Value analysis for AI applications in CIM environmmsts

The present state of applying AI in manufacturing industry is one of prototyping and internal exploitation. During this procedure, many worthwhile CIM projects have been turned down because the qualitative benefits could not be included in the justification procedure while the direct cost savings were insufficient to meet the financial hurdles set by the firm. This paper identifies the range of techniques that are in use to justify CIM investments and describes their appropriateness in therms of manufacturing technology, information technology and data uncertainty. In the main part, we apply a hybrid evaluation technique based on qualitative and quantative criteria to the CIWAI software developed in the course of ESPRIT project EP 2434. 1. Introduction In CIM applications, data uncertainty is extremely large: many of the advantages of the new information technologies lie not in the area of cost reduction but rather in more nebulous areas such as simpler scheduling and more consistent quality. CIM investment decisions, therefore, belong to the most challenging decision type anywhere: not because of the complexity of economic calculations and economic theory, however (which are very simple and only need basic mathematics). The difficulty is in the uncertainty of the assumptions; and, taken together, these combined uncertainties can multiply into a total uncertainty of critical proportions. This is where the element of risk enters. Decision support systems, including expert systems, are means to reduce risk and uncertainty in decision making. The methods and techniques investigated by Amficial Intelligence (AI) techniques, therefore, can be directly applied to the solution of economic problems as well, especially when related to data uncertainty. For to understand uncertainty and risk is to understand the key business problem - and the key business opportunity. The broad range of CIM justification techniques, therefore, can best be classified according to the type and accuracy of data which they are based upon and which they process. Using the classification of unreliable data developed in AI [l], these data may be quantitative, qualitative, uncertain and incomplete (Fig. 1).