AI and Infrared Tech Boost Metal 3D Printing

By Dr Silpaja Chandrasekar, PhDReviewed by Susha Cheriyedath, M.Sc.Jun 7 2024

In a paper published in the journal Materials & Design, researchers delved into metal additive manufacturing's thermal intricacies, highlighting the critical role of heat management in ensuring part quality. They proposed an innovative approach using affordable infrared (IR) imaging and machine learning (ML) for real-time detection and optimization of heat accumulation zones. Experimental validation across diverse geometries underscored the method's robustness and potential for enhancing the quality and reliability of printed parts.

Background

Previous research extensively explored metal additive manufacturing's thermal complexities. Powder bed fusion with laser beam (PBF-LB) emerged as a key technology, but concerns persisted due to variability. Thermal cycles and geometric features influenced heat accumulation, impacting part properties—IR thermography aided in defect detection and process control.

However, accurate detection remains challenging due to calibration issues. Automated approaches are proposed for real-time tracking and identification of localized heat accumulation events, aiming to enhance part quality. Further advancements in monitoring and control techniques are essential for widely adopting metal additive manufacturing.

Real-Time Heat Detection

In the experimental setup conducted at the Carnegie Mellon next manufacturing center off-campus laboratory, the Trumpf GmbH + Co. KG (TRUMPF ) TruPrint 3000 powder bed fusion - laser beam (PBF-LB) machine was utilized alongside a long-wavelength IR camera (FLIR A655sc) for real-time process monitoring. Due to space constraints within the chamber, the analysts mounted an IR camera on the build chamber with a 15° inclination angle, equipped with a Germanium IR long-pass filter for optimal transmission efficiency.

For real-time detection, the IR camera operated at a continuous frame rate of 200 Hz, resulting in a high temporal resolution of 5 ms and a spatial resolution of 0.3 mm after image preprocessing. The detection algorithm comprised three steps: image preprocessing, scan zone (SZ) tracking and heat accumulation zone detection, and mapping heat accumulation zones to the computer-aided design (CAD) geometry. Background subtraction with adaptive Gaussian Mixture Models (GMM) was employed for SZ tracking, allowing real-time detection of moving objects amidst multimodal backgrounds.

Subsequently, the identified SZ were classified into normal zones and zones with accumulated heat based on statistical characteristics such as standard deviation and kurtosis of their intensities. Rather than relying solely on maximum intensity, the approach focused on features characterizing the shape of the intensity distribution within the SZ. Classification methods included thresholding statistical measures or utilizing unsupervised/supervised algorithms such as k-nearest neighbors, support vector machine (SVM), or artificial neural network (ANN). Finally, the detected zones of local heat accumulation were mapped onto the CAD geometry for further analysis and validation.

Heat Accumulation Insights

The study employed a test part featuring various geometrical intricacies prone to heat accumulation during printing, designed without support structures. The printed part's orientation within the chamber showcased diverse features, from walls of varying thicknesses to overhangs and a hollow cylinder, all designed to examine heat accumulation effects comprehensively.

Analysis of local heat accumulation zones within printed layers revealed significant insights into the thermal behavior during the process. Focusing on specific areas where heat accumulates, the study delved into factors influencing this phenomenon, including processing parameters, material properties, and intricacies of scan strategy. Thermal imaging unveiled distinct characteristics of zones prone to heat accumulation, displaying higher intensity and variance due to rapid temperature elevation and limited heat dissipation, particularly evident at intersections and thin-walled sections.

The proposed real-time detection algorithm demonstrated robust performance in identifying heat accumulation zones, with a high detection rate of 90% compared to manually labeled ground truth data. Additionally, comparative analysis with traditional thermal history-based methods showcased superior accuracy and efficiency of the real-time approach, paving the way for potential closed-loop control applications and computational efficiency in process optimization.

Exploration into the influence of stripe angles on heat accumulation provided valuable insights into optimizing scan strategies for minimizing localized heat buildup. The correlation between stripe angles and zone attributes highlighted the importance of scan vector lengths in allowing adequate cooling time, with wider angles resulting in more uniform thermal distributions and reduced heat accumulation.

Furthermore, the study's findings illuminate the intricate interplay between process parameters, geometrical features, and thermal dynamics, offering actionable insights for optimizing metal additive manufacturing processes. Manufacturers can enhance part quality, minimize defects, and improve overall process efficiency by understanding and mitigating the risks associated with localized heat accumulation. The comprehensive approach adopted in this study, combining experimental analysis with advanced real-time detection algorithms, underscores the importance of multidisciplinary research in advancing additive manufacturing technologies toward greater reliability and scalability.

Conclusion

To summarize, this study introduced an ML-based method for real-time detection of heat accumulation zones in metal additive manufacturing using IR imaging. It explored the influence of scan strategy on heat accumulation and its impact on porosity formation, validated through experiments on stainless steel 316L (SS316L) alloy parts. The algorithm achieved over 90% accuracy in detecting heat accumulation spots, revealing insights into the correlation between stripe angle, scan zone area, and average temperature.

The study emphasized the importance of optimizing scan strategy to enhance thermal uniformity and mitigate hot spots, particularly at edges and around local features. However, challenges remained in detecting heat accumulation in thin walls and small features, warranting further research. This work contributed to understanding and optimizing metal additive manufacturing processes for improved part quality and performance.