A Data-Driven Digital Twin for Predicting Manufacturing Process Efficiency

Manufacturing process optimization is often limited by the limited availability of high-quality models. To address this challenge, we propose a methodology to build digital twins exploiting the data collected from manufacturing processes. This paper presents a data-driven digital twin framework that predicts energy consumption and production time for Computer Numerical Control (CNC) drilling operations. A real-world dataset from drilling experiments on 3D-printed polymer workpieces was augmented using a Conditional Wasserstein Generative Adversarial Network with Gradient Penalty (cWGAN-GP) to ensure both physical plausibility and statistical consistency. A Deep Neural Network (DNN) trained on the enriched dataset learned the nonlinear relationships between process parameters and efficiency metrics, outperforming traditional machine learning models with an R2 score greater than 0.96. The trained model is integrated into a simulation framework (i.e., Frost) as a component to build a digital twin of a manufacturing process. This integration enables rapid design space exploration and informed decision-making.