Physics-informed data-driven models for defect detection and process optimization in metal additive manufacturing (AM)
Loading...
Date
2026-07-10
Authors
Πορευόπουλος, Βασίλειος
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The adoption of wire laser-based Directed Energy Deposition (Wire DED-LB) Additive Manufacturing (AM) processes by the industry rises rapidly since they support high deposition rates and resource-efficient fabrication of parts. Despite these benefits, the holistic adoption of the process is bounded by the strong dependence of process stability on heat accumulation and working distance deviations, which affect the melt pool dimensions, the bonding of successive layers and finally the part quality. For the better understanding of the effect of process variables on process phenomena and the optimization of part quality, this thesis establishes workflows to correlate in-line process monitoring data with a physics-based simulation tool, enabling automated defect detection and process control. The first step of the structured methodology involves the generation of process data coming from the sensors of the experimental set up, consisting of an off-axis vision camera and an embedded load cell. These sensors capture the melt pool features at the top surface of the fabricated part while the load cell captures the forces applied on the wire feeding system. The monitoring data and the extracted knowledge are linked to the data generated by a physics-based simulation tool. The information extracted from the latter has been previously validated against in-situ melt pool data, aiming to correlate the calculated subsurface temperature field to the monitored upper surface characteristics and thus, evaluate the melt pool behavior at critical depth intervals. Recognizing the surface-to-subsurface quality characteristics, several control strategies are developed and evaluated across progressive stages. While single-variable regulation causes control saturation in high-aspect-ratio geometries, advanced multi-variable architectures tuning laser power and interlayer dwell times effectively eliminate process instabilities and consequently the surface quality of the end-part. The final control strategy incorporates moving averages and multi-level severity classifications to handle historical trends and minimize transient instabilities. The validation results point out an excellent agreement between the simulation driven targets and real-world experimental trials, paving the way for the development of physics-informed data-driven models for defect detection and process optimization.
Description
Keywords
Additive manufacturing, Wire laser-based directed energy deposition, Heat accumulation, Process data, Laser power, Dwell time