About this Abstract |
| Meeting |
MS&T25: Materials Science & Technology
|
| Symposium
|
Additive Manufacturing Modeling, Simulation, and Machine Learning: Microstructure, Mechanics, and Process
|
| Presentation Title |
Prediction Optimal Parameters for Wire-ARC DED Welding Using Multilayer-Perceptron Trained on Synthetic Data |
| Author(s) |
Daniil Gofman, Antonio Ramirez |
| On-Site Speaker (Planned) |
Daniil Gofman |
| Abstract Scope |
The global demand for 3D printing technologies has grown significantly, influencing various industries and extending into the welding sector through additive manufacturing (AM). AM involves constructing metal parts layer by layer using wire or powder feedstock. This method offers advantages such as material efficiency and design flexibility, but bead overlapping between layers often results in defects like surface irregularities and weakened structural integrity.
This paper presents an automated approach to optimize welding parameters using a multi-layer perceptron (MLP) neural network. Traditional methods for selecting parameters like current, voltage, and wire feed rate are time-consuming. To overcome limited data availability, synthetic data is generated using a generative adversarial network. By training the MLP on synthetic data, the system predicts parameters that improve bead quality and inter-layer bonding. This approach helps to save on resources by reducing the need for extensive physical experimentation, minimizing material waste, and shortening development cycles. |