Model-based explanation of plant knowledge
|Author:||Huuskonen, Pertti J.|
|Online Access:||PDF Full Text (PDF, )|
|Publish Date:|| 2005-05-20
|Thesis type:||Doctoral Dissertation
|Defence Note:||Academic dissertation for the degree of Doctor of Technology to be presented, with the permission of the Department of Electrical Engineering, University of Oulu, for public discussion in the Auditorium L10, Linnanmaa, on May 30th, 1997, at 12 o’clock noon.
Professor Matti Pietikäinen
Professor Martti Mäntylä
Doctor Raimo Korhonen
Professor Kari Kuutti
Doctor Raimo Korhonen
This thesis deals with computer explanation of knowledge related to the design and operation of industrial plants. The needs for explanation are motivated through case studies and literature reviews. A general framework for analysing plant explanations is presented. Prototypes demonstrate key mechanisms for implementing parts of the framework.
Power plants, steel mills, paper factories, and high energy physics control systems are studied to set requirements for explanation. The main problems are seen to be either a lack or an abundance of information. Design knowledge in particular is found to be missing in plants. Support systems and automation should be enhanced with ways of explaining plant knowledge to the plant staff.
A framework is formulated for analysing explanations of plant knowledge. It consists of three parts: 1. a typology of explanation, organised by the class of knowledge (factual, functional, or strategic) and by the target of explanation (processes, automation, or support systems), 2. an identification of explanation tasks generic for the plant domain, and 3. an identification of essential model types for explanation (structural, behavioural, functional, and teleological). The tasks use the models to create the explanations of the given classes.
Key mechanisms are discussed to implement the generic explanation tasks. Knowledge representations based on objects and their relations form a vocabulary to model and present plant knowledge. A particular class of models, means-end models, is used to explain plant knowledge. Explanations are generated through searches in the models. Hypertext is adopted to communicate explanations over dialogue based on context.
The results are demonstrated in prototypes. The VICE prototype explains the reasoning of an expert system for diagnosis of rotating machines in power plants. The Justifier prototype explains design knowledge obtained from an object-oriented plant design tool. Enhanced access mechanisms into on-line documentation are studied with examples from high-energy physics experiments. The Lepo prototype explains the behaviour of automation logic in various kinds of plants.