工業技術研究院

技術簡介

在工廠排程問題中，當場域規模達到數百部機台時，龐大的解空間會使得精確解 (exact method) 和meta-heuristic都無法在理想的時間內得出可接受的排程結果。商用排程軟體常使用dispatching rules進行排程，雖然可以快速地得到排程結果，但是結果常是不盡理想的。因此本研究提出基於智慧樹狀搜尋法的生產排程技術，首先建立大數據資料關聯流程，會將數百張關於需求、動態資訊、製程、機台/治具、條件限制的MES/ERP報表資料，進行關聯，以符合產線實務應用; 再融合工程師多年產線經驗的關鍵rule設定和搭配訂單依時段動態分批進單的特性，發展可拆解/可續接的智慧樹狀搜尋排程演算法。並應用AI技術(如MCTS)等，透過不斷模擬排程與選點，可以找到最佳生產目標和好的投料決策，解決瓶頸站點產能問題。系統導入面板產線應用，相較於過去人工排程提升全產線稼動產出3.3％，評估可增加單廠一年近千萬元營收。再以傳產導入智慧排程為例，應用在鋼鐵產線，可減少加熱站點能耗達8.7％，落實工廠綠能轉型; 於製藥產線應用，則可增加10％的產能規劃，提升旺季接單能力，估計月營收增加800萬以上。

Abstract

In a problem of job scheduling, when production lines reach hundreds of machines, a large solution space makes a traditional exact method and meta-heuristic method unable to calculate good scheduling results in an acceptable time. Commercial scheduling software often adopts dispatching rules, although the results can be quickly obtained, but are often not ideal. Therefore, we proposes a tree-search based intelligent scheduling technology to calculate a good scheduling results within the acceptable time. Firstly to satisfy real practical application, we establish a flow of Big data relationship that links hundreds of MES/ERP data tables, e.g., requirements, machine, fixture, production limitation, etc. Then, we apply critical dispatching rules and characteristics of product order in batch to develop our intelligent scheduling algorithm. And we apply AI technology (e.g., MCTS)，via continuous playout simulation, it achieves better scheduling KPIs and solves problem of bottleneck sites. We import this system in a panel production line, and confirmed 3.3％ utilization and throughput can increased efficiently (i.e., revenue increases 10 million). Besides, in a steel production line, we decrease energy consumption by 8.7％ and achieve green energy transformation; in a pharmaceutical industry, we increase 10％ planning throughput and ability to take orders (estimated revenue increases 8 million per peak season)

技術規格

智慧排程系統，包括以下功能規格： •基於產線資訊與生產條件限制模擬不同的生產計畫排程。 •提供資料驗證工具，可自動分析資料特性，找到關聯異常 (可回饋補值充份運用資料) 和訂單扣抵bank/WIP。 •可接續指定的產線狀態 (如線上在製品和庫存等資訊) 進行重新續排。 •提供使用者介面呈現排程解圖報表、編輯基礎資料和排程演算法必要參數設定。

Technical Specification

Smart scheduling system, including the following functional specifications: •Simulate different production schedules based on production line information and production conditions. •Provide data verification tools, which can automatically analyze data characteristics, find associated anomalies (can return supplementary value and sufficient application data) and order deduction bank/WIP. •Can continue to specify the status of the production line (such as online work in progress and inventory information) to re-schedule. •Provide a user interface to present the schedule solution chart report, edit the basic data and the necessary parameter settings of the schedule algorithm.

技術特色

工研院自主研發的人工智慧樹狀搜尋生產排程技術，可以根據生產資訊 (如WIP) 與條件限制 (如治具限制)，進行產線排程派工，兼具Dispatching rule求解快速的優點和優化機制，透過在求取排程解的exploitation過程中，適度的加入exploration因子，使得排程解可以跳開Local optimal，提高找到更接近Global optimal的機會。

應用範圍

可應用於建立訂單未來排程，預測委外製作時機，評估場域執行現況，提升訂單準交成果。

接受技術者具備基礎建議（設備）

伺服器Core i7 10 core*2，32GB RAM，4TB儲存空間等級以上。

接受技術者具備基礎建議（專業）

C＋＋程式開發人員、具最佳化理論的專業人才、具生管排程實務的專業人才。

聯絡資訊