Scientific direction Development of key enabling technologies
Transfer of knowledge to industry

PhD : selection by topics

Technological challenges >> Factory of the future incl. robotics and non destructive testing
6 proposition(s).

See all positions [+]

Situation awareness and task planning for a mobile manipulator in uncertain logistic environment

Département Intelligence Ambiante et Systèmes Interactifs (LIST)

Laboratoire de Contrôle et Supervision Robotique

01-03-2021

SL-DRT-21-0459

eric.lucet@cea.fr

Factory of the future incl. robotics and non destructive testing (.pdf)

The proposed research project concerns the autonomous evolution of a mobile robot in a logistic context, for example an automated forklift truck. Based on an analysis of the situation, the system will have to be able to autonomously find the sequence of actions allowing it to move towards an object of interest, grasp it and place it in a predefined region. This must be done while avoiding collisions and possibly clearing the path to make possible the movement of the mobile base and gripper. In case of failure, the replanning inherent in the method must be able to find a new sequence of actions. It will thus be necessary to implement a statistical model of the current and future navigation situation in the proximity of a mobile robot equipped with a gripper system, as well as a task planning algorithm based on this model. The Situation Awareness model will be based on contextual data from the perception modules, the process and task models, the agents (robots and humans) present in the environment and their states, the intrinsic data of the robot and the geometric model of the environment. The detection of particular situations (anomalies, etc.) can be handled by data analysis and automatic learning algorithms, possibly with a training phase based on feedback and a priori knowledge. In particular, the Hierarchical Planning in the Now (HPN) approach integrates task and motion planning and deals with uncertainty. It avoids trying to find optimal solutions for the POMDP (which is insoluble), by constructing a deterministic approximation of the dynamics (model of the situation), establishing a sequential plan, and executing this plan while observing possible changes in relation to the expected results, and re-planning it when deviations occur. In addition, to deal with uncertainty about the current state, planning must be done within the belief space, which is the probability of distributions over the states of the world. Thus, based on preliminary work on this subject, an improvement of the hierarchical plan, as well as a better understanding and formulation of models of the modification of belief states resulting from these actions, will be investigated.

Download the offer (.zip)

Shell finite elements for real time simulations in virtual reality

Département Intelligence Ambiante et Systèmes Interactifs (LIST)

Laboratoire de Simulation Interactive

01-09-2021

SL-DRT-21-0594

anders.thorin@cea.fr

Factory of the future incl. robotics and non destructive testing (.pdf)

With growing interest in interactive simulation comes the need to simulate shells (i.e. volumes with a thickness smaller than the two other dimensions) in real time. The objective of this thesis is to simulate the dynamics of thin shells in small strains, in the context of large displacements (geometric non-linearity) in real time. The work will be carried out in three parts, i) provide criteria to determine what can be simulated in real time or not (with a given number of degrees of freedom and a given precision), ii) identify or develop finite elements that are both robust to handle different use cases and efficient in terms of computational cost, iii) provide an implementation and examples of simulations that cannot be simulated in real time to date, with equal hardware.

Download the offer (.zip)

Optimization of acquisition parameters in measurement and X-ray imaging through simulation

Département Imagerie Simulation pour le Contrôle (LIST)

Laboratoire Simulation et Modélisation en Electro-magnétisme

01-10-2021

SL-DRT-21-0723

anthony.touron@cea.fr

Factory of the future incl. robotics and non destructive testing (.pdf)

The objective of the thesis is to set up a numerical method for optimizing acquisition parameters in X-ray control (XR). The DISC is developing a simulation software for Non Destructive Testing which integrates a module for X-ray imaging. This module, which corresponds to a digital twin of the testing, allows the operator to reproduce an experimental configuration (source, detector, geometry and material of the object, defects, ...) in order to estimate the result he would observe on a real acquisition and thus evaluate the sensitivity of the method, the limit of detectability of defects or the impact of different parameters (tension and filtration of the tube, acquisition geometry). To define these configurations, the operator has to rely on his own experience, which results in the realization of potentially non-optimal acquisitions. This limitation related to user dependency is particularly critical in cases where the signal sought is very weak, for example in the identification of a small quantity of element in X-ray fluorescence or for X-ray tomography which involves a large number of acquisition parameters. The objective is to make the experimental measurements more robust, by adding intelligence to the existing simulation tool, which would then integrate a functionality to optimize the acquisition parameters. The PhD student will have to propose a numerical method allowing to define, from a set of simulations, the parameters maximizing a given quality criterion. This criterion will be defined as a function of the targeted testing (typically the amplitude of the fluorescence peak or the transmission rate of the X-ray signal) and validated experimentally.

Download the offer (.zip)

Simulation-assisted imaging for SHM by elastic guided waves: tomography and shape derivative

Département Imagerie Simulation pour le Contrôle (LIST)

Laboratoire Méthodes CND

01-10-2021

SL-DRT-21-0738

tom.druet@cea.fr

Factory of the future incl. robotics and non destructive testing (.pdf)

Structural health monitoring (SHM) consists of monitoring the health of a structure using integrated sensors. In this context, the LIST is working on innovative elastic guided wave tomography methods to image corroded areas on simple geometries: plates and pipes. The subject of the thesis is to adapt to our dynamic problems the so-called "shape derivative" method, mostly used on static data, to develop iterative imaging based on a digital twin. This method allows a high quality reconstruction for complex geometries but implies the resolution of an expensive numerical problem. The quality of the image is also highly dependent of the initialization, often based on the perfect structure, potentially far from the current state. We propose to address these two points as follows: the initialization of the shape derivative will be obtained from existing tomography methods, thus improving convergence by limiting the number of iterations and the effect of local minima. The numerical problem will be solved using transient high-order spectral finite elements, allowing fast and inexpensive computations. The definition of metrics suitable for the comparison of simulated and experimental signals will also be part of the study. The performances will be validated on experimental data representative of complex cases of interest for the industry (corrosion under support, etc.).

Download the offer (.zip)

Deep learning based approach for sparse view Computed Tomography

Département Imagerie Simulation pour le Contrôle (LIST)

Laboratoire Simulation et Modélisation en Electro-magnétisme

01-10-2021

SL-DRT-21-0739

caroline.vienne@cea.fr

Factory of the future incl. robotics and non destructive testing (.pdf)

X-ray Computed Tomography (CT) is a well-established contactless and non-destructive inspection technique for various industrial parts. With its roots in the medical field, it evolved to industry as a key tool for three dimensional (3D) characterization and inspection. Industrial progress brings the need for verification of various objects on the production line and in some cases only the X-ray CT can fit the requirements. However, CT scans take a long time, often up to several hours, to achieve reasonable data quality. This long scanning time is an obstacle preventing integration of CT into production lines. In-line tomography implies strong constraints on the acquisition geometry, such as limited angle and sparse views configuration. Due to missing data, reconstructed images suffer from important artifacts that make the inspection task difficult. Many approaches have been investigated to reduce such artifacts. Among them, those based on deep learning seem promising. Convolutive neural network such as U-Net and generative adversarial networks will be studied for recovering missing projection data. The considered framework consists in creating a first version of the sinogram by projecting the initial reconstruction obtained by a FDK algorithm applied to the original incomplete projections and using the complete sinogram as target. Following the previous training of the neural network on simulated data, a second step will consist in using a transfer learning approach and experimental data sets to train a model able to deal with real data.

Download the offer (.zip)

Multi-architecture programming for high performance passive tomography reconstruction

Département Imagerie Simulation pour le Contrôle (LIST)

Laboratoire Développement Informatique

01-03-2021

SL-DRT-21-0745

hamza.chouh@cea.fr

Factory of the future incl. robotics and non destructive testing (.pdf)

Structural Health Monitoring is a set of non-destructive testing methods that aim to be directly integrated within the structures to control. This allows for continuous structural testing without disabling any equipment or involving additional human resources or testing hardware. Passive ultrasonic tomography uses structural noise of specimens to monitor their thickness. This method is dedicated to specimens that can be described as wave-guides. It is particularly useful to detect corrosion and erosion defects. The testing process implies dealing with large amount of numerical signals and applying various algorithms to them. In order to embed SHM controls into lightweight and low energy equipment, this thesis aims the development of a full passive ultrasonic tomography toolchain. This will require the evaluation of several hardware architectures (GPU, Low power GPU, FPGA) to find the best-suited ones. To this end, the student will produce high performance implementations of the tomography algorithms on the selected architectures and compare them with the performance obtained via a generic approach like Sycl. The development process should be flexible enough to provide maintainable and scalable software in order to ease the integration of future evolutions of the passive ultrasonic tomography method.

Download the offer (.zip)

See all positions