PhD Position : « Multi-architecture programming for high performance reconstruction in passive tomography »

Multi-architecture programming for high performance reconstruction in passive tomography

Date limite de candidature : 01/07/2021
Date de début : 01/10/2021
Date de fin : 30/09/2024

Pôle : Signaux et statistiques
Type de poste : Thèses
Contact : GAC Nicolas (

la fiche

Multi-architecture programming for high performance reconstruction in passive tomography

Universite Paris Saclay PhD thesis (CEA funding)
autumn 2021/autumn 2024

Keywords: parallel computing, GPU, FPGA, HLS, tomography, SHM (Structural Health Monitoring)

CEA-List (Saclay)
Departement Imagerie Simulation pour le Contrôle (DISC)
Laboratoire de Developpement Informatique (LDI)
CEA supervisor:

Laboratoire des Signaux et Systemes (L2S)
Inverse Problem Group (GPI)
L2S supervisor: de these)

Structural Health Monitoring (SHM) is an approach to non-destructive testing that aims to integrate inspection tools directly into the targeted structures to facilitate data acquisition and avoid the regular mobilization of inspection resources (human, material) the immobilization of equipment. As part of this approach, passive ultrasonic tomography exploits the structural noise of waveguide-like parts to monitor variations in their thickness, to detect the appearance of corrosion or erosion defects. This process involves using several signal processing algorithms applied to large amounts of data and integrated into an embedded system.

PhD Goal. Intending to integrate SHM controls in compact and energy-ecient equipment, this PhD thesis aims at developing an embedded signal processing chain that meets the needs of passive
tomography. Therefore, it will be necessary to determine the most suitable hardware architectures and realize highly-optimized implementations of the algorithms involved in the processing chain by making them evolve according to the performance needs. To this end, di erent hardware architectures (GPU, low power GPU, FPGA) will be studied during the thesis by comparing generic programming approaches (Sycl) and implementations dedicated to each architecture.

A collaborative work. This PhD will take place at CEA-List (employer) and L2S (PhD direction). The Groupe Problemes Inverses (GPI) at L2S has mixed expertise in architecture and signal processing with long-term experience in tomography reconstruction. One research axis of this team is Algorithm Architecture co-design (Adequation Algorithme Architecture). The AAA methodology aims to find the best match between the algorithms and the targeted architecture to respect constraints in computing time, energy consumption, cost and result accuracy. The GPI has applied it mainly for tomography reconstruction on GPU [1] or FPGA boards through High Level Synthesis tools [2], receiving for this work with a student team an Intel price in the HPC category [3]. The Departement d’Imagerie et de Simulation pour le Contrôle (DISC) is part of the French nuclear energy commission (CEA) and focuses on various research topics linked to nondestructive testing, including control techniques and physical models for simulations with a focus on intensive computation on both CPUs and GPUs. This expertise led to the development of the CIVA software platform [4] that provides multiple non destructive testing simulation and analysis tools. It is distributed worldwide and serves as a de facto standard in the nondestructive testing industry. Some of its software components are used on industrial embedded nondestructive techniques solutions that enable real-time treatments for ultrasonic imaging. During the past few years, the DISC has also been developing a novel technique for structural health monitoring that produces tomography images from structural noise [5]. This work will be part of the collaborative process carried out by the SACHEMS platform led by the CEA and funded by the Ile de France Region [6].

PhD Candidate Profi le. Such an interdisciplinary project in parallel computing applied to passive tomography will require strong skills in computer science (e.g. CUDA/OpenCL) and a good level in signal processing.

[1] C. Chapdelaine et al., \TomoBayes – logiciel de reconstruction en tomographie CT, » Oct. 2019.
[2] M. Martelli et al., \3d tomography back-projection parallelization on intel fpgas using opencl, »
Journal of Signal Processing Systems, 2018.
[3] M. Martelli, \An OpenCL design for tomographic reconstruction on FPGA. » Innovate Europe
Design Contest Winners, 2017.
[4] \Civa, »
[5] T. Druet et al., \Passive guided wave tomography for structural health monitoring, » The Journal of
the Acoustical Society of America, vol. 146, no. 4, pp. 2395{2403, 2019.
[6] \Sachems, »