Abstract The characterization of fractographic surfaces typically requires experts to evaluate the characteristics of fracture surfaces. However, these evaluations are influenced by human factors, such as subjectivity, and suffer from a lack of reproducibility. In this context, machine learning (ML), which has been established in various disciplines within materials science over the past few years, is a promising field enabling a more objective and reproducible evaluation. This study will evaluate the use of ML for the evaluation of fracture surfaces of notched Charpy specimens based on digital camera images. Image sections of the two reference regions “upper shelf” (ductile) and “lower shelf” (brittle) will serve as the database. In a first step, data visualization will be performed and data separability will be verified using unsupervised ML. On this basis, supervised ML will be used to train models to distinguish brittle and ductile fractures. These models will then be applied to determine ductile und brittle portions in mixed fracture modes, with the results being in good agreement with the expert consensus achieved in the round robin test.
[1]
Adrien Gruson,et al.
Deep Learning for fractographic classification in metallic materials
,
2020
.
[2]
Thomas G. Dietterich.
What is machine learning?
,
2015,
Archives of Disease in Childhood.
[3]
G. Pantazopoulos.
A Short Review on Fracture Mechanisms of Mechanical Components Operated under Industrial Process Conditions: Fractographic Analysis and Selected Prevention Strategies
,
2019,
Metals.
[4]
Elizabeth A. Holm,et al.
Microstructure Cluster Analysis with Transfer Learning and Unsupervised Learning
,
2018,
Integrating Materials and Manufacturing Innovation.
[5]
Juan Carlos Fernández,et al.
Multiobjective evolutionary algorithms to identify highly autocorrelated areas: the case of spatial distribution in financially compromised farms
,
2014,
Ann. Oper. Res..
[6]
H. Abdi,et al.
Principal component analysis
,
2010
.