Digital Image Correlation (DIC) is a well-known non-contact experimental technique. Its most common implementation is the subset-based approach, which performs a local least-squares fitting of a simple displacement model — usually an affine transform — to identify the displacement components at the center of the small area under investigation (the subset). Because of the statistical approach, DIC is usually able to provide reliable results even when theoretical prerequisites are not fully satisfied or in the presence of noise. However, the least-squares algorithm is not robust when the data set contains multiple statistical distributions. Indeed, the algorithm does not discriminate between them and process all the input data; thus, the resulting solution is usually unsatisfactory. A typical example implying the described context is the presence of cracks or shear bands inside the subset: as different sections of the inspected area move in different directions, the algorithm is unable to select a solution and the correlation between the reference and the test image is usually poor. This work proposes using RANSAC, a well-known robust algorithm, to select the largest domain of the subset. Because a similar problem has to be faced when computing strain components by the polynomial-fitting method, a simple modification of the main algorithm is suggested to handle also this problem.
Robust Algorithms for Digital Image Correlation in the Presence of Displacement Discontinuities
Baldi A.
2020-01-01
Abstract
Digital Image Correlation (DIC) is a well-known non-contact experimental technique. Its most common implementation is the subset-based approach, which performs a local least-squares fitting of a simple displacement model — usually an affine transform — to identify the displacement components at the center of the small area under investigation (the subset). Because of the statistical approach, DIC is usually able to provide reliable results even when theoretical prerequisites are not fully satisfied or in the presence of noise. However, the least-squares algorithm is not robust when the data set contains multiple statistical distributions. Indeed, the algorithm does not discriminate between them and process all the input data; thus, the resulting solution is usually unsatisfactory. A typical example implying the described context is the presence of cracks or shear bands inside the subset: as different sections of the inspected area move in different directions, the algorithm is unable to select a solution and the correlation between the reference and the test image is usually poor. This work proposes using RANSAC, a well-known robust algorithm, to select the largest domain of the subset. Because a similar problem has to be faced when computing strain components by the polynomial-fitting method, a simple modification of the main algorithm is suggested to handle also this problem.File | Dimensione | Formato | |
---|---|---|---|
1-s2.0-S0143816619317841-main.pdf
Solo gestori archivio
Descrizione: Versione editoriale dell'articolo
Tipologia:
versione editoriale (VoR)
Dimensione
4.99 MB
Formato
Adobe PDF
|
4.99 MB | Adobe PDF | Visualizza/Apri Richiedi una copia |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.