About this Abstract |
| Meeting |
MS&T25: Materials Science & Technology
|
| Symposium
|
Additive Manufacturing Modeling, Simulation, and Machine Learning: Microstructure, Mechanics, and Process
|
| Presentation Title |
Neural Network-Based Optimization of Stepover Distance for Wire-Arc Additive Manufacturing |
| Author(s) |
Rida Adhami, Kaue Riffel, Rakhi Bawa, Justin Chan, Daniil Gofman, Antonio Ramirez |
| On-Site Speaker (Planned) |
Rida Adhami |
| Abstract Scope |
In wire-arc directed energy deposition (DED), selecting optimal parameters like travel speed, wire feed speed, and stepover is critical for consistent bead geometry and high-quality multilayer builds. Traditional tuning relies on expert knowledge and costly trial-and-error. This work presents a supervised machine learning framework that predicts optimal stepover distance using a small experimental dataset. Starting with a 16-sample Design of Experiments (DOE), later expanded to 24, Ridge regression models generate 500+ synthetic samples spanning wire feed and travel speeds. These samples map to deposition shape descriptors such as valley area and bead height ratio. A feedforward neural network trained on the synthetic dataset performs multi-target regression to evaluate deposition quality across stepover values. A graphical user interface (GUI) enables users to input process parameters and visualize predicted outcomes. This method enables faster, data-driven parameter optimization in additive manufacturing. |