ABSTRACT
This chapter describes the development of an intelligent hybrid system that uses expert system techniques to incorporate various pattern recognition and neural networks as tools for data analysis and classification. This system is a knowledge-based interactive problem-solving system. The overall solution to a problem consists of a sequence of steps with a set of methods used at each step. The solution is represented in the form of a decision tree and each node of the solution tree represents a partial solution to the problem. The solution decision tree is formulated by the user. At each step, the user may be brought in to do exploratory investigations. Thus, the paradigm of the system is an optimized divide-and-conquer approach – to provide the user an opportunity to subdivide the problem on hand into arbitrarily small subproblems and allow the user to select the best methods available for the solution of each subproblem. As the problem solving is a highly intelligent process which requires manipulation of symbolic and numerical routines, as well as specific problem-solving strategies, the system contains domain knowledge for assisting the user to explore the problem solving methods, select computation tools, evaluate results, and make decisions, or to help the user make decision for either repeating the process or proceeding to the next level. Using a commercial expert system shell (the KEE system), a system for data estimation and pattern classification applications was implemented to illustrate the concept and used in real world problems of industrial inspection applications. Such approach enables to explore a variety of methods, both classical pattern recognition methods and artificial neural network models, and compare the results to select a best solution to the problem. In many cases, this approach led to remarkable results.
