Answers are grounded in this paper and cite their sources.
This article proposes an approach that uses machine learning to determine the optimal process parameters for the laser deposition of thin walls. The following machine learning models were tested: gradient boosting implemented with CatBoost, a fully connected neural network, and a recurrent neural network with long short-term memory block. To obtain training dataset data, thin walls were deposited from stainless steel. Data on wall geometry (height, width, standard deviation of height), obtained through optical monitoring, and process parameters - laser power, scanning speed, and wire feed rate - were used. For the recurrent neural network, a modified dataset was used, including characteristics of the current layer and the history of previous layers. The study showed that with a history depth of 9 layers, the recurrent neural network achieved the highest prediction accuracy, outperforming gradient boosting and the fully connected neural network across all metrics: the mean absolute error for predicting laser power was 1.5 percentage points, for scanning speed of 0.072 mm/s and for wire feed rate of 0.064 mm/s. The results demonstrate that incorporating the temporal history of previously deposited layers significantly improves the prediction of process parameters compared with models that consider each layer independently. The proposed approach provides a promising basis for adaptive closed-loop control of laser metal deposition using optical monitoring
No takes yet. Share an insight, caveat, or question.
Bannikov et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: