Sustainability analysis and Kolmogorov–Arnold Network (KAN) machine learning algorithm application for dry, EMQL, LCO 2 and HNEMQL assisted drilling of GLARE fiber metal laminates

Annals of Forest Science · Springer Nature · 2026-10-26 · DOI: 10.1007/s10845-026-03010-2

Abstract

This study investigates the hole quality in GLARE fibre metal laminates utilizing three cooling/lubrication techniques: dry, Electrostatic Minimum Quantity Lubrication (EMQL), cryogenic liquid carbon dioxide (LCO 2 ) and Hybrid Nanofluid Electrostatic MQL (HNEMQL). Under these cutting environments, the sustainability and machinability aspects of machining are comprehensively compared in terms of techno-economic and carbon emission analysis. The impact of the feed rate and spindle speed on cutting forces, power consumption, and surface finish was also evaluated. Additionally, a novel machine learning technique of Kolmogorov-Arnold Network (KAN) and MultiLayer Perceptron (MLP) was used for prediction of several output parameters utilizing experimental data. It was found that the cost while machining under EMQL, LCO 2 and HNEMQL conditions increased by up to 21.41%, 19.77%, 31.51% respectively in comparison with dry machining. The carbon emissions under EMQL, LCO 2 and HNEMQL conditions increased by up to 46.5%, −1.7%, 55.8%, respectively in comparison with dry condition. Thrust force and torque were highest under LCO 2 machining, increasing by up to 69.8% and 66.3%, respectively, compared with dry drilling. Surface roughness was minimal using HNEMQL (~ 0.7 µm), representing an improvement of up to 33.3% relative to dry drilling, while LCO 2 achieved the lowest cutting power consumption, reducing energy usage by up to 34.7%. Overall, HNEMQL emerged as the most effective cooling/lubrication strategy for drilling GLARE laminates owing to its superior surface finish, lower drilling loads, and improved energy efficiency compared with conventional dry and EMQL machining. The KAN and MLP models were compared based on their predictions, and it was found that KAN got R 2 -Value of 98.71% whereas MLP predicted the outputs only with R 2 -value of 60.64%. Hence, KAN outperformed Traditional MLP for finding complex patterns among features and targets and predicting thrust force, power consumption, and surface roughness simultaneously.

Read at publisher →

Similar articles