Skip to Main content Skip to Navigation

Filtrage à incertitudes stochastiques et bornées : application au diagnostic actif en automobile

Quoc Hung Lu 1 
1 LAAS-DISCO - Équipe DIagnostic, Supervision et COnduite
LAAS - Laboratoire d'analyse et d'architecture des systèmes
Abstract : The problem of filtering applied to automotive diagnostic is studied in this thesis, for linear or nonlinear, discrete-time dynamical systems, in a context of mixed uncertainties, i.e. uncertainties can be stochastic or bounded (in intervals). This context allows us to combine two well-known approaches of filtering: stochastic and set-membership approach. Through this thesis, we show that they complement rather than compete each other. Two models from the automotive industry are used in the applications along the thesis: bicycle vehicle model and suspension model. Mixed filtering methods are first developed and presented in this work, namely Optimal Upper Bound Interval Kalman Filter (OUBIKF) and Reinforced Likelihood Box Particle Filter (RLBPF), one is dedicated to linear systems and the other to nonlinear systems. The former is based on interval Kalman filter and enhances it by using developed properties and optimization strategy of upper bounds of all admissible covariance matrices belonging to a given interval matrix. The later proposes a general scheme of box particle filter and develops a reinforcement methodology to the likelihood computation, the crucial step of the scheme, to enhance the filter performance. The second part of this thesis is dedicated to fault detection. The previous filters are used and combined with a X^2-based hypothesis testing method with adaptive degrees of freedom, namely Adaptive Degrees of Freedom X^2-statistic (ADFC), to deal with fault detection in linear or nonlinear systems. It is a passive fault detection method enhanced by the adaptive threshold technique in the decision making stage. This method allows the detection of single or multiple additive faults on the sensors. In the last part of this work, a methodology of active diagnosis is developed, that is the ADFC-based Active Fault Diagnosis (AFD) using auxiliary signals. This methodology, a preliminary study to the active approach, is limited to single fault detection. However, its contributions are multiple: isolation (localization) and identification (estimation) of the fault, reduction of false alarms and improvement of the state estimation by returning the estimated fault as a feedback signal to the filter used. Our future researches focus specifically on this approach.
Document type :
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Wednesday, July 20, 2022 - 4:25:11 PM
Last modification on : Thursday, August 11, 2022 - 3:22:59 AM


Version validated by the jury (STAR)


  • HAL Id : tel-03729541, version 2


Quoc Hung Lu. Filtrage à incertitudes stochastiques et bornées : application au diagnostic actif en automobile. Systèmes embarqués. Université Paul Sabatier - Toulouse III, 2022. Français. ⟨NNT : 2022TOU30043⟩. ⟨tel-03729541v2⟩



Record views


Files downloads