Machine learning based predictive modeling of dimensional quality in direct energy deposition with SUS316L
Cited 0 times inCited 0 times in
- Machine learning based predictive modeling of dimensional quality in direct energy deposition with SUS316L
- Choi, Tae-Yang
- Kim, Namhun
- Issue Date
- Graduate School of UNIST
- This study aims to predict the dimensional quality of Direct Energy Deposition (DED) process on single and multi-track which are basis for the final product via machine learning. DED is a complex process of spraying powder onto a substrate and melting the material through a laser. Many process parameters (Laser power, Powder feed rate, etc.) affect output quality such as geometry (width, height, angle), mechanical properties (relative density, tensile strength, etc.). In order to see this effect, the DOE method, in which only one result is correlated with multiple factors, is used before, but machine learning is more effective in additive manufacturing in which multiple qualities needs to be predicted simultaneously.
In this study, a predictive model was generated through machine learning by using process parameters (Laser power, Powder feed rate, Coaxial gas) and dimensional qualities(Width, Height, Angle for single track, Height for multi-track) for the input and output data, respectively. After collecting the data, we trained the model using the five algorithms, Support Vector Machine(SVM), Random Forest(RF), Gradient Boosting Regression Tree(GBRT), and Artificial Neural Network(ANN), most commonly used as regression models in machine learning. After examining and comparing each generated prediction model through a goodness-of-fit test, the model generated using ANN was finally selected. When the selected model predicted the height of the multi-track most prominently, the r-square was as high as 96.63%. Afterwards, more than 4000 new datasets were created to derive optimal process parameter that met the dimensional objectives. The results are 300 W, 3.7 g/min, and 6 l/min for laser power, powder feed rate, and coaxial gas, respectively.
By using machine learning to predict output more accurately and faster than conventional methods, optimal process variables can be effectively derived in terms of time and cost. And, ultimately, it can be a cornerstone in researching future technologies that change process parameters in real time and monitor output results.
- Department of Mechanical Engineering
- Go to Link;
- Appears in Collections:
- Files in This Item:
Machine learning based predictive modeling of dimensional quality in direct energy deposition with SUS316L.pdf
can give you direct access to the published full text of this article. (UNISTARs only)
Show full item record
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.