Online surveillance of steam generator installation based on Bond Graphs

In industrial IT, process supervision is an advanced control and monitoring application (in terms of fault detection and localization). In industrial processes, the supervision system is known as SCADA (Supervisory Control And Data Acquisition), which includes hardware components (data collection) and software (alarm generation based on monitoring algorithms). Supervisory software generally runs on a computer in communication, via a local or remote industrial network, with one or more pieces of equipment:

• Programmable Logic Controller (PLC) ;

• computer ;

• specialized card.

In the industrial world, automatic control systems of the PID (Proportional Integral Derivative) or advanced (observer or fuzzy, for example) type are well mastered. The industrial application of on-line monitoring systems, on the other hand, is underdeveloped. They are often limited to variable thresholding systems and/or variable monitoring: the human operator is still the main "supervisor". The robustness and reliability of the many academic tools developed in the literature [AG 3 550] , as well as safety constraints, are technological obstacles to be overcome.

The aim of this article, which is a sequel to [AG 3 550] , is to apply bond graph models to the on-line monitoring of a complex real plant representing the operation of a thermal power station. It is organized as follows:

The first part presents the description and dynamic modeling of the plant, taking into account the parametric uncertainties of the system. The plant is then represented by an uncertain bond graph model in the form of Linear Fractional Transformations (LFT). The interest of bond graphs for this task is justified by the complexity of the non-linear process (due to the coupling of several phenomena: mechanical, thermodynamic, thermal, fluidic...) and the modular approach to graphically representing uncertainties.

The models obtained are used in the second part to systematically generate robust fault indicators and adaptive thresholds...