Denis Berdjag

A general fault tolerant linear quadratic control strategy under actuator outages

In many applications, faults of interest are associated with a known set of hardware configurations. Indeed, actuator outages, sensor losses, and more generally, any kind of fault when the reconfiguration strategy is used, result in only a subset of components being available for achieving the system mission, and therefore the hardware configurations of interest are defined by the power set of the components whose faults are considered. Whatever the design objective (stability, disturbance attenuation, H ∞ or H 2 optimality, etc.), a passive scheme designs one single control law that guarantees stability and performances for the set of all N configurations, while an active scheme designs N control laws, each of them being dedicated to one configuration. Considering the case of actuator outages, this article introduces a general frame where several controllers are designed, each of them being dedicated to a subset of faults. This frame includes the classical passive and active strategies, which are respectively associated with the single set of all faults and with N sets of single faults, and extends the reliable control strategy where only one subset of faults is considered. This approach is also applicable when all faults are not recoverable. Its performance evaluation and optimisation is carried out by setting a decision frame in which the trade-off between the Fault Tolerant Control (FTC) complexity and the probability to face a non-recoverable situation is explicitly addressed.

Fault detection and isolation for redundant aircraft sensors

A method for failure detection and isolation for redundant aircraft sensors is presented. The outputs of the concerned sensors are involved in the computation of flight law controls, and the objective is to eliminate any perturbation before propagation in the control loop when selecting a unique flight parameter among a set (generally 3) of redundant measurements. The particular case of an oscillatory failure is investigated. The proposed method allows an accurate fault detection and isolation of erroneous sensor and computes a consolidated parameter based on the fusion of data from remaining valid sensors. The benefits of the presented method are to enhance the data fusion process with FDI techniques which improves the performance of the fusion when only few sources (less than three) are valid.

An evidential network approach to support uncertain multiviewpoint abductive reasoning

The paper proposes an approach to support human abductive reasoning in the diagnosis of a multiviewpoint system. The novelty of this work lies on the capability of the approach to treat the uncertainty held by the agent performing the diagnosis. To do so, we make use of evidential networks to represent and propagate the uncertain evidence gathered by the agent. Using forward and backward propagation of the information, the impact of the evidence over the different symptoms and causes of failure is quantified. The agent can then make use of this information as additional hints in an iterative diagnosis process until a desired degree of certainty is obtained. The model is compared with a deterministic one in which evidence is represented by binary states, that is, a symptom is either observed or not.

Reducing the Reliability Over-Cost in Reconfiguration-Based Fault Tolerant Control Under Actuator Faults

This note addresses reconfiguration-based fault tolerance under actuator faults, for linear time invariant systems. Based on the concept of bottom-up extensible controls, a fault tolerance scheme that mixes the passive and active approaches is designed to recover all recoverable faults, and an algorithm is proposed to reduce the reliability overcost with respect to previous designs.

Algebraic approach for model decomposition: Application to fault detection and isolation in discrete-event systems

Algebraic approach for model decomposition: Application to fault detection and isolation in discrete-event systems This paper presents a constrained decomposition methodology with output injection to obtain decoupled partial models. Measured process outputs and decoupled partial model outputs are used to generate structured residuals for Fault Detection and Isolation (FDI). An algebraic framework is chosen to describe the decomposition method. The constraints of the decomposition ensure that the resulting partial model is decoupled from a given subset of inputs. Set theoretical notions are used to describe the decomposition methodology in the general case. The methodology is then detailed for discrete-event model decomposition using pair algebra concepts, and an extension of the output injection technique is used to relax the conservatism of the decomposition.

PACT : a passive / active approach to fault tolerant stability under actuator outages

This paper addresses actuator outages, or any actuator fault under the reconfiguration strategy. Based on the lattice structure of the set of system configurations, a frame that guarantees stability by mixing both the passive and active fault tolerance strategies is proposed. The proposed PACT designs a bank of controllers, with minimal number of control laws, and minimal instability during switching.

Evaluation and Optimal Selection of Reliable Control Design Specifications

Abstract In the Linear Quadratic (LQ) control frame, reliable control guarantees stability and performance against actuator faults that belong to a given design specification (DS) set. This paper proposes an evaluation frame and a multicriteria decision process for the optimal selection of DS, which is based on a decomposition of faults that do not belong to this set.

Nonlinear Model Decomposition for Fault Detection and Isolation System Design

Model-based fault detection and isolation methods relies on residuals that must be robust against uncertainties or unknown inputs while remaining sensitive to the faults. Two approaches may be applied to design a residual generator (RG): the first one uses the global model of the system and consider the robustness/sensitivity constraints in the RG synthesis. The second one relies on two steps: First, a structural decomposition of the global system is performed with respect to robustness/sensitivity criteria. Second, residuals generators are designed based on the subsystems generated by the previous decomposition. This paper deals with the second approach. The structural decomposition presented here is achieved through special mathematical tools, namely the algebra of functions. These tools are presented and the computational aspects are deeply investigated. An application on an academic example is provided to illustrate the methodology

An algebraic approach for behavioral model decomposition

Abstract This paper presents a constrained decomposition methodology for discrete-event models. Partial models are obtained and used for model-based fault detection and isolation. The constraints of the decomposition ensures that the resulting partial model is decoupled from a given subset of inputs. The constrained decomposition method is formulated using pair algebra concepts as an iterative algorithm for easy implementation.

An Algebraic Method for Nonlinear System Decomposition

Abstract This paper is about model decomposition with a robustness and a fault sensitivity criteria. The model to decompose is nonlinear. The objective is to obtain the part of the model which is robust to the perturbations and to only one fault, while being sensitive to all other faults. This is especially useful for system monitoring, where the fault isolation and robustness to unknown signals is a major concern. To achieve the decomposition, an algebraic method is used, based on algebra of functions. However, this method, in its original form, will not always reach a solution even if this solution exists. In this paper, a solution is proposed to fix that issue. The initial model is rewritten into an adequate form using variable elimination techniques. Both the algebraic decomposition and the proposed improvement are presented and illustrated through an example.