Development of defect detection models in aeronautical industrial images through the use of Artificial Intelligence

IND2023/TIC-27224

CAM

As part of the quality assurance process, industrial products are subject to numerous inspections such as visual, liquid penetrant, eddy current, ultrasonic or X-ray inspections. In production, the analysis of the X-ray images is done by certified inspectors using expensive computer (medical) displays and under low light conditions, in order to capture the maximum contrast as possible to detect fine-grain defects. Usually, there are also limitations on the maximum time that the inspector can spend reviewing the images before a compulsory break is required to avoid visual tiredness and fatigue. On a single work turn, a human inspector may review a few hundred of images in a process that can be monotonous and error prone.

Hopefully, the use of machine learning (ML) techniques, in particular, Deep Learning (DL) methods have shown an impressive advance in the last decade to resolve Computer Vision (CV) problems while achieving or even exceeding human performance. One of the major machine Learning model groups is Discriminative models, which discriminate between different kinds of data by calculating a probability of how likely a label is on the input instance. In particular, convolutional neural networks (CNN) are now the industry standard to solve classification, object detection, and segmentation
problems.

Models based on Deep Learning will be developed to provide increasing support to human inspectors in carrying out their work until eventually, the model takes responsibility for partial inspections of the industrial parts manufactured by ITP Aero.