In Direct Energy Deposition (DED), the melt pool temperature is a critical control parameter that affects deposition rate, porosity formation, residual stress, and microstructure in the final parts. In this thesis, a data-driven approach using Machine Learning (ML) models is used to predict the melt pool temperature using experimental data. This thesis presents the integration of the laser-based DED system using metal powder feedstock, the determination of the process parameter window for the setup, and the development of an ML pipeline to predict the melt pool temperature based on its history.

In the system integration for the DED system, a laser generator, powder feeder, deposition head, and sensors (i.e., an IR camera and a 2-wavelength pyrometer) were integrated into an existing 3-axis motion stage. Python-based software was developed to control the laser generator and to read data from the sensors. The software calibrates the IR camera’s temperature, which is highly dependent on the emissivity, by leveraging the data from the 2-wavelength pyrometer. To determine the process parameter window, 150 single-layer clads were deposited;​ clads’ crosssections were polished and etched and optical microscopy was used to measure the clad’s height, melt pool’s depth, and dilution ratio. Analysis was conducted on the correlation of the process parameters, laser power, scan speed, flow rate, and the measured properties of the clads. The process parameters with the minimal dilution of (5-25%) were selected to obtain clads with proper geometry and bonding to the substrate.

Finally, the temperature data of a 6-layer thin wall with the obtained process parameters were used to train several ML models, including Dense Neural Networks (DNN), 1-Dimensional Convolutional Neural Networks (1D-CNN), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). LSTM shows better performance among these models;​ therefore it was implemented in the ML pipeline for temperature prediction. The Model can predict the trend and fluctuations of the melt pool temperature with higher accuracy compared to the existing models for melt pool temperature prediction in the DED process.