Kads a methodolofogry d evelopilnagr gaei systems.pdf

KADS: A METHODOLOGY FOR DEVELOPING AI SYSTEMS LARGE

BORN GARY
SD-SCICON PEMBROKE HOUSE, PEMBROKE BROADWAY KINGDOM CAMBERLEY, GU15 3XD UNITED SURREY FAX:(0276) 683511
ABSTRACT

This paper discusses the KADS Methodology, which resulted from the largest-ever project devoted to a methodology for developing AI systems. KADS provides support for high-level analysis and design, includingknowledge acquisition. It is supported by a set of development tools acting as a knowledge engineer’s workbench. This paper describes the methodology itself, particularly its life-cycle model, model of expertise, interpretation models and use of modality. It describes how it has been used within SD-Scicon for system development. This has resulted in two useful systems based on AI techniques. Finally, itindicates some future directions for KADS.

engineering. Finally, it indicates some -future directions for KADS, including one where it is likely to be integrated into conventional system development methodologies.
HISTORY OF KADS

KADS originated from work carried out at the University of Amsterdam on the modelling of problem solving behaviour and was originally funded by ESPRIT Project 12beginning in 1983. ESPRIT Project 304 continued the work and was later converted into ESPRIT Project 1098 in 1985. During the years the acronym KADS has had changes of meaning as the project concentrated on different aspects; it is currently defined as Knowledge Acquisition and Design Support. Although the project ended early in 1990, the methodology is actively in use throughout Europe and isgrowing in popularity in the USA.

INTRODUCTION

SD-Scicon have been members of the project since 1985. The other partners are listed at the end of this paper.

This paper summarises the KADS Methodology for developing AI (Knowledge Based) Systems. KADS is the result of a six-year collaborative project sponsored by the European Commission’s ESPRIT programme. It is the largest-ever project devotedto the production of a methodology for developing such systems. Ending early in 1990, the project has produced a viable methodology for constructing large AI systems. It provides support for high-level analysis and design, including what is usually referred to as knowledge acquisition. It is supported by a set of development tools known as Shelley, which act as a knowledge engineer’s workbench.The methodology has been used on over 60 projects to date, and training courses are available in several countries. This paper describes the methodology itself, particularly its life-cycle model, model of expertise, interpretation models and use of modality. It describes how it has been used within SD-Scicon for system development. This has resulted in two useful systems based on AI techniques: fornetwork management and for support of software

OVERVIEW OF THE METHODOLOGY
CYCLE MODEL LIFE

KADS traditionally has employed a step-wise Life Cycle Model (LCM), consisting of analysis, design, implementation,installation, use and maintenance. This is shown conceptually in Figure 1. In the final stages of the KADS project, this was expanded within Barry Boehm’s spiral model framework [l],where an iterative model of risk assessment enables a highly controlled and limited form of prototyping. Within each stage of the LCM, a series of activities are laid out with specified guidance on

.
. .
166

When the activity should be performed Why and how the activity should be undertaken What input and outputs are expected.

CH2867-0/90/0000/0166$01 0 1990 IEEE .OO

and testedsystem
amiysed knowledge

installed

system

1
Figure 1 - KADS Top-Level Life Cycle Model (from Reference [2]) abstraction. These are linked by an overall model of expert knowledge, with four layers of description (Figure 2):

Considerable emphasis is given to the management of development within this framework. MODEL EXPERTISE OF

.
. .

The Domain layer, with all the domain-de...