Real-time femtosecond laser cleaning of microcontaminants using deep learning
Real-time femtosecond laser cleaning of microcontaminants using deep learning
Laser cleaning is used extensively across industrial manufacturing for the removal of contaminants, including rust, paint, and surface impurities. However, conventional laser cleaning typically applies higher-than-necessary laser energy to ensure the complete contaminant removal, leading to excessive energy consumption, unintentional material removal, and potential substrate damage. A more precise and adaptive cleaning strategy can enhance energy efficiency while preserving substrate integrity.Recent advancements in deep learning have significantly improved process control in laser materials processing, particularly in real-time monitoring, predictive visualization, and parameter optimization. These developments offer a promising insight to addressing key challenges in laser cleaning by dynamically adjusting laser parameters to optimize contaminant removal while minimizing energy consumption and collateral damage.This work introduces a new femtosecond laser cleaning method enhanced by deep learning, where a neural network, integrated into a real-time imaging system, automatically detects and identifies particle contaminants. Each identified particle is then selectively removed using a single laser pulse with precisely calibrated pulse energy, ensuring effective and efficient cleaning without damaging the substrate. Cleaning trials were conducted on samples containing a mixture of 15 µm polystyrene microbeads and 20 µm polymethacrylate microbeads. The neural network, trained on high-resolution imaging data, can predict the optimal laser energy required for removing each type of microbead by analyzing the appearance of the microbeads in real time. This adaptive approach ensures effective removal while minimizing energy consumption and substrate damage. The proposed method is particularly suited for high-precision applications in aerospace, electronics, vehicle maintenance, and medical device manufacturing, where selective material removal is essential.
Liu, Yuchen
1efd4b12-3f11-4eb1-abea-0f5b40a1a9f1
Grant-Jacob, James A.
c5d144d8-3c43-4195-8e80-edd96bfda91b
Xie, Yunhui
c30c579e-365e-4b11-b50c-89f12a7ca807
Chernikov, Fedor
a5a56a14-d8cf-4a11-8946-dbb145dbda91
Zervas, Michalis
1840a474-dd50-4a55-ab74-6f086aa3f701
Mills, Ben
05f1886e-96ef-420f-b856-4115f4ab36d0
10 June 2025
Liu, Yuchen
1efd4b12-3f11-4eb1-abea-0f5b40a1a9f1
Grant-Jacob, James A.
c5d144d8-3c43-4195-8e80-edd96bfda91b
Xie, Yunhui
c30c579e-365e-4b11-b50c-89f12a7ca807
Chernikov, Fedor
a5a56a14-d8cf-4a11-8946-dbb145dbda91
Zervas, Michalis
1840a474-dd50-4a55-ab74-6f086aa3f701
Mills, Ben
05f1886e-96ef-420f-b856-4115f4ab36d0
Liu, Yuchen, Grant-Jacob, James A., Xie, Yunhui, Chernikov, Fedor, Zervas, Michalis and Mills, Ben
(2025)
Real-time femtosecond laser cleaning of microcontaminants using deep learning.
LAMP 2025: The 9th International Congress on Laser Advanced Materials Processing, Sinfonia Technology Hibiki Hall Ise, Ise-city, Japan.
10 - 13 Jun 2025.
1 pp
.
Record type:
Conference or Workshop Item
(Other)
Abstract
Laser cleaning is used extensively across industrial manufacturing for the removal of contaminants, including rust, paint, and surface impurities. However, conventional laser cleaning typically applies higher-than-necessary laser energy to ensure the complete contaminant removal, leading to excessive energy consumption, unintentional material removal, and potential substrate damage. A more precise and adaptive cleaning strategy can enhance energy efficiency while preserving substrate integrity.Recent advancements in deep learning have significantly improved process control in laser materials processing, particularly in real-time monitoring, predictive visualization, and parameter optimization. These developments offer a promising insight to addressing key challenges in laser cleaning by dynamically adjusting laser parameters to optimize contaminant removal while minimizing energy consumption and collateral damage.This work introduces a new femtosecond laser cleaning method enhanced by deep learning, where a neural network, integrated into a real-time imaging system, automatically detects and identifies particle contaminants. Each identified particle is then selectively removed using a single laser pulse with precisely calibrated pulse energy, ensuring effective and efficient cleaning without damaging the substrate. Cleaning trials were conducted on samples containing a mixture of 15 µm polystyrene microbeads and 20 µm polymethacrylate microbeads. The neural network, trained on high-resolution imaging data, can predict the optimal laser energy required for removing each type of microbead by analyzing the appearance of the microbeads in real time. This adaptive approach ensures effective removal while minimizing energy consumption and substrate damage. The proposed method is particularly suited for high-precision applications in aerospace, electronics, vehicle maintenance, and medical device manufacturing, where selective material removal is essential.
Text
Yuchen Liu_Abstract_LAMP2025
- Author's Original
More information
Submitted date: 7 February 2025
Accepted/In Press date: 11 March 2025
Published date: 10 June 2025
Venue - Dates:
LAMP 2025: The 9th International Congress on Laser Advanced Materials Processing, Sinfonia Technology Hibiki Hall Ise, Ise-city, Japan, 2025-06-10 - 2025-06-13
Identifiers
Local EPrints ID: 509112
URI: http://eprints.soton.ac.uk/id/eprint/509112
PURE UUID: 4c550e6a-e52f-4ddb-ac6a-a0ee2330b2e0
Catalogue record
Date deposited: 11 Feb 2026 17:49
Last modified: 12 Feb 2026 03:20
Export record
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics
