Topic 1: Digital Twin of Manufacturing System
In recent manufacturing industries, intelligent production technologies are rapidly being adopted to accurately predict process conditions and respond proactively to changes. As a result, digital twin–based process optimization and monitoring have become key tools in modern manufacturing systems.
A digital twin aims to accurately replicate real manufacturing processes in a virtual environment. To achieve this, physics-based models of key machine components—such as the controller and feed drive system—are developed together with process models that describe machining phenomena. These models are built in a modular manner and integrated to form a complete digital twin.
The figure below shows the main modules of a digital twin for a milling machine and example simulation results. Using this digital twin, various machining-related phenomena—such as machined geometry, cutting forces, surface quality, vibrations, and tool wear—can be predicted in advance.
Topic 2: Physical-Informed AI with Evolving Digital Twin
Real machine tools change their dynamic behavior over time due to factors such as thermal effects, wear, tool condition variations, and system aging. To address this, we develop an Evolving Digital Twin (EDT) with AI-based parameter updating, enabling the digital twin to continuously adapt to the actual machine condition.
Using the EDT, machining outcomes under various cutting conditions can be predicted accurately in a virtual environment. This allows efficient process optimization by applying algorithms such as Genetic Algorithms, Particle Swarm Optimization, and Bayesian Optimization to determine optimal machining parameters that balance productivity, surface quality, and process stability.
In addition, the EDT is used to generate reliable virtual data for physics-informed AI (PIAI)–based process monitoring, alleviating data scarcity and poor generalization issues in conventional machine learning approaches. By integrating multi-sensor information through Parallel Multilayer Data Fusion (PMDF), the proposed framework enhances the robustness and reliability of machining process state monitoring.
The figure below shows the overall framework for EDT-based prediction, optimization, and PIAI-based process monitoring.
Topic 3: Smart Factory
The figure below shows an overview of a digital twin–based unmanned machining system consisting of a machining center (MCT), an automated guided vehicle (AGV), and a collaborative robot. Tasks traditionally performed by human operators—such as machine door operation, chip removal, workpiece handling, and panel control—are autonomously executed by the collaborative robot mounted on the AGV.
The digital twin generates optimal robot paths and machining conditions while simulating machine states (e.g., MCT axis loads and robot joint torques) and machining quality, including geometric errors and surface finish. By comparing simulation results with real-time sensor data, abnormal situations such as collisions, tool breakage, and chatter can be predicted and diagnosed. The digital twin is continuously updated, enabling autonomous responses through coordinated robot actions.
The main research focuses on (1) developing an integrated digital twin that accurately models the dynamics, geometry, and machining processes of the unmanned system, and (2) generating optimal task paths and monitoring execution using digital twin–based prediction combined with reinforcement learning and optimization techniques.
