Page 20 - IJAMD-2-2

International Journal of AI for
Materials and Design AI-driven material development for AM

From fatigue strength modeling to process parameter driven learning, will be essential for further improving
optimization, AI has proven to be a powerful tool in predictive accuracy and expanding the applicability of AI
enhancing microstructural control and mechanical property in AM alloy design.
refinement. Despite these advancements, challenges
remain in data completeness, model interpretability, 5.2. Polymer materials for AM
and generalization across different AM processes and Table 5 compiles a range of AI-driven strategies applied
material systems. The continued development of hybrid AI to polymer AM, covering both compositional design and
approaches, integrating physics-based models with data- property improvement.

Table 5. Summary of artificial intelligence applications for polymer materials for additive manufacturing
AM Material Optimization AI method Target Model performance References
process type
PBF Multi-material Design GMM and PCR Elastic property Poisson’s ratio error≈16% Chen et al. 81
optimization
VPP RPU+SilDN Design VAE and BO Elastic moduli tailoring Poisson’s ratio error≈6.4%, E Xue et al. 82
(Young’s modulus and error≈11.3%
Poisson’s ratio)
VPP TPU Design MLP Novel metamaterial with - Fleisch et al. 83
variable compression
properties design
MET PLA Performance Bayesian ML Super-compressibility and R2≈0.988 Bessa et al. 84
recoverability design
MET PLA Performance RF, KNN, ADA, Tensile and flexural LSTM achieved best Sharma et al. 85
DT, and LSTM strength prediction and performance: R =0.9169,
2
optimization MAPE=2.85%, RMSE=2.44;
other models (RF, KNN, ADA,
DT) showed R <0.75 and
2
MAPE >5%.
MET Technomelt PA Performance LiR, GPR, RR, Tensile property prediction LiR/RR best: <10% error Nasrin et al. 86
6910 and KNN (Young’s modulus, yield
stress, yield strain, tensile
stress, and tensile strain)
MET ABS Performance LiR, DT, RF, and Hardness prediction RF best: R² ≈ 0.91, RMSE≈0.99, Veeman et al. 87
ADA AdaBoost close: R² ≈ 0.90,
RMSE≈1.09
LR & DT lower: R² ≈ 0.84 &
0.77
MET PLA Performance CNN and RF Process parameters– RF accuracy: 94% (TS), 89% Butt and Mohaghegh 88
property correlation (hardness); CNN 88% (TS), 88%
(TS and hardness) (hardness)
MET PLA Performance LSTM TS prediction R =89.4% Zhang et al. 89
2
VPP Resin Performance GA+NN Modulus and strength R =0.9978 Lee et al. 90
2
optimization
MJT Multi-material Performance GA Tunable deformation and - He et al. 91
antibacterial performance
MJT Multi-material ANN and RSM Shore hardness and ANN: MSE=0.36% (Shore A), Goh et al. 92
compressive modulus 0.98% (E); RSM: MSE=1.3%
optimization (Shore A), 4.4% (E)
Abbreviations: ABS: Acrylonitrile butadiene styrene; ADA: Adaptive design algorithm; ANN: Artificial neural network; BO: Bayesian optimization;
CNN: Convolutional neural network; DT: Decision trees; GA: Genetic algorithms; GMM: Gaussian mixture model; GPR: Gaussian process regression;
KNN: K-nearest neighbors; LiR: Linear regression; LSTM: Long short-term memory; MAPE: Mean absolute percentage error; MET: Multi-exponential
theory; MJT: Material jetting; ML: Machine learning; MLP: Multilayer perceptron; MSE: Mean squared error; NN: Neural network; PBF: Powder bed
fusion; PCR: Principal component regression; PLA: Polylactic acid; RF: Random forest; RMSE: Root mean square error; RPU: A commercial hard
polyurethane; RR: Ridge regression; RSM: Response surface methodology; SilDN: A custom soft silicone; TPU: Thermoplastic polyurethane;
TS: Tensile strength; VAE: Variational autoencoder; VPP: Vat photopolymerization.

Volume 2 Issue 2 (2025) 14 doi: 10.36922/IJAMD025100007