> Markdown version of [/jobs/ext/4755350-intership-learning-ultrasonic-acquisition-patterns-for-3d-interface-characterisation-saclay-h-f](https://www.wearedevelopers.com/jobs/ext/4755350-intership-learning-ultrasonic-acquisition-patterns-for-3d-interface-characterisation-saclay-h-f). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md). --- # Intership Learning ultrasonic acquisition patterns for 3D interface characterisation-Saclay-H/F - **Company:** CEA Industrie - **Location:** Saclay, France - **Contract:** Internship / Graduate position - **Skills:** Computing Platforms, Experimental Data, Hardware Design, Image Quality, Machine Learning, Scientific Computating, Signal Processing, Software Engineering, Reinforcement Learning, Pytorch - **Published:** November 17, 2026 - **Apply:** https://www.emploi.cea.fr/offre-de-emploi/emploi-intership-learning-ultrasonic-acquisition-patterns-for-3d-interface-characterisation-saclay-h-f_42585.aspx ## About the Role Final-year engineering student or master's student in machine learning, signal processing or applied mathematics; proficiency in Python/Pytorch; interest in ultrasonics, scientific computing. ## Description The Acoustics for Inspection and Characterization Laboratory (LA2C) develops ultrasonic inspection and characterization methods, as well as associated robotics and sensors. It has significant expertise in hardware and software development, as well as current material and industrial problems. Its current principal focus is on ultrasonic imaging for complex industrial scenarios. The Intelligent, Distributed and Embedded Instrumentation Laboratory (LIIDE) is dedicated to developing a hybrid hardware-software platform to design the instrumentation functionalities of the future. The laboratory works on two complementary fronts: 1. Hardware development, focused on versatile and modular electronic boards together with the necessary software for their operation, in order to cover a wide range of sensor technologies; and 2. Innovative artificial intelligence functionalities for distributed measurement and frugal, decentralized learning. These laboratories are embedded within a rich ecosystem centered on digital instrumentation for control, monitoring, and diagnostics. The department it belongs to leverages a broad spectrum of sensors (optical fibers, piezoelectric sensors, eddy-current probes, X-ray systems) as well as cutting-edge experimental platforms. Its main application areas are non-destructive evaluation (NDE) and structural health monitoring (SHM)., As an intern at the CEA, you will have the opportunity to work within a world-renowned research environment. Our teams consist of passionate and dedicated experts, providing an ideal setting for learning and collaboration. You will have access to state-of-the-art equipment and top-tier research resources to carry out your assignments. The work conducted may potentially lead to a scientific publication. Objectives and proposed work For parametric surfaces, the internship will leverage existing semi-analytical estimation of planar and cylindrical surface parameters [1], and use reinforcement learning to identify sparse acquisition patterns that minimise errors in the estimated parameters, surface geometry and refracted travel times. For general surfaces, the internship will investigate how geometric envelope methods [2] can be extended to sparse acquisition settings, where spatial derivatives of the travel-time field must be estimated on a graph of sparse array elements, notably through weighted local regression. Regular, random, deterministic and optimised patterns from the 2D sparse-array literature [3] will serve as initialisations, baselines and relative penalties during training. All strategies will be compared at equal acquisition cost on simulated and then experimental data, using errors in surface parameters, surface points and normals, together with the induced travel-time error in an internal region. Methodology and outlook The work will combine geometrical modelling, signal processing, optimisation and reinforcement learning. A first policy will learn a fixed acquisition pattern tailored to a given family of surfaces; extending it to sequential and adaptive selection will be a natural next step. Reconstructed TFM image quality may subsequently provide an additional criterion for refining the interface. In the longer term, surface estimation and refraction-point estimation could be integrated into a learnt amortised solver. References [1] E. Iakovleva, D. Roué, and M. Marmonier, "Time efficient autofocus TFM imaging for ultrasonic inspection of planar and cylindrical components," Ultrasonics, vol. 166, art. 108118, 2026, doi: 10.1016/j.ultras.2026.108118. [2] J. Camacho, J. F. Cruza, J. Brizuela, and C. Fritsch, "Automatic dynamic depth focusing for NDT," IEEE Trans. Ultrason., Ferroelectr., Freq. Control, vol. 61, no. 4, pp. 673-684, 2014. [3] A. Ramalli, E. Boni, E. Roux, H. Liebgott, and P. Tortoli, "Design, implementation, and medical applications of 2-D ultrasound sparse arrays," IEEE Trans. Ultrason., Ferroelectr., Freq. Control, vol. 69, no. 10, pp. 2739-2755, 2022, doi: 10.1109/TUFFC.2022.3162419.