Hybrid artificial intelligence architecture for diagnosis and decision-making in manufacturing

Abstract

This paper presents an architecture which combines artificial neural networks (ANNs) and an expert system (ES) into a hybrid, self-improving artificial intelligence (AI) system. The purpose of this project is to explore methods of combining multiple AI technologies into a hybrid intelligent diagnostic and advisory system. ANNs and ESs have different strengths and weaknesses, which can be exPloited in such a way that they are complementary to each other: strengths in one system make up for weaknesses in the other, and vice versa. There is, presently, considerable interest in ways to exploit the strengths of these methodologies to produce an intelligent system which is more robust and flexible than one using either technology alone. Any process which involves both data-driven (bottom-up) and conceptdriven (top-down) processing is especially well suited to displaying the capabilities of such a hybrid system. The system can take an incoming pattern of signals, as from various points in an automated manufacturing process, and make intelligent process control decisions on the basis of the pattern as preprocessed by the ANNs, with rule-based heuristic help or corroboration from the ES. Patterns of data from the environment which can be classified by either the ES or a human consultant can result in a high-level ANN being created and trained to recognizet/aat paitem on faJture-occurrences, in subsequent cases in which the ANN and the ES fail to agree on a decision concerning the environmental situation, the system can resolve those differences and retrain the networks and/or modify the models of the environment stored in the ES, Work on a hybrid system for perception in machine vision has been funded initially by an Oak Ridge National Lab0ratoryseed grant, and most of the system components are operating presently in a parallel distributed computer environment.

An introduction to neural networks and their applications in manufacturing

“…However, it has been realized that due to their nature NNs may be integrated with the conventional computing systems and other AI schemes (such as expert systems), so that hybrid intelligent structures may be developed which are more powerful than today's tools (Spelt and Glover, 1991).…”