研究目的
To overcome the challenges of in situ and real-time monitoring of laser processes by combining fast hard X-ray imaging with acoustic sensors and state-of-the-art machine learning.
研究成果
The innovative approach combining AE and ML is promising for in situ and real-time monitoring of laser welding processes, with classification accuracy ranging from 74 to 95%.
研究不足
The signals are sensitive to environmental noises, and correlating the acoustic emission signal with real events is challenging.
1:Experimental Design and Method Selection:
Laser welding experiments were conducted using a single-mode fiber laser source with high-speed X-ray radiography combined with AE measurements. Four categories of laser welding processes were defined based on X-ray imaging.
2:Sample Selection and Data Sources:
An aluminum-magnesium alloy was chosen for its relative low Z-number. Samples had dimensions 50 x 20 x 2 mm
3:List of Experimental Equipment and Materials:
StarFiber 150 P fiber laser, piezo acoustic sensor Pico, data acquisition unit and software from Vallen, high-speed X-ray imaging at the beamline ID19 of the ESRF.
4:Experimental Procedures and Operational Workflow:
The acoustic signals were recorded and synchronized with the video sequences from X-ray imaging. The signals were preprocessed using wavelet packet transform and analyzed using gradient boost with two learners: classifications and regression trees CART, and trees based on independent component analysis (ICA).
5:Data Analysis Methods:
The classification of the AE signals into the four predefined categories was performed using gradient boost with both ICA and CART.
独家科研数据包,助您复现前沿成果,加速创新突破
获取完整内容