動態生產環境下適應性交期指派法則之發展

Abstract

交期指派（Due Date Assignment，DDA）是現場管理決策的要務之一，決定合適的訂單交期並將產品準時送交客戶，將可有效地提昇客戶服務水準並強化競爭優勢。然而，交期指派確是一個困難的決策，尤其是在動態且複雜的生產環境下。因此，本論文將針對動態且複雜的生產環境設計新的交期指派法則，以提高交期預測的準確性。本論文主要分成兩部分。 第一部分將針對以迴歸為基礎之交期指派法則（regression-based DDA method）進行研究與改善。這類型的交期指派方法擁有易於使用與解釋的優點，所以已經廣泛地被使用於實務與學術研究中。本研究認為迴歸模式的預測績效會受到訂單抵達時的系統狀態影響。因此，本研究將利用機器學習工具，分析迴歸模式的架構與系統狀態之間的關連性，進而提出兩個適應性以迴歸為基礎之交期指派法則（adaptive regression-based DDA method），分別為單因子適應性迴歸模式（single variable adaptive regression method）及多因子適應性迴歸模式（multiple-variables adaptive regression method）。本論文所提出之適應性迴歸模式具有線上調整（on-line tuning）迴歸係數之功能，意即其迴歸係數會根據訂單抵達時的系統狀態進行調整。從模擬及統計分析的結果發現適應性迴歸模式比傳統的迴歸模式有較高的交期預測準確性，而且訂單延誤的狀況可大幅改善。 第二部分針對一個新興機器學習工具—案例式推理（Case-Based Reasoning，CBR）進行研究與改善。案例式推理主要根據過去的求解經驗來解決問題。本研究為了提高案例式推理對連續型數值的預測能力與效率，提出一個新的案例標示方法—樹標法（tree-indexing approach）。本論文使用三個UCI資料集合來驗證樹標法的效益，研究結果顯示案例式推理搭配樹標法使用，將能有效地降低案例式推理的預測誤差，並且能顯著地縮短案例擷取時間。本研究並將案例式推理搭配樹標法應用於交期指派問題，實驗結果顯示此一案例式推理系統的表現明顯優於傳統交期指派方法。

Due date assignment (DDA) is the first important task of shop floor control. Due date related performance is impacted by the quality of DDA methods. Assigning exact due dates and timely delivery of goods to the customers enhances customer satisfaction as well as providing a competitive advantage. However, DDA is a difficult decision. Therefore, a lot of previous studies focused on this field and developed DDA methods to do it. The purpose of this dissertation is to develop more accurate and precise DDA methods to predict job due dates in dynamic and complicated production environments, such as a job shop and a wafer fabrication plant. There are two parts in this dissertation. First, many regression-based methods to date have been proposed to solve DDA problem. The advantages of regression-based DDA methods are that they are easy to both put into practice and comprehend. However, relatively little scheduling research has focused on improving the performances of regression-based DDA methods. The performance of a regression-based DDA method could be improved if its values of regression coefficients could provide a more accurate and precise flowtime estimation for each individual job. The difficulty in doing this stems from the dynamic and stochastic nature of production environment that precludes accurate estimation. Therefore, in this study two novel regression-based DDA methods are developed by using machine learning techniques, one of which is a single-variable adaptive regression method and the other is a multiple-variables adaptive regression method. In particular, the adaptive regression-based DDA methods are able to adjust the values of regression coefficients dynamically to best predict the job due date according on the condition of the shop at the instant of job entry. Based on the simulation and statistical results, the adaptive regression-based DDA methods outperformed the traditional regression-based DDA methods with respect to mean absolute lateness (MAL) and mean squared lateness (MSL). The second part of this dissertation discusses in details of a novel machine learning tool -- case-based reasoning (CBR) that uses previous case to solve new problems and has drawn great attention in recent years. A novel case indexing approach, called tree-indexing approach, is proposed in this study in order to improve the efficiency and prediction errors of CBR in numeric prediction. The experiments, using three real world problems from the UCI repository, showed that the CBR with the tree-indexing approach (T-CBR) was superior to the conventional CBR. This study also applied T-CBR to solve the DDA problem in a dynamic job shop environment in order to investigate whether T-CBR’s expected benefits can be observed in practice. The results of the experiments showed that our proposed T-CBR can indeed more accurately predict job due dates than the other DDA methods presently in use.