在面板的陣列( array)製程中，產品的良率與產品電性(Ion及Ioff)有顯著關係。而電性的變異往往相當大，因此面板廠往往需要耗費大量的人力在品質檢驗上，造成原本吃緊的人力在著重於品質的原則下使其產量下降。而面板電性與陣列製程中幾個關鍵機台的控制特徵與環境特徵有相當密切的關係，因此本研究擬提出一適用面板陣列製程之特徵篩選與品質預測模型，協助廠商由目前完成品檢驗模式改為針對在製品檢驗以提升良率並降低生產成本。
面板陣列製程在製品的電性不只受單一站別影響而是受所有前站影響。亦即若前站在製品電性不良，則會影響到下一站在製品的電性。故本研究提出漸進式特徵預測，即上站最佳解特徵及上站預測結果加入至下一站特徵協助預測，每站擁有所有前站的最佳特徵，目的為保留過往資訊。
本研究將與國內某主要面板廠合作，藉由其所收集之關鍵機台的特徵資料，先運用權重隨機森林演算法衡量各特徵重要程度，再利用向前選擇法找出影響電性的重要特徵並用以預測電性。希望能藉由所訓練之隨機森林演算法，預測各站製品的電性以提升完成品的良率。
本研究的成果可替台灣面板廠在陣列段製程建立一新的智慧生產模式。因在製造過程中即可預測產品電性，故工程師可針對預測異常品進行品質控管，利用機台量實際測量該產品，並在數站皆顯示電性異常可立即報廢，此舉不但提高產品良率並可替公司節省相當大生產成本，更可進而提高商譽增加顧客黏著度。

In the TFT/LCD panel array process, the yield of the product is significantly related to the product's electrical properties (Ion and Ioff). Electricity variation is often quite large in the output of the process, so panel factories often need to spend a lot of manpower in the quality inspection resulting in the decline of output quantity. However, the control and environmental parameters of several key machines in the process have quite close relationship with the panel electrical properties. Therefore, this research intends to propose a feature engineering and quality prediction model. Such that the final output inspection mode applied in the current process can be transferred to WIP inspection mode in order to improve yield and reduce production costs.
The project will use the infield data collected from a major panel manufacture in Taiwan to identify the main influence parameters by applying weighted RandomForest algorithm to evaluate the importance of each feature, and a novel progressive feature engineering approach. Then, by applying RandomForest algorithm approach to forecast the electricity.
The research can provide a novel intelligent model for Taiwan panel manufacturers in the array process, which not only improves product yield but also saves considerable costs for the company.

Booth, A., Gerding, E., & McGroarty, F. (2014). Automated trading with performance weighted random forests and seasonality. Expert Systems with Applications, 41(8), 3651-3661. doi:10.1016/j.eswa.2013.12.009
Bosch, A., Zisserman, A., & Muñoz, X. (2007). Image Classification using Random Forests and Ferns.
Chuang, H.-W. (2002). 薄膜電晶體液晶顯示器製程.
de Villiers, J., & Barnard, E. (1993). Backpropagation neural nets with one and two hidden layers. IEEE Trans Neural Netw, 4(1), 136-141. doi:10.1109/72.182704
Hecht-Nielsen, R. (1990). On the algebraic structure of feedforward network weight spaces Advanced Neural Computers (pp. 129-135): Elsevier.
Herrera, M., Torgo, L., Izquierdo, J., & Pérez-García, R. (2010). Predictive models for forecasting hourly urban water demand. Journal of Hydrology, 387(1-2), 141-150. doi:10.1016/j.jhydrol.2010.04.005
Huang, G. B. (2003). Learning capability and storage capacity of two-hidden-layer feedforward networks. IEEE Trans Neural Netw, 14(2), 274-281. doi:10.1109/TNN.2003.809401
Huang, P. B., Chen, J. C., Yu, C.-W., & Chen, C.-C. (2015). Integration of simulation and neural network in forecasting the throughput for TFT-LCD colour filter fabs. International Journal of Computer Integrated Manufacturing, 29(3), 298-308. doi:10.1080/0951192x.2015.1032356
Jeong, Y.-S., Shin, K. S., & Jeong, M. K. (2015). An evolutionary algorithm with the partial sequential forward floating search mutation for large-scale feature selection problems.
Liu, Y., & Ge, Z. (2018). Weighted random forests for fault classification in industrial processes with hierarchical clustering model selection. Journal of Process Control, 64, 62-70. doi:10.1016/j.jprocont.2018.02.005
Maragoudakis, M., & Serpanos, D. (2010). Towards Stock Market Data Mining Using Enriched Random Forests from Textual Resources and Technical Indicators. 339, 278-286. doi:10.1007/978-3-642-16239-8_37
Prasad, A. M., Iverson, L. R., & Liaw, A. (2006). Newer Classification and Regression Tree Techniques: Bagging and Random Forests for Ecological Prediction. Ecosystems, 9(2), 181-199. doi:10.1007/s10021-005-0054-1
Sasaki, N. (2006). A New Measurement Method for Ion Density in TFT-LCD Panels. Molecular Crystals and Liquid Crystals Science and Technology. Section A. Molecular Crystals and Liquid Crystals, 367(1), 671-679. doi:10.1080/10587250108028688
Stathakis, D. (2009). How many hidden layers and nodes? International Journal of Remote Sensing, 30(8), 2133-2147. doi:10.1080/01431160802549278
Surkan, A. J., & Singleton, J. C. (1990). Neural networks for bond rating improved by multiple hidden layers. Paper presented at the Neural Networks, 1990., 1990 IJCNN International Joint Conference on.
Winham, S. J., Freimuth, R. R., & Biernacka, J. M. (2013). A weighted random forests approach to improve predictive performance. Statistical Analysis and Data Mining: The ASA Data Science Journal, 6(6), 496-505.
中華民國經濟部統計處. (2016). 產業發展經濟統計簡訊.