예측 모델 기반 섀플리 값을 이용한 타이어 설계 인자 분석

Abstract

In the tire manufacturing industry, factor analysis plays a crucial role in deriving design candidates that can increase tire performance by deriving design factors, including variable importance and directions of design change. Typically, expert knowledge and finite element analysis (FEA) are necessary to derive design factors. However, they required a high computation load. Although several studies have been performed using machine learning and variable selection methods, they have limitations in recommending design change directions for each variable. In this study, we propose a predictive model coupled with Shapley additive explanations (SHAP) for tire design factor analysis. We first construct a predictive model using tire design data, and then applied SHAP for deriving important variables and directions of design change. To demonstrate the effectiveness of our methodology, we evaluate the design candidates using FEA. Results show that our methodology is effective for a tire factor analysis.