Architecture and design issues in a hybrid knowledge-based expert system for intelligent quality control

A manufacturer's goals are to provide high quality products, operate at low cost and respond rapidly to market place demands. We designed an intelligent system aimed to help a manufacturer increase productivity by assisting the operator and technicians to improve process control methods, reduce production costs by preventing defects, and support management actions by detecting and predicting potential faults. Combining expert systems, neural networks, conventional programs, databases, fuzzy logic and adding other resources we provided a powerful approach for performing intelligent quality control. Using a hybrid approach that is a combination of text, keyword search, pattern association, and rule-based techniques, it is possible to avoid many of the drawbacks pertaining to each paradigm. The objective of the design was to define the best architecture to overcome not only the drawbacks of each technique, but also to become a more valuable and accessible tool. The neural network models are used for prediction of outputs for dynamic and complex systems under real-time constraints. We designed an optimizer to control the process. The expert system rules are used to derive recommendations to the operators to support decision-making and improve quality monitoring and control. Our architecture has been tested off-line with data from two different manufacturing companies. The architecture of the hybrid knowledge-based expert system satisfies the requirements for various industries: chemical, steel, cement, food processing.