머신러닝 및 XAI 기반 최적화를 통한 L-PBF 박판 3.3% Si 전기강판의 제조 공정 및 특성 분석

In this study, electrical steel was additively manufactured using the Laser Powder Bed Fusion (LPBF) process with 3.3% Si powder. The effects of varying laser power, scanning speed, and Y-axis scanning angle on the microstructure, density, surface roughness, and hardness were investigated. Machine l...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:대한금속·재료학회지, 63(8) Ročník 63; číslo 8; s. 568 - 582
Hlavní autoři: 김현아, Hyeonah Kim, 박세민, Semin Park
Médium: Journal Article
Jazyk:korejština
Vydáno: 대한금속재료학회 05.08.2025
대한금속·재료학회
Témata:
Machine learning
Magnetic materials
Powder processing
Scanning electron microscopy(SEM)
Si steels
재료공학
ISSN:1738-8228, 2288-8241
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:In this study, electrical steel was additively manufactured using the Laser Powder Bed Fusion (LPBF) process with 3.3% Si powder. The effects of varying laser power, scanning speed, and Y-axis scanning angle on the microstructure, density, surface roughness, and hardness were investigated. Machine learning and explainable artificial intelligence (XAI) techniques were employed to model and optimize processing conditions to achieve desirable density, surface roughness, and hardness, and the results were compared with those of commercial electrical steel. Microstructural observations revealed melt pools with cellular structures, dendrites, and porosity along the solidification heat flow direction. As the number of layers increased, primary pores formed and extended continuously along the build direction (BD). Stable values of porosity, surface roughness, and hardness were achieved when the laser power exceeded 250W and the scan speed ranged from 300 to 1000 mm/s. Increasing the Y-axis scanning angle led to a stepped layer structure due to variations in deposition thickness, which negatively affected surface roughness and porosity, although the impact on hardness was minimal―indicating the feasibility of thin-layer fabrication. Optimization using machine learning and XAI reduced porosity to 5.49%, surface roughness to 4.6 μm, and increased hardness to 305.9 HV. The predicted and experimental values showed deviations within 5%, validating the effectiveness of the machine learning approach. By prioritizing high density and low roughness, the optimal processing conditions were suggested to be 260W, 30o, and 700 mm/s. Compared with commercial electrical steel, the additively manufactured samples showed a 2% lower density, fourfold higher roughness, and 10% higher hardness, suggesting that with improved roughness control, L-PBF-based additive manufacturing could become a viable production route for electrical steel. (Received 15 May, 2025; Accepted 19 June, 2025)
Bibliografie:The Korean Institute of Metals and Materials
ISSN:1738-8228
2288-8241
DOI:10.3365/KJMM.2025.63.8.568