EPJ Web of Conferences 6, 31006 (2010)
https://doi.org/10.1051/epjconf/20100631006

Digital image correlation: displacement accuracy estimation

¹ Institut Pprime, UPR 3346 CNRS, Université de Poitiers, ENSMA, D3 PEM, SP2MI, Téléport 2, Bd M. et P. Curie, BP 30179, F86962 Futuroscope Chasseneuil CEDEX, France
² Laboratoire Navier, École des Ponts ParisTech, Université Paris-Est, 77544 Marne-la-Vallée, France
³ Université de Toulouse; INSA, UPS, Mines Albi, ISAE; ICA (Institut Clément Ader); Campus Jarlard, F-81013 Albi, France
⁴ Ecole Mines Albi, Campus Jarlard, F-81013 Albi, France
⁵ Laboratoire de Mécanique et Génie Civil, UMR CNRS 5508, Université Montpellier 2, Place E. Bataillon, 34095 Montpellier, France

^a corresponding author, e-mail : jean-christophe.dupre@lms.univ-poitiers.fr

Abstract

The aim of this collaborative work is to study the uncertainties associated with Digital Image Correlation techniques (DIC). More specifically, the link between displacement uncertainties and several correlation parameters chosen by the user and relative to the image analysis software and several image characteristics like speckle size and image noise is emphasized. A previous work [1] has been done for situations with spatially fluctuating displacement fields which dealt with mismatch error linked to the discrepancy between the adopted shape function and the real displacement field in the subset. This present work is focused on the ultimate error regime. To ensure that there is no mismatch error, synthetic images of plane rigid body translation have been analysed. DIC softwares developed by or used in the French community were used to study a large number of settings. The first observations are: (a) bias amplitude is almost always insensitive to the subset size, (b) DIC formulations can be split up into two families. For the first one, the bias amplitude increases with the noise while it remains constant for the second one. For both families, the mean value of the random error increases with the noise level and with the inverse of the subset size. Furthermore, the random error decreases with the radius of the speckle for the first family, while it increases for the second one. These two different behaviours of the tested DIC package are probably due to their underlying DIC formulation (interpolation, correlation criteria, optimisation process).