ABSTRACT
A new problem-solving method is proposed which utilizes a neural network trained by not only precedent samples but also knowledge rules to draw conclusions. The motivation for this work is based on the limitations of mathematical optimization and expert system methods. This chapter introduces the basic concepts and techniques of such rule-combined neural networks in the context of engineering design optimization. In particular, the relationship between rules and samples is established. Knowledge rules are shown to play an important role in fast approaching the expected performance of neural computation.
