10.1 Introduction

Increasing attention to safety and the need for reduction of energy production costs motivate the research into methodologies that enable one to provide long-term monitoring and early detection of faults and abnormal process behaviour in power plants. This chapter focuses on analytical redundancy techniques that can be conveniently employed to detect failures in the heater line of a 320 MW power plant, regarded as a testbed. The main topics are modelling, identification, and design of estimators that may be used to diagnose faults in the plant.

Grey-box modelling allows one to account for different levels of knowledge regarding a plant. In the literature, many works on grey-box identification techniques are available that deal with linear models (Tulleken, 1993; Gawthrop et al., 1993) and hence are unsuitable for modelling real complex physical processes. A few investigations have focused on non-linear systems and proposed multiple-hypotheses statistical identification techniques (Bohlin, 1994a, b; Bohlin and Graebe, 1995). Such methods evaluate models of differing structure and complexity from a statistical point of view so as to select one that is acceptable in terms of a given criterion. In any case, the on-line adaptation of the model remains in general rather difficult.

As will be explained later on, our approach is substantially different from the aforementioned and may be split into two phases. In the first phase, a model is built that is as consistent as possible with the physical reality of the various components of the...