Computer Science, Prifysgol Cymru Aberystwyth University of Wales

CS46310 (1995-96 session)
Model-Based Reasoning for Physical Systems

Brief Description

A problem at the heart of the research agenda for artificial intelligence is how to reason about the physical world. This module assesses the adequacy of models for this domain and aims to give students an understanding of the issues involved in effectively modelling and reasoning about physical systems. The course will be organised around a number of key real world application domains and will show how the requirements of knowledge and software reuse lead to the need for both compositional models and multiple models of phenomena.

Aims, Objectives, Syllabus, Booklist

Further Details

Number of lectures: 20
Number of seminars/tutorials: 3
Number of practicals: 3 x 1-hour
Coordinator: Dr. Fred Long
Other staff involved: Not yet known
Pre-requisites: CS46010
Co-requisites: None
Incompatibilities: None
Assessment: Assessed coursework - 20%
Written exam - 80%
Timing: This module is offered only in Semester 2

Aims

A problem at the heart of the research agenda for artificial intelligence is how to reason about the physical world. This module assesses the adequacy of models for this domain and aims to give students an understanding of the issues involved in effectively modelling and reasoning about the physical systems. The course will be organised around a number of key real world application domains and these will be used to focus on specific problems that arise in modelling real systems, to show how existing techniques can be used and where such techniques prove inadequate.

Objectives

On successful completion of this module students should:

be capable of assessing the adequacy of models for effectively reasoning in this domain;
understand the consequences of different ontological choices;
be aware of different models of behaviour, and methods for reasoning about change;
understand how the requirements of knowledge and software reuse lead to the need for compositional modelling and multiple models of phenomena;
have detailed knowledge of attempts to model behaviour in specific electrical and spatial domains;
understand the issues involved in automating the modelling process;
understand the relationship between model-based reasoning and non-monotonic reasoning, and the practical issues that arise out of building model-based systems;
appreciate how model-based reasoning can be used in diagnosis systems.

Syllabus

Introduction - 1 Lectures: The problem with rule-based approaches. Overview of: reasoning from first principles; component libraries; formalisation and automated modeling; qualitative descriptions of behaviour; domain-independent model-based problem solvers.
Ontologies - 3 Lectures: Formal languages for conceptual modelling. Component based modelling: component models and interaction modelling. Context-free models. Processes. Bond graphs. Causality. Ontologies for modelling fluids, spatial occupancy and topology.
Behaviour Models - 3 Lectures: Difficulties of numeric modelling. Qualitative values, constraints and ambiguity. Order of magnitude reasoning. Interval reasoning.
Case Study - Spatial reasoning about mechanisms - 2 Lectures: Mechanics and kinematics. Topological and metric information. Configuration spaces, qualitative reasoning and occupancy arrays.
Reasoning About Change - 3 Lectures: Evolving qualitative states. State transitions, ambiguity and multiple time scales. Non-linear systems and phase portraits.
Compositional and Multiple Modelling - 2 Lectures: Model properties, assembling model fragments, multiple models and model selection. Logical relations between relational models.
Model-Based and Non-Monotonic Modelling - 3 Lectures: Non-monoticity. Persistence. Minimum violation of normality. Default logic. Truth maintenance. Assumption-based truth maintenance.
Model-Based systems and Diagnosis - 3 Lectures: Tasks and problems: diagnosis, repair, conceptual and innovative design. failure modes and effects analysis, sensor placement and selection, task generation and testing. Diagnosis: basic concepts, fault models, complexity issues, hierarchies, conflicts and focusing. Managing multiple models.