An extreme learning machine for predicting kerf waviness and heat affected zone in pulsed laser cutting of thin non-oriented silicon steel. (November 2020)

• Record Type: Journal Article
• Authors: Nguyen, Tan Hoai; Lin, Chih-Kuang; Tung, Pi-Cheng; Nguyen-Van, Cuong; Ho, Jeng-Rong
• Is Part Of: Optics and lasers in engineering. Volume 134(2020)
• Publication Date: 2020-11
• Subjects: Heat affected zone (HAZ) -- Kerf waviness -- Pulsed laser cutting -- Extreme learning machine (ELM) -- Thin non-oriented silicon steel sheet -- Random forests method; Lasers in engineering -- Periodicals; Optical measurements -- Periodicals; Optics -- Periodicals; Lasers en ingénierie -- Périodiques; Mesures optiques -- Périodiques; Optique -- Périodiques; 621.36605
• DOI: 10.1016/j.optlaseng.2020.106244
• Languages: English
• ISSNs: 0143-8166

Abstract

Highlights: Extreme learning machine is practical and effective for predicting laser cutting properties. Laser power is most significant factor on cutting properties. Pulsed laser is powerful for machining thin silicon steels. Abstract: Due to lower core loss and higher flux density and permeability, thin non-oriented silicon steels are becoming more and more important for soft magnetic materials. Recently, laser has been emerged as a cost-effective tool for machining thin silicon steels, especially for the low-volume and high-variety motor manufacturing. Based on experimental data, this study aims at developing an extreme learning machine (ELM) for predicting the laser cutting qualities of silicon steels with a thickness of 100 μm. The three parameters considered were the laser power, cutting speed and pulse repetition rate and the two qualities monitored were the kerf waviness and heat affected zone (HAZ). Each parameter was designated at four levels and totally 64 sets of experimental parameter were performed. Experimental results showed that both cutting qualities were positively correlated with these three parameters. We randomly took 80% of the experimental data for model training while the remaining 20% was for model testing. To verify the ELM's appropriateness and advantage, two auxiliary models, artificial neural network and full quadratic multiple regression analysis (MRA), were also developed based on the same dataset for comparison. Results revealed that ELM well predicted waviness and HAZ and provided the most accurate predictions among the three models. The errors for waviness and HAZ were 2.90% and 4.16%, respectively. Consequently, the developed ELM was practical and effective for the waviness and HAZ estimations. Moreover, based on the random forests method, the relative significance of inputs associated with the responses was also addressed.