Model-Based and Intelligent Fault Diagnosis

Model-Based and Intelligent Fault Detection and Isolation: Residual Generation, Observers and Industrial Case Studies. Seminar of two hours for the University of Science & Technology Beijing - USTB.
 

Abstract

• Fault detection and isolation play a crucial role in ensuring the safe and reliable operation of modern engineering systems, where incipient faults, strong coupling, and tight performance requirements demand systematic monitoring approaches. This contribution presents a tutorial overview of model-based and intelligent methods for fault detection and isolation, with emphasis on residual generation and evaluation. After recalling basic definitions, fault types and the main tasks of fault diagnosis, the talk reviews analytical redundancy principles and the construction of residual signals from state-space process models. Unknown input observers and Kalman filters are then introduced as powerful tools for robust residual generation in the presence of disturbances and modelling errors. Residual evaluation and change detection are discussed through statistical tests, minimum detectable fault analysis and reliability assessment. Knowledge-based and data-driven techniques are also considered, focusing on neural networks used for residual evaluation and fault classification within a model-based framework. The ideas are illustrated by case studies from industrial and aerospace applications, including a gas turbine prototype, a small commercial aircraft and a spacecraft attitude-control system. These examples show how residual design, fault isolability analysis, and performance indices can be combined to quantify sensitivity, false alarms, and isolation capability. The talk concludes with a discussion of current challenges and trends, including robustness to uncertainty, incipient fault sensitivity, the link to fault-tolerant control and the growing role of hybrid model-based and data-driven approaches.

Keywords

• Fault detection and isolation, model-based fault diagnosis, analytical redundancy, unknown input observers, Kalman filtering, neural networks, industrial processes, aerospace systems.

Main Points

• Overview of the basic concepts, classifications and tasks of fault detection and isolation in safety-critical industrial systems;
• Presentation of analytical redundancy and residual generation using state-space models, unknown input observers and Kalman filters;
• Discussion of residual evaluation, change detection, minimum detectable faults and reliability assessment for robust monitoring;
• Integration of neural networks and other intelligent techniques to support residual evaluation and fault classification within a model-based framework;
• Industrial and aerospace case studies (gas turbine, aircraft and spacecraft) demonstrating fault sensitivity, isolability and performance indices;
• Open issues and future directions, including robustness to uncertainty, incipient faults, fault-tolerant control and hybrid model-based/data-driven strategies.

Downloads: Talk Slide Files

• Seminar Slides (PDF file, 23 MB).

Selected References

• Isermann, R. (2005). Model-based fault-detection and diagnosis - status and applications. Annual Reviews in Control, 29(1), 71-85.
• Chen, J., & Patton, R. J. (1999). Robust Model-Based Fault Diagnosis for Dynamic Systems. Kluwer Academic Publishers.
• Simani, S., Fantuzzi, C., & Patton, R. J. (2003). Model-Based Fault Diagnosis in Dynamic Systems Using Identification Techniques. Springer.
• Isermann, R. (2011). Fault-Diagnosis Applications: Model-Based Condition Monitoring: Actuators, Drives, Machinery, Plants, Sensors, and Fault-Tolerant Systems. Springer.
• Yan, W., Wang, J., Lu, S., Zhou, M., & Peng, X. (2023). A review of real-time fault diagnosis methods for industrial smart manufacturing. Processes, 11, 369.

Previous Page

Simani Home Page

Dipartimento di Ingegneria