Journal
APPLIED ACOUSTICS
Volume 180, Issue -, Pages -Publisher
ELSEVIER SCI LTD
DOI: 10.1016/j.apacoust.2021.108125
Keywords
Laser ultrasound; Grain size distribution; Neural network; Lognormal distribution; Metallic materials
Categories
Funding
- National Natural Science Foundation of China [51975043]
- Beijing Natural Science Foundation [3182026]
- Fundamental Research Funds for the Central Universities [FRF-TP-19-002A3]
- China Scholarship Council (CSC)
Ask authors/readers for more resources
This study proposes a grain size distribution characterization method based on machine learning using laser-induced ultrasonic technology. It avoids the problem of choosing different scattering attenuation mechanisms for grain size distribution and demonstrates the feasibility of characterizing grain size distribution.
A single average grain size has been used to characterize all grain sizes of polycrystalline materials. However, the grain size distribution also affects performance and quality. A grain size distribution characterization method was investigated based on a machine learning approach using a laser-induced ultrasonic technology. A pulsed laser was used to generate ultrasound inside the specimens and the ultrasonic signals were detected using a two-wave-mixing interferometer. The grain size distribution was quantified using the expectation and standard deviation of the logarithmic normal distribution function. The attenuation coefficients in different frequencies of ultrasonic signals were set as inputs and the expectation and standard deviation of grain size distribution were set as outputs. The grain size distribution prediction model was built with a neural network optimized by the particle swarm optimization algorithm. 90 data samples were selected as training data (75%) to train the characterization model and 30 data samples were set as test data (25%). The method does not require a physical model and avoids the problem of choosing different scattering attenuation mechanisms for grain size distribution. The results show the machine learning method has the feasibility to characterize the grain size distribution. (C) 2021 Elsevier Ltd. All rights reserved.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available