| Summary: | 博士 === 國防大學中正理工學院 === 國防科學研究所 === 99 === The execution of the cost planning by design to cost (DTC) concept is extremely important for industries when estimating the cost of product lifecycles. Current cost estimation methods exists some drawbacks, like the solution quality and the robusty and flexibility of estimation model, etc. Also, the actual cost database in many industries includes many outlier data which could not be used directly. Thus, this research applied a novel method of machine learning, the least squares support vector machines (LS-SVM) to the cost estimation model for product lifecycles. This research also attempted to combine the LS-SVM with data transformation (DT) mechanisms for smoothing the data distribution on the cost database to get more accurate estimation results. In the case studies, this research used numerous cost data sets and simulated many possible conditions to verify the feasibility and accuracy of the LS-SVM. The data sets used included four real cases of cost estimation examined by past studies, and simulated the cost data sets of the airframe structural parts for design and manufacture process in the development phase. To compare LS-SVM with current cost estimation methods, this research also used the back-propagation neural network (BPN) and regression analysis (REG) to build cost estimation models. The test results showed that the LS-SVM estimation model can provide more accurate performance. And, the LS-SVM estimation model was able to adapt to highly complex parameters, fewer sample size, and mixed cost databases. Additionally, fewer parameters need to determine in the LS-SVM modelling procedure, and the robust solution can be provided. The test result of combining the LS-SVM with DT mechanisms showed that more accuracy and flexibility can be provided than the pure LS-SVM model. The mechanisms of cost estimation proposed by this research could provide an important reference for various industries to planning and controlling product life cycle costs.
|
|---|
