5. Conclusion
The simultaneous development of several products, long duration of product development, and usually limited amount of resources cause that the effective management of NPD projects is a challenging goal. CBR provides methodology for supporting product design by adapting previously solutions to new problems during the NPD process. This study presents the use of ANN to calculate attribute weights in a case-based reasoning approach in the context of case retrieval and reuse to cost estimation of NPD projects. In turn, cost estimation is used to select the most promising portfolio of NPD projects and then in the revision phase of CBR for monitoring the NPD process. The presented approach is able to specify a smaller number of k-NN with less prediction errors than an approach based on equal weights for attributes. Conducted experiments also illustrate that the use of the presented algorithm for calculating attribute weights improves the accuracy of cost estimation in comparison with MRA models and ANN trained according to the gradient descent with an adaptive learning rate algorithm. Consequently, more precise estimation of the new product cost helps the project manager select the most promising portfolio of NPD projects. Moreover, the retrieved cases are used to obtain additional information about developing a new product, for example, required materials and technological process, which the R&D department can use to revise requirements related to designing and testing a new product. The presented CBR approach allows the project manager during product design to refer existing problems to solutions that occurred in similar past NPD projects. Drawbacks of using the proposed approach can be considered in the perspective of collecting a sufficient number of similar NPD projects.
