A Deep Learning Model for Estimating the Quality of Bimetallic Tracks Obtained by Laser Powder-Directed Energy Deposition

During the fabrication of Inconel 718-AISI 316L bimetallic components via laser powder-directed energy deposition, understanding the relationships between processes, microstructures, and material properties is crucial to obtaining high-quality parts. Physical-chemical properties, cooling rates, and thermal expansion coefficients of each material may affect the microstructure of parts, generating segregations and cracks. This paper analyzes how the process parameters affect the dimensions, chemical composition, and microhardness of bimetallic tracks. We created a dataset that included laser power, powder feed rate, material, skeletal density, dimensional features, chemical composition, and microhardness. Then, a deep learning methodology using a multilayer perceptron was used to estimate the relationship between these factors. The architecture comprised four inputs in the input layer and five hidden layers with 20, 40, 30, 30, and 30 neurons, respectively. This architecture was used to estimate the dimensional features, chemical composition, and microhardness. The model precision was evaluated using the determination coefficient (R2) and the mean absolute error (MAE) function. Lastly, we used a random forest classifier to select the bead quality from the optimal process parameters. The results showed a significant decrease in training loss and validation loss between 50 and 100 epochs. This decreasing trend continued until 350 epochs. This paper contributes to understanding the relationships between process-structure properties in the bimetallic tracks of Inconel 718 and AISI 316L.