Spatiotemporal deep learning using infrared thermography for real-time monitoring and process parameter adjustment in material extrusion

Material Extrusion (ME) additive manufacturing has enabled the production of complex structures across various industries due to its low cost and versatility. However, ME processes are prone to defects such as under-extrusion and over-extrusion, which adversely affect the quality and performance of printed parts. Traditional monitoring methods rely heavily on human operators and lack real-time feedback mechanisms, which leads to inefficiencies and increased production costs. To address these challenges, we present a novel methodology that leverages deep learning to extract spatiotemporal features from thermographic data for real-time process deviation detection and correction in ME processes. An experimental setup is established alongside a novel data acquisition and preprocessing framework to consistently capture and store thermal data during printing. We designed and trained three deep learning architectures focusing on spatial and spatiotemporal feature extraction using preprocessed offline data. Our results demonstrate that incorporating spatiotemporal features significantly improves the accuracy and responsiveness of process deviation detection compared to existing methods. Furthermore, we report a closed-loop control system based on infrared thermography that utilizes the best-performing spatiotemporal deep learning model with 99.25% validation accuracy to categorize the severity of under-extrusion and over-extrusion in real-time and implement corrective actions. The proposed closed-loop control system demonstrates dynamic adjustment of the relative flow rate (RFR) within 3 s of process deviation and ensures consistent production of high-quality parts by reducing percentage volumetric change by a factor of 10. This work advances real-time quality control in ME processes and paves the way in achieving first-time-right manufacturing by reducing waste, lowering costs, and enhancing overall production efficiency.