Michèle Basseville

Optimal sensor location for detecting changes in dynamical behavior

We address the problem of optimal sensor location for monitoring the eigenstructure of a multivariable dynamical system. The criterions which are optimized are the power of new tests designed for detecting and diagnosing changes in the eigencharacteristics of a system [3], [12]. The key points are the choice of the parameterization for computing the criterion and the comparison of designs with a different number of sensors. The discussion of the numerical results for sensor location includes the analysis of the effect of the geometry of the unknown excitation.

Recursive and iterative estimation algorithms for multiresolution stochastic processes

A particular class of processes defined on dyadic trees is treated. Three algorithms are given for optimal estimation/reconstruction for such processes: one reminiscent of the Laplacian pyramid and making efficient use of Haar transforms, a second that is iterative in nature and can be viewed as a multigrid relaxation algorithm, and a third that represents an extension of the Rauch-Tung-Striebel algorithm to processes on dyadic trees. The last involves a discrete Riccati equation, which in this case has three steps: prediction, merging and measurement update. Related work and extensions are briefly discussed.<<ETX>>

Monitoring the Combustion Set of a Gas Turbine

Abstract In this paper, we investigate the problem of monitoring the combustion chambers of the gas turbine of an electrical power generator. We describe a general approach for detecting small changes, which can be applied to a wide class of nonlinear systems, can deal with a reduced order and biased signature of the monitored system, and is robust with respect to changes in the functioning mode. We also describe a statistical method for diagnosing the detected changes in the combustion chambers. We report about the processing of a significant amount of real data from two different types of turbines.

On Failure Detection and Identification: An Optimum Robust Min-Max Approach

Abstract In this paper, the Failure Detection and Identification (FDI) problem for noise corrupted Linear Time Invariant (LTI) systems is considered. We apply to this problem an optimum min-max robust likelihood ratio testing approach which is known optimal in the Gaussian case. The originality of this approach is that ‘detection probability’ and ‘false alarm probability’ (i.e. power vs. level) in presence of noise are considered when referred to ‘optimality’ and ‘robustness’. Based on this approach, we propose a batch processing algorithm to failure isolation. Then we show that the recursive version reduces to some LQ optimization problem for which we provide a recursive solution. Simulation results on an example related to the lateral motion of a light aircraft are also reported.

Convergence rates for eigenstructure identification using subspace methods

We apply the general results of a companion paper (Benveniste and Deylon (2000) on the relationship between identification and local tests, to the estimation of convergence rates for MIMO system eigenstructure identification using subspace algorithms. We provide a new and practical estimator for such convergence rates.

Handling the Temperature Effect in Vibration Monitoring of Civil Structures: A Combined Subspace-Based and Nuisance Rejection Approach

Abstract: : Monitoring the vibrations of a structure has been recognized as a useful monitoring approach. However, the temperature and other environmental factors are known to affect the way a civil structure vibrates. A vibration monitoring method based on a subspace residual and χ2-type global and sensitivity tests has been proposed by some of the authors. In this paper, an extended detection algorithm is presented, which rejects the temperature effect viewed as a nuisance parameter. Its relevance when applied to a simulated bridge deck example is investigated.

Monitoring Dynamic Systems with Polynomial Nonlinearities

Abstract The problem of fault detection for nonlinear dynamical systems is addressed through a combined algebraic and statistical approach. The considered systems have polynomial nonlinearities, modeled by differential-algebraic equations (DAEs) for continuous-time systems, and by state-space equations for discrete-time systems. Differential and commutative algebras are used to transform these models into input-output representations to which a local approach to change detection is applied.

Early Warning of Slight Changes in Systems and Plants

Abstract Techniques for early warning of slight changes in systems and plants are useful for condition based maintenance. In this paper we present an approach for this problem. This approach is based on the so-called “asymptotic local” approach for change detection previously introduced by some of the authors. Its original principle consists in characterizing a system via some identified model, and then monitoring its changes using some data-to-model distance also derived from identification techniques. We show here that this method is of much wider applicability: model reduction can be enforced, biased identification procedures can be used, and finally one can even get rid of identification and use instead some much simpler Monte-Carlo estimation technique prior to change detection.

Detection of slight changes using reduced models and biased identification

Techniques for early warning of slight changes in systems and plants are useful for condition-based maintenance. An approach to this problem based on the asymptotic local approach to change detection is presented. Its principle consists in characterizing a system via some identified model and then monitoring its changes using some data-to-model distance also derived from identification techniques. It is shown that this method can be used even when only poor identification procedures are available (with bias, with oversimplified models, etc.). An example from the gas turbine industry is discussed.<<ETX>>

Change Detection and Diagnosis for Vibration Monitoring

Abstract The two problems of detection and diagnosis of changes in the state transition matrix of a multivariable system with nonstationary unkown state noise, are addressed. New instrumental tests are derived and shown to be numerically powerful, even for small changes. The application to vibration monitoring of offshore platforms is described.