A neural-network-based approach for estimating the cost of assembly systems

Any product that consists of two or more parts requires some form of assembly. The selection of the most appropriate assembly system for a product or a family for similar products and the detailed design of the assembly system are important activities in the life cycle of many products. The selection of an assembly system is usually based on cost-benefit analysis. The analysis requires engineering expertise and decision-support systems for data collection and data processing. This paper demonstrates the potential value of using neural networks in cost estimation as opposed to traditional regression or engineering analysis. A network architecture is proposed by which the expected cost of six major types of assembly system is estimated. The performances of the network are compared to those of a regression model commonly used for cost estimation. The neural network consistently outperformed the regression model with respect to several performance measures.