Project Leader: BELKHERIA Kamel
Objectives
The strengthening of failure prevention policies tends to raise awareness across all sectors, from the scientific training of individuals to industrial production processes. In Tunisia, numerous decrees aim to promote and implement guidelines for the monitoring of production systems with a view to improving their safety.
In line with this policy, our proposal focuses on enhancing system monitoring by training young researchers through the exploration of new tools and methods in the field of system dependability (safety, reliability, availability, and maintainability).
The main objectives of this project are to:
- Develop state estimation and diagnostic methods for poorly known systems, i.e., systems with ill-defined models and uncertain measurements.
- Establish a coherent scientific framework for the integrated design of dependable systems, by unifying the underlying concepts.
- Optimize existing methods and develop new approaches capable of meeting the increasing dependability requirements of increasingly complex systems.
Summary
Dependability is, by nature, interdisciplinary and covers a broad spectrum, both in terms of methods and areas of application. Characterizing a system’s ability to provide a specified service, dependability is formally defined as the “quality of service delivered by a system such that users can place justified confidence in it.”
A dependable system avoids or eliminates danger and maintains its operation in a failure-free state, ensuring a maximum level of confidence in its performance.
This project is structured around four main actions, described below.
Action 1: System Fault Diagnosis through Data Analysis without A Priori Behavioral Models
Conventional methods used to automate the monitoring of complex systems generally fall into two main categories:
- Model-based approaches (internal methods), which rely on behavioral models built from the physical description of the system or expert knowledge.
- Data-driven approaches (external methods), which assume that the only available knowledge of the system comes from its past and present observations. These approaches do not rely on any explicit cause–effect model but instead exploit measured signals collected from the system under supervision.
For model-based methods, diagnostic performance—fault detection and localization—depends directly on the quality of the model used. To overcome the limitations related to model accuracy, an alternative is to use data-driven methods, which identify relationships (often linear) between system variables without explicitly formulating the model.
These methods make it easier to consider fault detectability and isolability criteria, and they are particularly suitable when detailed physical models are unavailable.
Action 2: Diagnosis of Nonlinear Systems Using Multi-Model Approaches
Fault diagnosis plays a key role in many application domains, such as industrial process monitoring or satellite autonomy.
Model-based fault detection methods relying on linear models have reached a certain level of maturity after decades of research. However, the assumption of linearity limits their relevance for complex real-world systems. Extending linear-model-based techniques to nonlinear systems is challenging.
An effective alternative is the multi-model approach, which represents the system using a set of simple local models, each describing behavior in a particular operating region (defined, for instance, by input or state variables). The global model is then constructed by interpolating between these local linear models.
This approach has already yielded promising results and allows for more realistic fault detection and isolation in nonlinear contexts.
Action 3: Diagnosis and Fault Tolerance for Automated Systems
This action is based on an application-oriented architecture that integrates, in addition to nominal system functions, modules for fault detection, localization, and diagnosis, as well as for mode change detection (particularly those caused by environmental variations). It also incorporates prognostics, fault accommodation, and control reconfiguration mechanisms, ensuring the desired level of system reactivity and resilience.
The overall fault management process, known as FDIR (Fault Detection, Isolation, and Recovery) or FTC (Fault-Tolerant Control), includes mechanisms aimed at maintaining dependability.
A fault-tolerant system is characterized by its ability to maintain or recover performance (both dynamic and static) close to that of normal operation, even in the presence of faults.
Classical approaches, known as passive fault tolerance, are often derived from robust control techniques. In contrast, active fault-tolerant approaches—which integrate a diagnostic module (FDI)—allow the control law to adapt once a fault is detected and localized, either by adjusting parameters or modifying the control structure.
However, few studies have considered time delays associated with the computation of corrective control actions. During this delay, the faulty system continues to operate under nominal control, which may lead to performance degradation or instability. This project seeks to address this limitation.
Action 4: Analysis of Uncertainties in the Performance Evaluation of High-Integrity Safety Systems
In the design phase, major progress has been achieved in reducing risks in hazardous situations through the implementation of active safety systems. This relies on the use of reliability databases, the consideration of influence factors, and the propagation of uncertainties.
A key challenge is to account for uncertainties in component reliability data when evaluating system dependability. This is addressed using fuzzy set theory, possibility theory, or evidence theory.
The primary focus of these studies is on Safety Instrumented Systems (SIS), where dependability requirements are of utmost importance.
Dependability performance analysis for high-integrity protection systems can be conducted using Markov models, which formally represent system states as functions of events (failure, test, maintenance, etc.) and parameters (failure rate, maintainability, common cause failures, and so forth).