Page 46 - IJAMD-2-2

International Journal of AI for
Materials and Design ML-driven optimization in additive manufacturing

Table 2. Summary of ML models for metal‑based 3D printing
Motivation Material AM ML model Model inputs Remarks References
method
Process SS316L, Ti-6Al-4V PBF Deep reinforcement Heat distribution, melt pool depth Control process parameters 138
optimization learning such as laser velocity and
power in real time
Metal powder PBF Multi-input neural Thermal images, process Feedback control based on 139
network parameters (scan vector length, multi-input data to achieve
heat trace length, heat transfer the desired thermal history
distance, cumulative scan time)
®
CarTech 718 DED XGBoost, LSTM Laser power, scan speed, layer Real-time melt pool 140
temperature prediction
index, time index, average height,
superalloy
average width
Ti-6Al-4V DED ConvLSTM Video frames of melt pool Detect melt pool anomalies 141
Autoencoder dynamics (wire dripping, arcing,
oscillation, stubbing) for
quality control
Inconel 625 PBF C-RNN Melt pool video frames Improve defect detection 142
using spatiotemporal analysis
of melt pool videos
SS316L, Ti-6Al-4V, PBF Vision transformer High-speed imaging of melt pool Improve process 143
Inconel 718 dynamics development and defect
detection in 3D printing of
new metal alloys through
in situ process mapping
SS316L DED CNN Coaxial camera images of melt Improve fault detection by 144
pool monitoring nozzle clogging
and abnormalities in the
powder stream
Inconel 718 DED Bayesian LSTM, Temperature history, laser power Develop a digital twin 150
BOTSPO profile, spatial location framework for real-time
predictive control, enhancing
process stability and efficiency
Property SS316L PBF DT (CART), RF, Build location, post-chamber Quality prediction based on 148
optimization XGBoost pressure drop, powder properties, the combination of build
gas flow characteristics location and post-chamber
pressure drop
Ti-6Al-4V PBF GPR, BO Laser power, scan speed, hatch Discovered an expanded 149
spacing, porosity data processing window that
optimizes mechanical
properties and density
Inconel 625 DED CNN, YOLOv8 Melt pool video frames, bead Predict geometric 150
geometry data characteristics and analyze
their correlation with process
parameters and bead geometry
for real-time process control
Ti-6Al-4V DED CGAN Laser power, powder feed rate, scan Predict and optimize surface 151
speed morphology to improve
surface quality and reduce cost
Design ER70S-6 steel DED SVR, NSGA-II Laser power, travel speed, wire Optimize process parameters 152
optimization feed rate for achieving desired layer
geometry
Abbreviations: BO: Bayesian optimization; BOTSPO: Bayesian optimization for time series process optimization; C-RNN: Convolutional recurrent
neural network; CART: Classification and regression tree; CGAN: Conditional generative adversarial network; CNN: Convolution neural network;
ConvLSTM: Convolutional long short-term memory; DED: Directed energy deposition; DT: Decision tree; GPR: Gaussian process regression;
LSTM: Long short-term memory; ML: Machine learning; NSGA-II: Non-dominated sorting genetic algorithm-II; PBF: Powder bed fusion;
RF: Random forest; SVR: Support vector regression; XGBoost: Extreme gradient boosting; YOLO: You Only Look Once.

Volume 2 Issue 2 (2025) 40 doi: 10.36922/IJAMD025130010