ORGANISATION/COMPANYUniversité Clermont Auvergne
RESEARCH FIELDComputer scienceMathematics
RESEARCHER PROFILEFirst Stage Researcher (R1)
APPLICATION DEADLINE28/06/2020 00:00 - Europe/Brussels
LOCATIONFrance › AUBIERE
TYPE OF CONTRACTTemporary
HOURS PER WEEK35 H
OFFER STARTING DATE01/10/2020
IS THE JOB RELATED TO STAFF POSITION WITHIN A RESEARCH INFRASTRUCTURE?Yes
Development of a convolutional neural network for the real-time measurement of
displacement and strain fields on the surface of deformed structures
Supervisor: Michel GREDIAC
Laboratory: Institut Pascal, UMR CNRS 6605
Email and phone: email@example.com, 0649232265
Co-advisor(s): François Berry, Benoît Blaysat
Abstract (up to 10 lines):
Digital Image Correlation (DIC) has progressively become the reference technique for
measuring displacement and strain fields on specimens subjected to various
thermomechanical loadings. A range of different industries such as robotics, civil
engineering, automotive industry or aeronautics are concerned. DIC has however some
serious limitations such as calculating time or metrological performance. This PhD work
will mainly consist of developing a convolution neural network dedicated to this type of
measurement, but with better global performance than DIC, as a recent feasibility study
appears to be suggesting. The final goal will be to implement this network on an
embedded system able to provide with high accuracy real-time displacement and strain
maps on specimens subjected to weak deformation.
- basic education in signal processing, image processing and/or mechatronics
- applicants showing a strong appetence for computational mechanics or for the
development of full-field measurement techniques will also be considered.
Convolutional Neural Network, Full-field measurements, Metrology, Photomechanics,
Description (up to 1 page):
This project deals with the use of images in order to measure displacement and strain fields
on the surface of flat objects subjected to thermomechanical loads. Various fields dealing
with mechanics of materials and structures are concerned such as aeronautics or civil
engineering. Various techniques have been developed to perform such measurements, in
particular Digital Image Correlation , have been developed to reach this goal. They
feature however some limitations, especially in terms of processing time and metrological
performance, the displacements and strains to be measured generally featuring a low
A preliminary feasibility study has been recently carried out. It has led to the definition of a
convolutional neural network and a daset suitable for training this network. This dataset
comprises speckle images, which are deformed through well-chosen sub-pixel displacement
fields. This network trained on this dataset features a global performance, which is
equivalent to, if not better than that of DIC . This gives us a glimpse of real-time
displacement and strain field measurement with better metrological performance that
In this context, this work consists of different tasks:
- The first goal will be to better understand the interaction between performance of
the network and definition of the synthetic images of the dataset in terms of nature
of the speckle, displacement fields used to generate the images, and computing
time. The assumptions under which these images are defined and their impact on
the quality of the final results will also be investigated. The speckle rendering system
developed by the supervising team  may be used for this purpose.
- The network proposed in  was developed on the basis of other networks
dedicated to other applications, in particular concerning displacements, which
amplitude was much greater than one pixel, while the challenge here is to properly
measure displacements with an uncertainty lower than one hundredth of pixel. It
will therefore be necessary to redesign completely the network, under the
constraint to limit as much as possible the number of layers, and thus the number
of coefficients to be determined during the training phase.
- The network will be embedded on a small computer (typicall a NVIDIA Jetson), to
have at the end a compact system providing real-time displacement and strain maps.
- Depending on the work progress, various cameras will be associated in order to get
a stereo system providing tridimensional displacement fields.
Various real experiments will be carried out in order to evidence the benefit of using this
new approach for measuring displacement and strain fields in mechanics of materials and
structures, with substantial industrial applications.
Supervision will be carried out by a pluridisciplinary team composed of experts in embedded
systems, signal processing (F. Berry) and photomechanics (B. Blaysat, M. Grédiac).
Références (up to ½ page):
 M. Sutton, J.J. Orteu, and H. Schreier. Image Correlation for Shape, Motion and
Deformation Measurements. Basic Concepts, Theory and Applications. Springer, 2009
 S. Boukhtache, K. Abdelouahab, F. Berry, B. Blaysat, F. Sur, M. Grédiac. When Deep
Learning meets Digital Image Correlation. Submitted, 2020.
 F. Sur, B. Blaysat, M. Grédiac. Rendering deformed speckle images with a Boolean model.
Journal of Mathematical Imaging and Vision, 2018.
How to candidate?
Contact the supervisor
EURAXESS offer ID: 528432
The responsibility for the jobs published on this website, including the job description, lies entirely with the publishing institutions. The application is handled uniquely by the employer, who is also fully responsible for the recruitment and selection processes.
Please contact firstname.lastname@example.org if you wish to download all jobs in XML.