Automatic inspection of aeronautical mechanical assemblies using 2D and 3D computer vision - Archive ouverte HAL Access content directly
Conference Papers Year : 2019

Automatic inspection of aeronautical mechanical assemblies using 2D and 3D computer vision

(1, 2) , (2) , (1) , (2) , (2)
Hamdi Ben Abdallah
Igor Jovančević
Jean-José Orteu
Benoit Dolives
  • Function : Author
Ludovic Brèthes


Quality control is of key importance in the aerospace industry. This paper deals with the automatic inspection of mechanical aeronautical assemblies. For that purpose, we have developed a computer-vision-based system made of a robot equipped with two 2D cameras and a 3D scanner. The 3D CAD model of the mechanical assembly is available. It is used as a reference and it describes the assembly as it should be. The objective is to verify that the mechanical assembly conforms with the CAD model. Several types of inspection are required. For instance, we must check that the needed elements of the assembly are present, and that they have been mounted in the correct position. For this kind of inspection we use the 2D cameras and we have developed inspection solutions based on 2D image analysis. We have found that some types of inspection cannot be performed by using only 2D image analysis. A typical example of such types is detecting the interference between elements. It requires to check if two flexible elements (e.g. cables,harnesses) or a flexible and a rigid element (e.g. pipe, support) are at a safe distance from each other. For this type of situations, we use the 3D data provided by the 3D scanner and we have developed an inspection solution based on 3D point cloud analysis. We have also developed a method to compute the best viewpoints for the sensor held by the robot, in order to obtain an optimal view of each component to be inspected. The view-point selection is performed off-line (before the on-line inspection) and it exploits the CAD model of the mechanical assembly. The proposed automatic computer-vision-based inspection system has been validated in a context of industrial applications. Our software solution for 2D image analysis has been deployed on the robot platform as well as in a hand-held tablet. Since it requires a 3D sensor, our approach based on 3D point cloud has been tested in the robotic context only.
Fichier principal
Vignette du fichier
NDT_Aerospace_2019.pdf (8.06 Mo) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-02374691 , version 1 (21-11-2019)


  • HAL Id : hal-02374691 , version 1


Hamdi Ben Abdallah, Igor Jovančević, Jean-José Orteu, Benoit Dolives, Ludovic Brèthes. Automatic inspection of aeronautical mechanical assemblies using 2D and 3D computer vision. NDT AEROSPACE 2019 - 11th Symposium on NDT in Aerospace, Nov 2019, Paris-Saclay, France. ⟨hal-02374691⟩
139 View
145 Download


Gmail Facebook Twitter LinkedIn More