Feature-based cost estimation for packaging products using neural networks

Abstract

Cost estimation plays an important role in the product development cycle. For instance, a proper cost estimation can help designers make good trade-off decisions regarding product structures, material, and manufacturing processes. In this paper, a feature-based product cost estimation using back-propagation neural networks is proposed. A system using this approach has been successfully developed for estimating the cost of packaging products. The cost-related features in both design and manufacturing aspects were extracted and quantified according to their cost drivers. The correlation between the cost-related features and the estimated costs of the product was obtained by training and validating a back-propagation neural network based on 60 existing products with their designs, process routings, and actual cost data. To illustrate, the testing results of the trained neural network based on 20 actual products are presented. The performances of the neural network are compared to those of the company's method and a linear regression model. The results show that the neural network model outperformed both the other methods in respect to performance measures such as average relative deviation and maximum relative deviation.