Data-Driven Modeling Using System Integration Scaling Factors and Positioning Performance of an Exposure Machine System

作     者：Tsai, Jinn-Tsong Chang, Cheng-Chung Chen, Wen-Ping Chou, Jyh-Horng 

作者机构：Natl Pingtung Univ Dept Comp Sci Pingtung 900 Taiwan Natl Kaohsiung Univ Appl Sci Dept Elect Engn Kaohsiung 807 Taiwan Met Ind Res & Dev Ctr Kaohsiung 811 Taiwan Natl Kaohsiung First Univ Sci & Technol Inst Elect Engn Kaohsiung 824 Taiwan Kaohsiung Med Univ Dept Healthcare Adm & Med Informat Kaohsiung 807 Taiwan 

出 版 物：《IEEE ACCESS》 (IEEE Access)

年 卷 期：2017年第5卷

页      面：7826-7838页

核心收录：

基　　金：Ministry of Science and Technology  Taiwan  R.O.C. [MOST 105-2221-E-151-024-MY3  MOST 103-2221-E-153-004-MY2  MOST 105-2221-E-153-005] 

主　　题：Data-driven modeling system integration scaling factors exposure machine uniform experimental design multiple regression back-propagation neural network adaptive neuro-fuzzy inference system 

摘      要：A data-driven modeling approach is proposed for using system integration scaling factors and positioning performance of an exposure machine system to build models for predicting positioning errors and for analyzing parameter sensitivity. The proposed approach uses a uniform experimental design (UED), multiple regression (MR), back-propagation neural network (BPNN), adaptive neuro-fuzzy inference system (ANFIS), and analysis of variance (ANOVA). The UED reduces the number of experimental runs needed to collect data for modeling. The MR, BPNN, and ANFIS are used to construct positioning models of an exposure machine system. The significant system integration scaling factors are determined by ANOVA. The inputs to the data-driven model are system integration scaling factors f(x), f(y), and f(q), and the output is the positioning error. The UED was used to collect 41 experimental data, which comprised 0.0595% of the full-factorial experimental data. Performance tests demonstrated the excellent performance of the UED in collecting data used to build the MR, BPNN, and ANFIS data-driven models. The data-driven models can accurately predict positioning errors during validation. In addition, a sensitivity analyses of parameters showed that design parameters f(x) and f(y) have the greatest influence on positioning performance.