Computers and Chemical Engineering 30 (2006) 548–557 Similarity concept for case-based design in process engineering Yuri Avramenko, Andrzej Kraslawski ∗ Department of Chemical Technology, Lappeenranta University of Technology, P.O. Box 20, FIN-53851, Lappeenranta, Finland Received 31 March 2003; received in revised form 31 July 2005; accepted 17 October 2005 Available online 28 November 2005 Abstract The design process requires significant engineering experience, intuition and creativeness. To facilitate the preliminary design, the authors have proposed approach based on the re-use of design experience. The design problems in chemical engineering are often quite difficult to represent as a well-structured list of features of one-type data. Many problems in chemical engineering are very large and complex, the problem description is often incomplete and uncertain, and sometimes it changes dynamically. In order to cope with design cases that have different structure representation the general similarity concept has been developed. The paper surveys the existing mathematical methods for similarity measurements and extends them to cope with different data formats. The main idea of this paper is an integration of representation and retrieval of the cases into one step. The proposed concept is illustrated by the examples of the model selection for synthesis of distillation systems and formulation of a fats and oils product. © 2005 Elsevier Ltd. All rights reserved. PACS: 07.05.K Keywords: Case-based reasoning; Similarity measures; Design support 1. Introduction Over the last few years, one of the methods of artificial intelligence, case-based reasoning (CBR) has attracted a gen- eral interest. The main idea of CBR is based on the assumption that the similar problems have the similar solutions. The prac- tice shows that often it is more efficient to solve a problem by starting with a solution of a previous, similar problem than to generate the entire solution from scratch. The central notion of CBR is a case. A case is problem- solving episode of experience that is represented as a pair: problem and its solution. Many cases are collected in a set to build a case library (case base). In solving a current problem, a CBR system retrieves a similar, past problem and its solution using a set of rules for measuring similarity between actual problem and those stored in case base. Usually it is unlikely that an exact match will occur, therefore the retrieved solution must be adapted. The adaptation rules, based on the problem domain theory, are applied to adjust ∗ Corresponding author. E-mail address: andrzej.kraslawski@lut.fi (A. Kraslawski). for any differences between the current case and the retrieved one. Finally, the CBR stores the approved solution to the current case, and it can then be used in solving future problems. There are several advantages of using CBR. Instead of relying on general knowledge of a problem domain, CBR employs the specific problem characteristics. CBR is beneficial when the problems are not completely understood so that a reliable model cannot be built. Moreover, the problem may not be completely defined before starting to search for possible solutions. However, if the case base does not have sufficiently similar cases, then the retrieved solution may be inappropriate. In addition, the library of cases must roughly cover the set of problems that may arise in the considered domain of application to achieve the acceptable solution. Due to the above-mentioned properties, CBR systems have attracted a great attention in legal and medical domains, especially as diagnostic and care systems, as well as in finance and insurance for customer support and credit assessment (Allen, 1994). In addition CBR has a diversity of applications in intelligent Web-based sales service (Watson & Gardingen, 1999; Wilke, Lenz, & Wess, 1998), in building and mechanical design (Rivard & Fenves, 2000; Mileman, Knight, Petridis, Cowell, & Ewer, 2000), in material science (Amen & Vomacka, 0098-1354/$ – see front matter © 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.compchemeng.2005.10.011