (Archive Staff Only)
 |
Edit View |