Research Center for Supervision, Safety and Automatic Control

ITECOTOFA

CICYT ref. DPI2002-03500.

Summary

This project has developed techniques for fault detection, diagnosis, and fault-tolerant control using uncertain models. These techniques have been applied to a pneumatic actuator, a physical-chemical laboratory system, and Barcelona's sewer network. To implement these, fault-tolerant control tools have been developed that allow evaluation of the situation, the degree of recoverability, and corrective actions when faults occur in the system.

In the field of fault detection, various techniques have been used: parity equations, observers, and identification. The first two could be considered direct image detection techniques, meaning they assess if the prediction or simulation of the model aligns with expectations. The third, identification, involves evaluating which set of parameters can drive the system to certain operating conditions and assessing if these parameters match the expected ones, also known as inverse image. Using both techniques independently and applying them to uncertain models has advantages and disadvantages, largely depending on the system type, fault type, and uncertainty type, and results cannot be generalized. As solutions considered in the project, work is being done on both techniques simultaneously, aiming to achieve their complementarity and obtain better detection results.

Fault-tolerant control is an emerging field where no general theory yet exists, and the literature only includes partial contributions to the various problems that must be addressed to add tolerance to a control system. This project has aimed to frame the problem of fault-tolerant control within the field of predictive control with constraints, which has enabled the use of existing tools in this field to assess performance degradation due to faults and to decide what corrective actions to take. Likewise, a fault-tolerant control system can be very naturally modeled as a hybrid system that represents faults as different modes of operation. Currently, there is an entire research line that allows hybrid systems to be integrated very easily with predictive control systems, enabling fault-tolerant control systems to be analyzed with all the analytical tools developed for these systems and control algorithms.

Achieved Objectives

Objective 1. Analysis of observability and degree of isolability

  • An algorithm has been developed to characterize and determine the minimum number of sensors necessary to isolate the maximum number of faults present in a dynamic process. The methodology used for this analysis is based on a structural analysis of the system.
  • To analyze the degree of isolability, a comparative study has been conducted between the technique based on structural analysis and another technique that uses the state-space representation of quasi-stationary systems. The advantages and disadvantages of each technique have been analyzed, and a new field of work is proposed dedicated to their integration.

Objective 2. Analysis of fault propagation and evaluation of recoverability and corrective actions for fault-tolerant control.

  • Algorithms based on constraint satisfaction have been developed that allow for the evaluation of recoverability in predictive controllers with constraints and to decide what type of corrective actions (accommodation/reconfiguration) are most suitable to address the impact of a fault in an actuator.

Objective 3. Improvement of robust detection techniques and their integration with diagnostic techniques.

  • Robust detection and diagnostic algorithms have been developed based on interval observers for linear and nonlinear systems using optimization techniques.
  • Improvement of the integration of fault detection and isolation techniques. Currently, fault detection and isolation algorithms are developed separately, and their integrated development is necessary to improve functionality.
  • Improvement of robust detection using interval methods that combine the direct image with the inverse image. Traditionally, these methods used the direct image of the interval model for detection. Calculating this inverse image is costly (global optimization, uncertainty propagation, etc.). The project proposes to complement this image with the inverse image using consistency-based techniques that are very effective at detecting inconsistencies between the model and experimental data.
  • Finally, it has been studied how the diagnostic system is affected by consecutive faults in the sensors used and how their placement should be designed to minimize the impact on isolation capability.

Objective 4. Robust design of controllers for tolerance

  • Use of Quantitative Feedback Theory (QFT) techniques to design robust controllers in the case of model parameter uncertainty.

Most Notable Publications

    • Travè-Massuyés, L., Escobet, T. Olive X., “Diagnosability Analysis Based on Component-Supported Analytical Redundancy Relations”, Systems, Man and Cybernetics, Part A: Systems and Humans, IEEE Transactions on, vol.36 (6): 1146-1160, <doi:10.1109/TSMCA.2006.878984>, 2006
    • Nejjari, F., Perez, R., Escobet, T., Travé-Massuyès, L., “Fault diagnosability utilizing quasi-static and structural modelling”, Mathematical and Computer Modelling, vol. 45 (5-6): 606-616, <doi:doi:10.1016/j.mcm.2006.06.008>, 2007
    • V. Puig, A. Stancua, T. Escobet, F. Nejjara, J. Queveda and R.J. Patton. “Passive robust fault detection using interval observers: Application to the DAMADICS benchmark problem”, Control Engineering Practice, vol. 14(6):621-633, <doi:10.1016/j.conengprac.2005.03.016>, 2006
    • Puig, V., Quevedo, J., Escobet, T., Pulido, B. “On the integration of fault detection and isolation in model based fault diagnosis”. 15th International Workshop on Principles of Diagnosis (DX’04). France. 2004. http://spiderman-2.laas.fr/DX04/Proceedings-DX04/Articles/129.pdf.
    • Puig, V., Schmid, F., Quevedo, J., Pulido, B. “A new fault diagnosis algorithm that improves the integration of fault detection and isolation”. Submitted to ECC-CDC’05. Seville. Spain.
    • Puig, V., Mrugalski, M., Nejjari, F., Quevedo, J., Korbicz, J. “A GMDH neural network-based approach to passive robust fault detection using a constraint satisfaction backward test", Engineering Applications of Artificial Intelligence, vol. 20(7):886-897, <doi:doi:10.1016/j.engappai.2006.12.005>, 2007.