多細緻度模型最佳化於彈性製造系統之探討

Multi-fidelity Optimization with Ordinal Transformation and Optimal Sampling in FMS

陳宏銘(Chen, Hung Ming)

指導教授 ： 林則孟

國立清華大學/工學院/工業工程與工程管理學系/碩士(2015年)

摘要

模擬最佳化經常被使用對複雜系統中進行最佳化求解，然而在複雜系統中所包含之問題特性相當錯綜複雜，因考量特性不同而存在許多種不同細緻度的模擬模型。當運用越高細緻度模型進行模擬系統評估時雖然能夠較準確，卻會需要耗費較長的系統評估時間以及越高的成本。而低細緻度模型雖有偏誤但其績效趨勢可能會反映部份高細緻度模型之績效趨勢。因此本論文將探討如何有效結合多細緻度模型，並以低、高細緻度模型之相關趨勢於模擬最佳化中以節省求解時間、提高求解效率。
本研究以彈性製造系統(Flexible Manufacturing System)的同步進行機台與車輛排程為對象，透過區域控制以及替代機台的特性，進行本研究之高、低細緻度模型設置。為了能有效利用低、高細緻度模型之間的關係，本論文將引用MO2TOS(Multi-fidelity Optimization with Ordinal Transformation and Optimal Sampling)架構對此FMS同步排程問題進行求解。
在MO2TOS架構中，如何依據低細緻度模型績效進行抽樣與分組將會是運用MO2TOS進行模擬最佳化的求解重點。由於模型之間相關程度將可反映出模型之間績效趨勢一致性。本研究提出以Adaptive Sampling藉由求解過程中持續增加的高細緻度模型績效，並已逐漸收斂之模型相關程度進行抽樣方法的更新，將可避免選擇到不適當之抽樣方法。導入Adaptive Sampling後除了求解品質有效提升外，在高與低相關程度模型下平均可以節省45.39%與22.71%模擬資源的使用。在分組方法部分，MO2TOS採用等距方式進行方案分組，然後高細緻度模型績效中相似解之個數並不一定相等，使用等距分組易使組間差異不明顯，進而使資源分配不佳。本研究提出Adaptive Grouping以持續抽樣所得到模型相關資訊更新組別，在高與低相關程度之多細緻度模型下，導入Adaptive Grouping平均可以節省28.45%與17.94%模擬資源的使用。

關鍵字

多細緻度模型 ； MO2TOS ； Adaptive Sampling ； Adaptive Grouping ； K-means分群

並列摘要

In this research, Multi-fidelity Optimization with Ordinal Transformation and Optimal Sampling (MO2TOS) is exploite to solve the simultaneous scheduling problem of machines and automated guided vehicles (AGVs) in flexible manufacturing system (FMS). Fidelity represented the degree to which a simulation replicates reality. Because of FMS contains lots of system characteristics and flexibilities, and there are many different fidelity models exist which considerd different system features.
Evaluating system via higher fidelity model can be more accurately, but it will cause time-consuming and will bring higher cost. Although lower fidelity model may suffered bias, but faster evaluation and the performance can provide partial of trend between low and high fidelity models. It is important to enhance the efficiency of optimization by using multi-fidelity models.
In MO2TOS, applied an inappropriate sampling method will lead a poor quality of optimization. Hence Adaptive Sampling is proposed to update the sampling method by sample correlation coefficient. The correlation coefficient can present the trend between multi-fidelity models, updating the sampling method according to the sample correlation coefficient can enhance the efficiency of optimization and save 45.39% and 22.71% of simulation resources for higher and lower correlation models.
Grouping method is one of main factors which may affect the quality of optimization significantly. In this research, Adaptive Grouping is proposed to update the group after every iteration of MO2TOS. It may significantly enhance the gap between groups, further to allocate resource effectively and save 28.45% and 17.94 of simulation resources for higher and lower correlation models.

並列關鍵字

Multi-fidelity ； MO2TOS ； Adaptive Sampling ； Adaptive Grouping ； K-means

參考文獻

[2] 張祐翔，“應用模擬最佳化於FMS之機台與車輛同步排程”，清華大學工業工程與工程管理研究所碩士論文，2013。

[3] 許雅寧，“粒子群聚演算法於FMS之機台與車輛同步排程”，清華大學工業工程與工程管理研究所碩士論文，2014。

[5] Abdelmaguid, T. F., Nassef, A. O., Kamal, B. A., & Hassan, M. F. (2004). A hybrid GA/heuristic approach to the simultaneous scheduling of machines and automated guided vehicles. International journal of production research, 42(2), 267-281.

[7] Anwar, M. F., & Nagi, R. (1998). Integrated scheduling of material handling and manufacturing activities for just-in-time production of complex assemblies.International Journal of Production Research, 36(3), 653-681.

[9] Balabanov, V., & Venter, G. (2004, August). Multi-fidelity optimization with high-fidelity analysis and low-fidelity gradients. In 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference (Vol. 4459).