AI-powered quality analytics and control in 3D additive manufacturing

Ensuring consistent quality in 3D additive manufacturing(AM) is challenging due to process variability, material inconsistencies, and environmental fluctuations. To address these issues, this study proposes an AI-powered quality analytics and control framework integrating in-situ sensing, data-driven learning, and real-time process optimisation to improve the accuracy and reliability of 3D-printed components. The framework uses a dataset of 1480 process and quality records collected from an AM system, including parameters such as nozzle temperature, print speed, feed rate, vibration, and environmental metrics.Layer geometry and quality data are captured using in-situ point cloud scanning and pre-processed to remove environmental noise. The artefact surface is divided into a 6×6 grid, where each segment is classified as under-printed, normally printed, over-printed, or empty. A Fire Hawk Optimiser-driven Multi-hop Attention Graph Neural Network(FHO-MAGNN) evaluates segment-wise quality to support real-time process optimisation. The model demonstrates fast inference with a time of 0.48 ms.Compared with baseline models including RF,CNN, and 1D-CNN, the proposed FHO-MAGNN achieves improved segment-wise prediction with R² values between 0.90 and 0.95, indicating higher accuracy and reliability. Implemented in Python, the approach supports real-time quality assurance and reduces material waste. These results highlight the potential of AI-driven analytics for improving reliability in 3D additive manufacturing.