Case Based Design of Knitwear

This project explores the use of artificial intelligence to automatically generate designs using case based reasoning (CBR). CBR reuses past experience to solve new problems.

In the developed world, we are surrounded by objects that have been designed, yet design can be labour intensive.  Despite its creative nature, design is also repetitive; designers use other designs as sources of inspiration. If the reuse of past designs can be automated, then designers can spend less time being repetitive, and more being creative.  

The subject of this project is the design of hand knitted garments. Knitters use thousands of stitches to produce garments by following knitting patterns. The patterns have complex codes that are understood only by knitters. The knitting pattern design process is complex, and largely manual, utilising the experience of designers. Patterns must obey the technical rules of knitting, and enable the knitter to produce a fashionable garment.

Most knowledge about knitwear design is passed from designer to designer, without being written down. This “tacit” knowledge is typical of problems that are amenable to CBR.  In these problems, the human expert is so experienced that they no longer have to think about the details of how they perform tasks. In fact, this also applies to many everyday problems, for example experienced car drivers often change gears without conscious thought. In the past, many problems such as design were solved using rules. Programmers translated the knowledge of experts into rules in a computer program. However, because experts find it hard to explain their tacit knowledge, these programs were often ineffective.

CBR takes a different approach; it reuses specific examples of previous experience.  A CBR system begins with a store of problems and their solutions. When a new problem is presented, the system finds the most similar previously solved problem from the store.  It then reuses and “adapts” the old solution to the new problem.   This style of problem solving occurs in humans also, for example a doctor may diagnose a patient by remembering a previous patient with similar symptoms.

A knitwear computer-aided design (CAD) system was developed specifically for this project. The CAD system works by asking the user to complete a questionnaire about their requirements, and then rules are used to generate a design from this.  Since rules are imperfect, the design must be manually edited afterwards. To reduce this time-consuming editing, this project integrates CBR and CAD in an innovative way.  The questionnaire is used as the statement of the problem for CBR. The program uses a method known as rule difference replay, which was developed as part of this project. The design that was previously generated is reused, in conjunction with the rules, to produce a new design.

Experiments have compared the effectiveness of an integrated CBR-CAD system, with the use of CAD alone.  If a computer can anticipate a designer’s requirements, then this can have applications in other computer-aided design processes.

This work was jointly supported by funding from an EPSRC CASE studentship and Sirdar Spinning Ltd.

For more information on Research degrees in the School of Engineering and Applied Science please click here