Neuro-fuzzy modeling in model-based fault detection, fault isolation and controller reconfiguration

Project members: R. Hallouzi, S. Kanev, V. Verdult, R. Babuška, J. Hellendoorn, M. Verhaegen

Sponsored by: STW

The aim is the development of fast and reliable algorithms for fault detection and diagnosis (FDD) and controller reconfiguration (CR). In control systems, faults are events that could cause unwanted behavior or a catastrophe of the controlled system. The design of FTC systems has therefore the purpose to prevent the degradation from simple faults into serious system failures, since system failures might lead to huge economical and human losses. A fault-tolerant system consists of two main parts (see Figure 7): one that has the task to detect and diagnose faults that occur in the control system, and another that reconfigures the controller accordingly, whenever faults occur in the system, so that the performance of the reconfigured faulty closed-loop system is preserved at some desired level.

Figure 7: Fault-tolerant control system

The goal in the project is the development of numerically fast and robust algorithms for on-line implementation, applicable to the problems of FDD and CR in cases of both abrupt and incipient system faults in the sensors, actuators and physical parameters in the system. The project is subdivided into two work packages, one dealing with fault detection and isolation (researcher R. Hallouzi, started in 2004), and another focused on the problem of controller reconfiguration (researcher S. Kanev, 1999-2003).

Within Work Package I the main focus is put on the following items:

• the augmented Kalman filter for the estimation of multiplicative and additive sensor and actuator faults,
• LPV based FDI for dealing with non-linear systems.
• FDI methods that provide information on the uncertainty of the identified faults.
• evaluation of FDI methods on a non-linear aircraft model that may include component faults.

Within Work Package II different approaches have been developed:

• FTC based on multiple-model estimation and predictive control,
• reconfiguration strategies for robust LQ regulator/Kalman filter,
• a BMI approach to passive FTC,
• an ellipsoid algorithm for probabilistic robust controller design,
• active LPV-based FTC in the presence of uncertainty in the FDI,
• a randomized approach to robust output-feedback MPC.