MACHINE LEARNING IN DIGITAL TWIN FOR DETECTION AND PREDICTION OF DEFECTS IN METAL 3D PRINTING

Abstract

The Directed Energy Deposition (DED) 3D printing process has emerged as an innovative technology for manufacturing complex metal components in today's industries. However, its broader industrial adoption is limited by challenges such as defects in printed parts, material wastage, and inefficiencies in production processes due to some problems that will be mentioned afterward. This project aims to address these issues by developing a machine-learning-enhanced digital twin concept for real-time defect detection and prediction. The proposed digital twin concept integrates machine learning models, simulations from ANSYS Workbench, and a digital twin of the DED printer into one system that will predict the crack anomalies in an object. The machine learning models are trained on simulated data that will be taken from ANSYS to predict defects like cracks and deformation based on input parameters. A user-friendly software interface will connect all components of the system. It enables real-time monitoring and control over the printing object and lets us choose optimized input parameters that will give no cracks. Future work will include designing a detailed 3D model of the DED printer in SOLIDWORKS and expanding the database with experimental data collected from real-world printing. Moreover, the machine-learning algorithms will be refined for improved accuracy. By validating the system through physical tests, this project seeks to enhance the reliability, efficiency, and sustainability of the DED 3D printing process. That will contribute to its wider adoption in industrial applications.