The Study of Reducing First Article Production Time Based on Virtual Metrology

  • 黃 冠哲

Student thesis: Doctoral Thesis

Abstract

For the manufacturing industries first article production helps the enterprises to ensure the production process is correct and is an essential phase before they can go into mass production phase However it is worth noting that adjusting machine is usually based on rule of thumb or on the non-standard operating procedures When adjustment process becomes complicated some additional costs may become inevitable In 2005 virtual metrology (VM) has been proposed to predict the quality characteristics based on the previous metrology information instead of physical measurement Recently artificial intelligence such as neural network machine learning and deep learning has created a huge impact on different areas Extreme Gradient Boosting Tree (XGBoost) developed in 2016 is a gradient ensemble tree method and can be embedded as a classification or regression model However XGBoost is seldom applied in virtual metrology field In this study we propose a novel virtual metrology system including off-line training phase and on-line detecting phase Two models are proposed in off-line training phase using XGBoost Random Forest (RF) Classification and Regression Tree (CART) Deep Neural Network (DNN) respectively Firstly the model 1 is used to predict the product quality index If inferior product quality characteristics are expected model 2 is applied to predict magnitude of machine adjustment for each index immediately Consequently we propose an offset compensation approach to improve predicting preformance by simulation and Bayesian optimization XGBoost performs well in model 1 and RF performs good robustnees in model 2 We use an factory as a case study to apply our models from December 2019 to February 2020 The results show that using Virtual Metrology system with offset (VM-offset) has the best performance among different methods In contrast to past manual adjustment VM-offset can improve 69 7% of adjustment efficiency and reduce 82 4% of first article production time
Date of Award2020
Original languageEnglish
SupervisorTai-Yue Wang (Supervisor)

Cite this

'