Page 23 - ESAM-1-1

Engineering Science in
Additive Manufacturing ML in MAM monitoring and control through images

consistent part quality. Factors such as appropriate powder Table 4. Summary of process control in MAM
layer thickness, particle size distribution, and powder
diffusion technology play a crucial role in optimizing melt Control algorithms LPBF DED
pool behavior and minimizing defects. PID 168 169-171
Sliding mode control 172,173 174
Cruz et al. utilized an artificial neural network
161
(ANN)-based logic controller to adjust welding speed Predictive control 175-178 22,179,180
in WAAM, ensuring the desired bead width and high Adaptive control 181,182 158,183,184
printed sample quality. Kershaw et al. developed an Rule-based 185,186 160,187
162
ML framework for controlling bead width by adjusting Abbreviations: DED: Direct energy deposition; LPBF: Laser powder
welding speed. Zhang et al. proposed a robust control bed fusion; MAM: Metal additive manufacturing; PID: Proportional
163
system driven by ML to regulate deposition height through integral derivative.
adjustments in arc current, travel speed, and external
wire feed speed. Tang et al. implemented a topography 5. Challenges and prospects
164
control strategy using a deep-image method to adaptively 5.1. Challenges
address surface topography quality, handling concave
and convex parts during the WAAM process. To address By analyzing the image data with ML algorithms, researchers
accumulation errors during printing, Dharmawan et al. can monitor the formation and solidification process of the
165
suggested a framework for iteratively learning the effects of melt pool in real time and identify potential defects such
process parameters on printing and correcting inter-layer as porosity, cracks, or unevenness. ML technology can
geometric deviations based on RL, ensuring satisfactory help predict the occurrence of these defects and adjust the
final output. Notably, Mireles et al. developed an process parameters in advance to avoid the occurrence of
166
ML-driven automatic feedback control system to manage defects. The quality and stability of the MAM process can
powder bed temperature variations in EBM through be improved through real-time feedback and automatic
modifications in processing parameters. control. Still, several challenges remain in the integration
of ML, limiting the efficiency of the overall process.
4.3. Others
In MAM, besides controlling the melt pool and deposited 5.1.1. Multi-source data fusion
layers, several other crucial control aspects demand Due to the complex non-linearity, robust coupling, and
consideration. For example, Imani et al. presented an multi-parameter properties of MAM, many factors
167
ML-based technique that successfully described surface have a complex effect on faults in as-built components.
finishes for ultraprecision machining quality control and Unfortunately, existing systems often restrict themselves
created a model that connected process parameters with to monitoring individual feature signals, leading to
fractal features of in-process photos. This linkage enables unreliable detection results. Overcoming this limitation
swift responses to process changes, subsequently reducing necessitates the implementation of advanced multi-sensor
the number of defective products. Moreover, atmosphere synchronous sensing techniques to seamlessly integrate
control plays a key role. Maintaining an appropriate data from diverse sources, enhancing the reliability of
printing environment atmosphere is essential for reducing defect detection results. Nevertheless, the complex
6,30
oxidation and pollution, enhancing printing quality, and interplay of various signals requires careful clarification, as
minimizing defects. mutual interference gives rise to numerous uncertainties
In summary, process control in MAM is essential and noise within the captured images.
for maintaining quality, consistency, and efficiency in 5.1.2. Limited amount of data
the production process. By monitoring and adjusting
parameters such as material quality, process settings, The scarcity of data can significantly impact the efficacy
temperature, and melt pool behavior, manufactured and performance of these models, impeding their ability
components with improved qualities can be achieved. Key to generalize, optimize process parameters, and accurately
aspects of control include melt pool control to manage melt identify crucial information within images. To begin with,
pool geometry and temperature, deposited layer control a constrained dataset can impede the model’s capacity
to ensure the quality and accuracy of each layer, and to generalize, potentially resulting in overfitting of the
atmosphere control to sustain appropriate environmental training data and consequently producing failed results
conditions and minimize oxidation and contamination. in practical applications. Moreover, feature extraction
Table 4 provides a summary of closed-loop control systems becomes challenging as a substantial number of data is
in MAM. typically required to accurately identify and represent

Volume 1 Issue 1 (2025) 17 doi: 10.36922/esam.8548