Transfer learning-enhanced physics informed neural network for accurate melt pool prediction in laser melting

The profile of the melt pool is essential in selective laser melting (SLM) to control the process quality and avoid defects. Physics informed neural network (PINN) method is proposed to address challenges in various science and engineering problems when traditional numerical calculations are time-co...

Full description

Saved in:
Bibliographic Details
Main Authors: Qingyun Zhu, Zhengxin Lu, Yaowu Hu
Format: Article
Language:English
Published: ELSPublishing 2025-01-01
Series:Advanced Manufacturing
Subjects:
Online Access:https://elsp-homepage.oss-cn-hongkong.aliyuncs.compaper/journal/open/AM/2025/am20250001.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849736305551867904
author Qingyun Zhu
Zhengxin Lu
Yaowu Hu
author_facet Qingyun Zhu
Zhengxin Lu
Yaowu Hu
author_sort Qingyun Zhu
collection DOAJ
description The profile of the melt pool is essential in selective laser melting (SLM) to control the process quality and avoid defects. Physics informed neural network (PINN) method is proposed to address challenges in various science and engineering problems when traditional numerical calculations are time-consuming, or deep learning (DL) methods have high demand for data. However, SLM process involves many complex physical phenomena. Low-fidelity data from low-fidelity models struggle to accurately reflect these phenomena, while high-fidelity data from high-fidelity models contains more physical equations, making it difficult for current PINN. This article proposed a transfer learning-enhanced PINN (TLE-PINN) method using high-fidelity data for precise and fast melt pool prediction. It contains the enhanced PINN (EPINN) and transfer learning framework. The EPINN model integrates the heat transfer law and boundary condition to loss function, imposing strong physical constraints on data. Then, the transfer learning framework, combining the concepts of PINN and DL, initially trains with PINN and then further fine-tunes it using DL method. Notably, it only uses a single model, which is more convenient to traditional methods that require two models. The developed solution demonstrates outstanding performance when compared with experiments and existing methods, showing significant potential for industrial applications.
format Article
id doaj-art-280bca0531db4907b1b3663b21dcd689
institution DOAJ
issn 2959-3263
2959-3271
language English
publishDate 2025-01-01
publisher ELSPublishing
record_format Article
series Advanced Manufacturing
spelling doaj-art-280bca0531db4907b1b3663b21dcd6892025-08-20T03:07:20ZengELSPublishingAdvanced Manufacturing2959-32632959-32712025-01-012110.55092/am202500011799796481847033856Transfer learning-enhanced physics informed neural network for accurate melt pool prediction in laser meltingQingyun Zhu0Zhengxin Lu1Yaowu Hu2The Institute of Technological Sciences, Wuhan University, 430072 Wuhan, ChinaThe Institute of Technological Sciences, Wuhan University, 430072 Wuhan, ChinaThe Institute of Technological Sciences, Wuhan University, 430072 Wuhan, ChinaThe profile of the melt pool is essential in selective laser melting (SLM) to control the process quality and avoid defects. Physics informed neural network (PINN) method is proposed to address challenges in various science and engineering problems when traditional numerical calculations are time-consuming, or deep learning (DL) methods have high demand for data. However, SLM process involves many complex physical phenomena. Low-fidelity data from low-fidelity models struggle to accurately reflect these phenomena, while high-fidelity data from high-fidelity models contains more physical equations, making it difficult for current PINN. This article proposed a transfer learning-enhanced PINN (TLE-PINN) method using high-fidelity data for precise and fast melt pool prediction. It contains the enhanced PINN (EPINN) and transfer learning framework. The EPINN model integrates the heat transfer law and boundary condition to loss function, imposing strong physical constraints on data. Then, the transfer learning framework, combining the concepts of PINN and DL, initially trains with PINN and then further fine-tunes it using DL method. Notably, it only uses a single model, which is more convenient to traditional methods that require two models. The developed solution demonstrates outstanding performance when compared with experiments and existing methods, showing significant potential for industrial applications.https://elsp-homepage.oss-cn-hongkong.aliyuncs.compaper/journal/open/AM/2025/am20250001.pdfselective laser meltingmelt pooldeep learningpinntransfer learning
spellingShingle Qingyun Zhu
Zhengxin Lu
Yaowu Hu
Transfer learning-enhanced physics informed neural network for accurate melt pool prediction in laser melting
Advanced Manufacturing
selective laser melting
melt pool
deep learning
pinn
transfer learning
title Transfer learning-enhanced physics informed neural network for accurate melt pool prediction in laser melting
title_full Transfer learning-enhanced physics informed neural network for accurate melt pool prediction in laser melting
title_fullStr Transfer learning-enhanced physics informed neural network for accurate melt pool prediction in laser melting
title_full_unstemmed Transfer learning-enhanced physics informed neural network for accurate melt pool prediction in laser melting
title_short Transfer learning-enhanced physics informed neural network for accurate melt pool prediction in laser melting
title_sort transfer learning enhanced physics informed neural network for accurate melt pool prediction in laser melting
topic selective laser melting
melt pool
deep learning
pinn
transfer learning
url https://elsp-homepage.oss-cn-hongkong.aliyuncs.compaper/journal/open/AM/2025/am20250001.pdf
work_keys_str_mv AT qingyunzhu transferlearningenhancedphysicsinformedneuralnetworkforaccuratemeltpoolpredictioninlasermelting
AT zhengxinlu transferlearningenhancedphysicsinformedneuralnetworkforaccuratemeltpoolpredictioninlasermelting
AT yaowuhu transferlearningenhancedphysicsinformedneuralnetworkforaccuratemeltpoolpredictioninlasermelting