.png){kind=icon}
.png){kind=icon}
Return
Implementation of a physics-informed neural network to predict relative densities for copper LPBF with a green laser system
.png){kind=icon}
OA
AI
M
L
S
E
T
E
DOI：10.1108/rpj-08-2024-0343.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
<jats:sec>
<jats:title>Purpose</jats:title>
<jats:p>The purpose of this paper is to present the implementation of a physics-informed neural network (PINN) to predict the relative density of copper following the laser powder bed fusion process (LPBF).</jats:p>
</jats:sec>
<jats:sec>
<jats:title>Design/methodology/approach</jats:title>
<jats:p>A 532 nm laser system is used to manufacture specimens from pure copper powder. The process parameters are varied to reduce the number of pores in the specimens. A specified set of the most relevant process parameters is used. Data from these experiments and literature data are combined to train the PINN. A control data set is used to compare the predictions of the PINN to experimentally determined data, as well as the predictions of other methods, like linear regression and a conventional artificial neural network.</jats:p>
</jats:sec>
<jats:sec>
<jats:title>Findings</jats:title>
<jats:p>PINN are capable to predict the relative density of parts based on a limited amount of data. For predictions of the relative density for copper samples manufactured on the 532 nm LPBF system, the RMSE is 2.19 percentage points for an artificial neural network, 1.80 percentage points for a linear regression and 1.46 for the implemented PINN. Similar improvements in the quality of predictions can be found for predictions based on data extracted from published research.</jats:p>
</jats:sec>
<jats:sec>
<jats:title>Research limitations/implications</jats:title>
<jats:p>The research conducted for this paper is limited to the LPBF process, but the implications suggest applicability to a much broader range of technologies which are limited by the number of possible experiments and are too complex to model or simulate otherwise.</jats:p>
</jats:sec>
<jats:sec>
<jats:title>Practical implications</jats:title>
<jats:p>PINNs pose the opportunity to analyze systems which could otherwise not be analyzed properly because of a lack of data. Therefore, the presented PINN approach can facilitate the development of a variety of other processes with similar problems, while reducing the number of experiments necessary to generate training data.</jats:p>
</jats:sec>
<jats:sec>
<jats:title>Originality/value</jats:title>
<jats:p>To the authors knowledge, no PINN has been applied for the prediction of process outcomes in LPBF based on process parameters, especially in the field of the additive manufacturing of pure copper.</jats:p>
</jats:sec>
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
3.6
Papers:
2.2K
Citations:
7.7K
Organization
Cited Papers