Abstract

We propose a mechanistic Artificial Intelligence (AI) framework, called Hierarchical Deep Learning Neural Networks or HiDeNN-AI [1,2,8] for discovering the multiscale linkage of process-structure-property of additive manufacturing systems. The HiDeNN-AI discovery has three sequentially executed steps: (1) using available data to characterize an unknown physical process in manufacturing, (2) enriching the database and training with mechanistic knowledge coming from the system identification in step (1) with uncertain parameters to create a reduced order model with uncertainty, and (3) using the reduced order model to generate sufficient data to discover new robust mathematical principles that are able to (a) perform predictive solutions for design and optimization, and (b) provide simple relationship for online monitoring and control. We have applied this HiDeNN-AI framework to address the Air Force Research Lab (AFRL) AM modeling challenges [3, 4, 5]; and for the prediction of the as-built mechanical properties [6]. To further enhance HiDeNN-AI, a reduced-order modeling method accounting input uncertainty, called the Tensor Decomposition (TD) [7], is being developed. The so-called HiDeNN-AI-TD is expected to solve the general engineering design and manufacturing problems in high dimensional space-time-parametric domains at deep discount in computational cost. Once the offline database is set up, the mechanistic machine learning module of HiDeNN-AI can be activated for process design, real time system monitoring and control or the identification of key processing parameters for the desired performance of the manufactured material systems with uncertainty quantification. Various results comparing the HiDeNN-AI-TD approach with the conventional machine learning models will be shown using real-time IR in-situ measurement, and high-frequency thermal signatures for the predictions of mechanical properties and the detection of lack of fusion and keyhole porosities. Similar applications to polymer matrix composites will be presented.

References

[1] Zhang, L., Cheng, L., Li, H., Gao, J., Yu, C., Domel, R., Yang, Y., Tang, S. and Liu, W.K., 2021. Hierarchical deep-learning neural networks: finite elements and beyond. Computational Mechanics, 67(1), pp.207-230.

[2] Saha, S., Gan, Z., Cheng, L., Gao, J., Kafka, O.L., Xie, X., Li, H., Tajdari, M., Kim, H.A. and Liu, W.K., 2021. Hierarchical Deep Learning Neural Network (HiDeNN): An artificial intelligence (AI) framework for computational science and engineering. Computer Methods in Applied Mechanics and Engineering, 373, p.113452. https://doi.org/10.1016/j.cma.2020.113452.

[3] Gan, Z., Jones, K.K., Lu, Y., and Liu, W.K., 2021, Benchmark study of melted track geometries in laser powder bed fusion of Inconel 625, Integrating Materials and Manufacturing Innovation, 10, pages 177–195.

[4] Saha, S., Kafka, O.L., Lu, Y., Liu, W. K., 2021, Microscale Structure to Property Prediction for Additively Manufactured IN625 through Advanced Material Model Parameter Identification, Integrating Materials and Manufacturing Innovation, 10, pages142–156 (2021).

[5] Saha, S., Kafka, O.L., Lu, Y., Liu, W. K., 2021, Macroscale Property Prediction for Additively Manufactured IN625 from Microstructure through Advanced Homogenization, Integrating Materials and Manufacturing Innovation, 10, pages 360–372 (2021). (https://doi.org/10.1007/s40192-021-00221-8)

[6] Xie, X., Bennett, J., Saha, S., Lu, Y., Cao, J., Liu, W. K., Gan, Z., 2021, Mechanistic data-driven prediction of as-built mechanical properties in metal additive manufacturing, npj Computational Materials, 7(1), 1–12. https://doi.org/10.1038/s41524-021-00555-z

[7] L. Zhang, Y. Lu, S. Tang, WK. Liu, HiDeNN-TD: Reduced-order hierarchical deep learning neural networks, Computer Methods in Applied Mechanics and Engineering, 389, (2022) 114414

[8] Wing Kam Liu, Zhengtao Gan, Mark Fleming, “Mechanistic Data Science for STEM Education and Applications,” Springer, 2021