Design Knowledge Acquisition: Task Analysis and a Partial Implementation Yoram Reich Department of Civil Engineering Carnegie Mellon University Pittsburgh, PA 15213 Knowledge Acquisition: An International Journal of Knowledge Acquisition for Knowledge-Based Systems, 3(3):237-254, 1991 Abstract: Design is conceptualized as an ill-structured process that requires diverse knowledge that is hard to acquire. Systematic analysis of design and the knowledge requirements is has in general and in the context of bridge design shows that the knowledge needed can be semi-automatically acquired by using machine learning techniques. Although there are limitations to the approach, preliminary results in the bridge design domain are promising and can potentially transfer to other design domains. 1 INTRODUCTION The process of knowledge acquisition for any type of expert system is time-consuming and tedious. This effort increases when dealing with design domains that are ill-structured by their nature. One approach that promises to alleviate the difficulty of the knowledge acquisition process is the introduction of learning into system development and maintenance stages (Reich and Fenves, 1989b; Shalin et al., 1988; Witten and MacDonald, 1988). For tasks that are relatively well understood and specified, the above promise can be realized by using a single learning method. For example, one can use the learning program ID3 (Quinlan, 1986) to create decision trees for classifying symptoms into malady classes. Single learning techniques have also been used for very restricted aspects of design (e.g., knowledge refinement (Mitchell et al., 1985) or the acquisition of simple design rules (Arciszewski et al., 1987)); however, no overall approach for the acquisition of design knowledge has been proposed. The goal of this research is to develop a framework for assisting in identifying learning methods that can automate or assist in design knowledge acquisition. Understanding the issues involved in design knowledge acquisition requires a systematic analysis of the task, deriving the knowledge it uses, the knowledge dynamics and application (Gaines, 1987). The requirements formulated by this analysis are then mapped into specific knowledge acquisition 1