Thesis defense by Manon Jubert

Thesis title:

"Algorithm for planning the digitization and alignment of 3D point clouds for in-situ inspection of mechanical parts during machining"

Summary:

In this thesis, we address the problem of digitization planning and also tackle the issue of 3D point cloud alignment. These problems can be studied and generalized to meet the needs of many fields of study.

We are addressing the following issue: how can a mechanical part be automatically inspected during machining? Or more precisely: how can an accurate representation of the part be obtained so that its conformity can be verified during the machining process?
This question gives rise to two main sub-problems that we are seeking to address. First, we need to define a way to digitize the part in its environment. Second, we need to reconstruct the model of the part so that geometric measurements can be made.

The first contribution to this study is the development of an algorithm for automatically planning the digitization of a part based on the checks to be performed on it. The approach we propose can be adapted to any optical digitization tool (lidar, camera, profilometer, fringe projection sensor, etc.). In our study, we show the results of this method on several fringe projection optical sensors and on several industrial parts. The proposed method is general and can also be adapted to any industrial environment in which the part to be inspected is located.

The second contribution focuses on the specific case of data from optical sensors, namely 3D point clouds. Based on the digitization plan, we develop a strategy for aligning the point clouds with each other. We address the issue of cases where point cloud alignment is not possible and attempt to resolve this problem.

The final contribution of this work is the integration of these methods into a product currently being marketed. We identify the key points that make the solution robust for specific industrial cases.

Finally, we show several industrial applications of these algorithms and discuss the accuracy of the proposed methods in these cases. We propose several avenues for further work, particularly in specific cases involving non-alignable or large-scale parts. Ideas for improving and strengthening the algorithms are put forward for further research.

Thesis jury:

Claire LARTIGUE, Professor, École Normale Supérieure de Paris-Saclay, LURPA

Stéfanie HAHMANN, Professor, Grenoble National Polytechnic Institute, LJK

Géraldine MORIN, Professor, National Polytechnic Institute of Toulouse, IRIT

Raphaëlle CHAINE, Professor, University of Lyon 1, LIRIS

Michel DEMESY, Valduc Atomic Energy Commission

Jean-Luc MARI, Professor, Aix Marseille University, LIS

Jean-Philippe PERNOT, Professor, Ecole Nationale SupérieureArts et Métiers, LISPEN

Arnaud POLETTE, Associate Professor, École Nationale SupérieureArts et Métiers, LISPEN}

Dominique NOZAIS, I-MC

Keywords:

CAD model, Point cloud, View planning algorithm, Optical sensor, 3D reconstruction, Part inspection, Alignment algorithm