2.2.3 Applying Machine Learning Methods to the Airframe Structural Design Cost Estimation – A Case Study of Wing‐Box Project

Abstract

AbstractThis research used two machine leaning methods, the Support Vector Regression (SVR) and Back‐Propagation Neural Network (BPN), to create the cost prediction models for airplane wing‐box structural design, and verified the feasibility and efficiency for both methods. In the case study, four different main structural part groups of the wing‐box, Spars/Ribs/Skins/Stringers, were chosen. In the parts data base, the part dimensions were included and used for classifying the part groups. Each part group has 150 bill of parts, 100 bill of parts used for training samples, 50 bill of parts used for predicting samples, to test there accuracy. After verified through wing‐box case study, the results showed either SVR or BPN can precisely predicting the design costs. But compare to the BPN, SVR can get the global optimal solution while using less decision parameters. This can save lots of time for searching the best parameters combination when creating the prediction model.