Effect Of Measurement Uncertainty On Artificial Vision Methods, For Quality Control On Composite Components

Antonella Gaspari, Emanuela Natale, Armando De Silvestri , Giulio D’Emilia

Abstract

In this work, an inspection strategy by a vision system is analysed, for the identification of surface and aesthetical defects, with reference to composite components for automotive and aeronautical industrial sectors. Attention is paid to the background identification, since the specificity of the application requires particular care in order to avoid misunderstandings and false negatives during the detection phase. The evaluation of the parameters setup effects is used for the identification of the main uncertainty contributions, which is a strong support for the most suitable choice of the monitoring strategy. The robustness of the approach is studied with reference to several laboratory datasets, representing some commonly found issues for an easy in-field transfer. To this aim, some commercial tools available in Matlab ® environment have been used. The obtained results encourage to monitor the variability of the performances rates, depending on the qualitative levels to be achieved during the operating conditions and on the desired reliability of the approach.

Keywords

Measurement uncertainty; Artificial intelligence; Deep learning; Vision systems

Download

IMEKO-TC10-2020-027.pdf

DOI

-

IMEKO TC

TC10 - Measurement for Diagnostics, Optimization and Control

Back to the proceedings