A case-based reasoning approach to the designing of building envelopes

Building-envelope design is an information-intensive process that requires experiential knowledge. Confronted with such a process, a human expert adds to well-known domain knowledge his own experience, or the experience of others, to support his reasoning process and guide him in typical situations. The problem-solving paradigm where reasoning is supported by reusing past experiences is called Case-Based Reasoning ( CBR), and it was added to the Artificial Intelligence (AI) methodology following research in cognitive psychology. Instead of relying solely on general knowledge of a problem domain, or making associations along generalized relationships between problem descriptors and conclusions, CBR is able to utilize the specific knowledge of previous experienced problem situations called cases. CBR is a technology that solves problem by storing, retrieving, and adapting past cases. CBR systems have been proposed as an alternative to rule-based systems whenever the knowledge engineering process of eliciting rules is difficult or unmanageable. Instead, many experiences (or cases) with solutions, warnings, plans, and so forth are collected and new situations are related to a stored recollection of these past cases. New solutions are adapted from the old ones. Research in Knowledge-Based Expert Systems ( KBES) for building-envelope design has shown a similar trend. While computerized assistance was imposed by the large amount of data to be processed, the success of rule-based implementations was hampered by the lack of abstract domain knowledge. Such fields where most of the knowledge is based on experience are often labeled as “weak theory domains,” and they are prime candidates for adopting a CBR approach. This thesis proposes a CBR framework for selecting the construction alternatives during the preliminary stage of the building-envelope design process. The methodology presented aims to find the most suitable design for a new building envelope from a library of prototypical building cases and adapts it to meet the requirements of ASHRAE Standard 90.1/1989 for energy efficient building design. The study outlines the potential benefits of using CBR technology and the key issues encountered while attempting to define the CBR model for building-envelope design. Developing a hierarchy of building-envelope components identifies cases and features for design. The envelope design problem is solved through decomposition, and by combining case-based and rule-based reasoning methods. In searching for a best match to achieve a higher degree of case filtering, a connection between case-based reasoning and Artificial Neural Networks (ANN) is proposed. An ANN-based filtering mechanism is designed to improve the quality of case-matching outcome while enforcing the economy of case representation. The framework proposed by this research has been implemented into the CRED software system demonstrating the feasibility and advantages of using CBR methodology for building envelope design. CRED blends several Al techniques (such as ANN, CBR and KBES) while aiming to offer expert assistance to building design professionals for browsing and selecting building-envelope alternatives.