結合機器學習與模擬最佳化之製程控制-以可逆冷軋機為例

Abstract

本研究以Digital Twin之概念出發，藉由實時的蒐集現實系統運作之數據，打造虛擬系統，建立可逆冷軋機之機器學習預測模型，並結合模擬最佳化演算進行板形優化、提高製程品質。軋機採用往復軋延使得最終鋼板達到目標板形，所謂板形即為衡量鋼板平整程度之指標，板形受先前軋延、機器參數及控制變數等影響。 現行軋機以內建之模糊控制系統進行製程控制，但時常發生最終板形與目標不符的問題。受限於問題之長時間、不穩定狀態且包含許多非線性因子等特性，無法運用微分方程處理，因此本研究以過去資料為基礎，導入人工智慧的方法來模擬且實時更新實際軋延系統之動態。 本研究以集成學習的XGBoost模型打造虛擬軋機，在給定必要參數後即可精準的模擬軋延成果，接著結合模擬最佳化的概念，運用Cross-Entropy演算法計算在不同入口板形及目標板形的情況下所對應最適之控制參數，提升整體製程品質，實現智能製造。

Based on the concept of digital twin, we collect and update the data of the actual system in real-time. In this research, we establish a virtual rolling system to simulate the output of the reversing cold rolling mill. By integrating the virtual rolling system and simulation optimization model to optimize the output shape and improve the process quality. The steel strip moves back and forth between the rollers to reduce the thickness incrementally and to be formed into the expected shape. Shape means the flatness and the thickness of the steel plate which is affected by previous rolling, machine parameters, and control parameters. The current control method of the reversing cold rolling mill is based on the built-in fuzzy control system, which often results in a gap between the exit shape and the target shape. Since this is a long-term and unstable nonlinear system, it cannot be processed by differential equations. Therefore, based on past data, we use artificial intelligence methods to simulate and update the dynamic of the rolling system. We use XGBoost model as the virtual rolling mill. As long as the entry shape, machine parameters, and control parameters are given, the virtual rolling system can accurately simulate the exit shape. Based on the virtual rolling mill, we implement the Cross-Entropy algorithm to compute the appropriate control parameters that make the exit shape and the target shape as similar as possible to improve the process quality.