Estimating the cost of steel pipe bending, a comparison between neural networks and regression analysis

Design to cost and cost based competition focus on the relationship between design decisions and the resulting cost of manufacturing, supporting and operating products. The efforts to optimize the value (measured in terms of cost/benefit ratio) of products early on in the design process motivated the development of a variety of cost estimation tools. Automatic tools linked directly to the computer aided design (CAD) system and producing cost estimates based on the CAD data are the ideal. However, most of the cost estimating systems available today cannot read the CAD data files directly - the input of these tools include two types of information: 1. Technical, objective information (independent of the person using the system, e.g. the geometry of parts, materials used and dimensions) and 2. Subjective information (e.g, an estimate of the quantities that will be manufactured each period during the product's life cycle). Modern cost estimating systems are based on a combination of a model base and a database. The users who collect the data on actual cost of products usually update the database. The model base has to be updated whenever the technology changes and new processes and materials are made available or when new estimation models are developed. The users cannot frequently update the model base of a cost estimating system. They are dependent on the software suppliers that release new versions of the software with the required updates. In this paper, we describe a cost estimating system that can be linked to the CAD data. The parameters used as input to the cost estimation model are objective and easily available. The proposed system is based on a neural network that learns how to modify cost estimates when a new technology is developed. A comparative study reveals that the proposed system outperforms traditional linear regression analysis models used for cost estimation. (C) 1999 Elsevier Science B.V. All rights reserved.