A First Step Towards Unmanned Intelligent Process Management: A Procedure for the Diagnostics and Prognostics of Energy Conversion Plants

Abstract

This paper describes the conceptual development and the prototype implementation of an Expert System that deals with the prognosis and diagnosis of a significant subset of the operative faults of a cogeneration power plant. The expert system receives both raw and organised real-time information on the “instantaneous” (actually, averaged over 60-seconds intervals) plant operating conditions from a non-intelligent “plant interface” consisting of a standard plant data acquisition system and of a specific plant simulation software. This solution was adopted in view of possible future applications to industrial plants, where a low number of intrusive sensors is desirable: in this case, the simulator provides the missing data.

The Inference Engine operates on the basis of a set of pre-defined rules that seek possible “faults chains” expressed as combinations of a pre-assigned number of continuously calculated performance indicators, like the air filter output pressure drop, the relative compressor, combustion and turbine efficiency decay with respect to their respective nameplate values, the compressor enthalpy gain and surge conditions, the pollutants concentration in flue gases, the relative pressure losses in both the primary and secondary water circuit, the TTD of the Heat Recovery Boiler, etc.: the complete list includes 29 indices. The rules establish whether a component’s behaviour is degraded by examining both the individual indicators and all of their relevant combinations. A graphic window displays a series of icons, one for each indicator, with a refresh rate of one real-time minute. The Expert System enables the user to determine in detail the instantaneous performance conditions. If performance deterioration is detected, it sends a message to the operator and provides some decision support via a customised graphic interface.

The structure of the code is Object Oriented, and each component as well as each flux are represented as the instance of a class. Both the reasoning and the controlling actions are taken in the same O-O environment.

The present paper presents the organization of an ES whose prototype version has been nicknamed PROMISE, from the Italian acronym for PROgnostic and Intelligent Monitoring Expert System, defines its goals and discusses both the results of the tests conducted so far and the implications for future applied research in this field.

Keywords

• Unmanned Intelligent Process Management, Expert System, diagnosis, cogeneration