多目標多任務最佳化遺傳演算法應用於加工排程

Abstract

隨著電腦工藝規劃與演算法進步，學界開始使用各種演算法對各種議題進行研究，對各種目標進行最佳化，如：使用NSGA-II對柔性工藝路線優化、蒙地卡羅樹狀搜尋對多目標彈性零工式工廠排程最佳化……。查到的文獻中演算法在排程領域上多利用單任務多目標最佳化遺傳演算法或多任務單目標最佳化遺傳演算法，並且於限制條件上較為理想，而多目標多任務最佳化遺傳演算法以及改善演算法限制條件貼近實際工廠運作規則少人研究。 近期環境工廠增加以「客製化」方式提高利潤，客製化為相似工件或製程於同一廠區內進行生產，因為相似工件之工件工序不同，每種客製化訂單之排程問題視為獨立多目標最佳化問題，又因為客製化工件排程問題因工件相似，工序排程背後明顯相互關連，因此可將多個客製化訂單視為一多目標多任務最佳化問題，藉由不同任務間知識轉移增加演算法搜索效率達到更有效地排程。此外，碳排放議題日益受關注，因此使用多目標多任務最佳化遺傳演算法於實際工廠的運作提供排程之建議，預期提升實際工廠作業效率、減少總加工時間以降低碳排放量達到減碳目的。 綜合上述問題，本研究將發展一套排程系統，根據現實工件工程圖與真實加工參數使用HyperMill產生CAD/CAM，匯出加工參數與工序加工時間與加工功率，利用Excel的VBA巨集判斷與調整加工參數、時間與功率，輸出符合實際加工機台之數據，最後利用MATLAB經由符合真實情形之遺傳演算法，提出該情況下的最適排程方式，並得出預期加工時間與碳排放。 模擬結果顯示，本研究所發展的排程系統，可針對真實加工機台與加工工件進行可加工與不可加工之判斷並進行加工參數調整，達到增加可加工機台選擇之可能性與增加實際可加工之效率。並藉由符合真實情形之遺傳演算法獲得最適之加工排程結果與最小化總體加工時間與最小化碳排放。

With the advancement of computer process planning and algorithms, the academic community has begun to utilize various algorithms to research diverse issues and optimize various goals. For instance, using NSGA-II for flexible process route optimization, Monte Carlo tree search for multi-objective flexible job shop scheduling optimization, etc. The literature often utilizes single-task multi-objective optimization genetic algorithms or multi-task single-objective optimization genetic algorithms in the field of scheduling. These algorithms tend to work well under ideal constraints, but there's limited research on multi-objective multi-task optimization genetic algorithms and the enhancement of algorithmic constraints closer to the actual operation rules of factories. Recently, "customization" has been employed in factories to boost profits. Customization involves the production of similar workpieces or processes within the same factory area. As the processing order of similar workpieces differs, each customized order's scheduling problem is treated as an independent multi-objective optimization problem. Due to the similarity in the workpieces, the sequencing behind the operations is significantly correlated. Therefore, multiple customized orders can be considered as a multi-objective multi-task optimization problem. Knowledge transfer between different tasks can enhance the efficiency of the algorithm search, achieving more effective scheduling. In recent years, carbon emissions have increasingly drawn attention. Therefore, utilizing multi-objective multi-task optimization genetic algorithms can provide scheduling suggestions based on actual factory operation rules. It is expected to improve the operational efficiency of actual factories and reduce the total processing time, thereby reducing carbon emissions and achieving carbon reduction. In summary, this study will develop a scheduling system that uses HyperMill to generate CAD/CAM based on real workpiece engineering drawings and real processing parameters. The system will export processing parameters, processing time, and processing power, use Excel's VBA macros to judge and adjust these parameters, and output data that match the actual machining equipment. MATLAB, equipped with a genetic algorithm that aligns with real-world scenarios, will then be used to propose the most suitable scheduling method under these conditions and predict processing time and carbon emissions. Validation results show that the scheduling system developed in this study can make feasible and unfeasible judgments for real machining equipment and workpieces and make parameter adjustments, thereby increasing the possibility of choosing more machines and increasing the actual machining efficiency. Moreover, it can derive the optimal machining scheduling results and minimize total machining time and carbon emissions through a genetic algorithm that aligns with real-world scenarios.