Author: DE LIRA RAMIREZ, SALVADOR
Data of defense: 31/10/2011
Abstract
Fuel cell systems are considered as clean and e¿cient power sources, which are under development by manufacturers for both stationary and mobile applications. Polymer Electrolyte Membrane (PEM) fuel cells are considered to have the highest energy density due to the nature of the reaction and the quickest startup time. These are the main reasons for being used in applications such as automotive engines, portable and backup power applications. Recent years have seen the proliferation of PEM fuel cell system (PEMFCS) optimization and control applications, where the aim is to obtain a better process performance. Nowadays, increases in safety, reliability and uptime process operation are requiring the inclusion of fault diagnosis algorithms. Because of the lack of space for physical redundancy and cost reduction such as in automotive applications, the industry is pushing and facing better techniques for fault diagnosis that makes its products compatible in the markets o¿ering to the ¿nal customer not only the best quality but also a reliable product. Here an alternative technique to hardware redundancy is the analytical redundancy which uses a mathematical model with input and output measurements as monitored system signals to generate a diag- nosis of the system. Analytical redundancy could also allow increasing the fault tolerance using recent methods of Fault Tolerant Control (FTC). The problem of fault diagnosis in a PEMFCS is addressed in this Thesis for parametric and additive faults. A model based fault diagnosis approach is proposed. The problem of robustness against modelling uncertainty is faced using robust fault detection and isolation techniques. Fault detection is based on a LPV interval observer, which can deal with the variation of parameter with respect to the operating point while preserving a linear structure of the model. Parametric uncertainty e¿ects are propagated to the observer outputs using a zonotope representation to approximate and propagate the set of possible states. The fault isolation approach proposed is based on the use of residual fault sensitivity analysis. Sensitivity o¿ers information, which could be essential to isolate faults where just binary information does not allow to isolate them. Fault isolation is based on a set of structured residuals that are analyzed using a relative fault sensitivity analysis approach. The proposed method- ology of fault diagnosis is applied as case study to a PEMFCS. Moreover, this thesis provides a state space model for a PEMFCS that can be used for a real time control and system diagnosis. The model is able to describe the dynamic behaviour of a PEMFCS taking into account the e¿ect of the operating point changes using material and energy balances. The model uses a set of parameters which can be identi¿ed solving a linear least- square problem and using a set of a PEMFCS sensor data obtained from the laboratory set-up, that corresponds to a commercial PEMFC prototype. This lab set-up is used to test the fault diagnosis proposed in this PhD thesis.
Share: