Simulation System for Cut-to-length Line based on Neural Network Prediction

[[abstract]]摘 要 定尺剪床為於鋼筋軋延產線(Rolling Mill)之後段製程設備之一，主要功能為客製化裁剪鋼筋。就剪床刀片而言，經過長時間的剪切撞擊鋼筋後，刀片會造成不同程度之破裂或磨耗，導致成品鋼筋切口會產生歪斜或翹曲的變化，不利於日後客戶的續接加工。因此，如何評估剪床刀片的磨耗程度，與刀片剩餘可用壽命(Remaining Useful Life RUL)，以事先安排維修時間，實為加工實務的重要挑戰。 本研究除基於實際軋延產線資料，採用Flexsim以建構一模擬產線，更利用類神經網路模型，預測產線中剪床刀片的RUL。藉由刀片加工特性所建立的預測模型中，所輸入參數將包含刀片已使用時間、每日鋼筋產量、鋼筋直徑、鋼筋溫度、鋼筋碳含量、錳含量、鋅含量、鋼筋等級、與鋼筋切斷面等，而預測模型的輸出參數則為該刀片的累積生產量。最後，再將所預測結果提供為模擬排班依據，以降低產線因刀片異常破損所產生的非預期停機時間。 實驗結果顯示，本研究所建構的剪床模擬產線產量與實際產量，於3個月、6個月與12個月的平均絕對誤差分別為3 12%、3 33%與3 47%，代表所建立的模擬產線具一定的產出分析能力。而當根據剪床刀片預測磨耗程度，事先安排刀片更換計畫以減少停機時間，則模擬總生產量於3個月、6個月與12個月時，可分別增加3 57%、2 55%與1 88%的總產量。因此，本研究之定尺剪床模擬與預測機制，將有助於提升產線產量。 關鍵字：定尺剪床、生產模擬、類神經網路預測、剩餘可用壽命。[[abstract]]Abstract Cut-to-length shear designed for cutting custom steel is a backend process of steel rolling production line After long-term shearing the shear tool will be worn or cracked resulting in skewed or warped variation of finished steel cut which could damage the following operations of customers Therefore challenges are how to assess the degree of tool wear and remaining useful life of tool for scheduling maintenance Based on actual production data this study not only builds a simulation line by Flexsim but also predicts the timing of tool replacement by neural network Utilizing cutting characteristics of tools the prediction model is inputted with the following features: tool usage time daily production amount steel diameter steel temperature carbon quantity manganese quantity zinc quantity steel grade and steel cut Finally the prediction model serves the simulation line with the accumulated production quantity to reduce unscheduled downtime which could be caused from tool issues Experimental results show that the mean absolute errors of production amounts of simulation and actual lines were 3 12% 3 33% and 3 47% by 3 6 and 12 months respectively It indicates that the capability of the simulation line is significant for output analysis After scheduling by tool wear the simulated production amounts can be increased 3 57% 2 55% and 1 88% by 3 6 and 12 months respectively Hence the prediction and simulation system for cut-to-length line is promising in increasing line capability Keywords: Cut-to-length shear production simulation neural network prediction remaining useful life