Artificial Vision to assess product quality in InterQ
Video Systems (videosystems.it), leveraging its consolidated experience in industrial monitoring through image processing techniques, proposes within WP3 (InterQ-Product), its artificial vision and AI solutions to extend the opportunities of dimensional or surface quality control of mechanical components present in the supply chain of the industrial use cases of InterQ.
The possibility to detect through visual inspection surface defects in an automatic and reproducible way, replacing human intervention, is of great interest in the field of quality production within a paradigm of zero-defect manufacturing. The requests to verify the performance of these techniques have been concentrated in particular in the aerospace use case, although not exclusively: this use case involves the control of the various stages of processing of high quality components both static (turbine casing) and rotating (turbine disk), to be assembled for the construction of engines, in order to obtain a quality hallmark of process, product and data, which is aimed to be the added value of InterQ for the quality tracking of the production process.
The video shows the experimental set-up used to acquire images of the surface of a turbine casing in the Video Systems laboratories. The properly programmed robotic arm allows the coverage of the surface to be scanned; the other hardware element of interest is the circular multi-sector illuminator, with each sector programmable with four colors (RGB + white), which allows to expand the dataset of available images and consequently extend the analysis techniques to be used for surface inspection.
For example, using Shapes from Shading (SFS) techniques, which can exploit the separation of color planes by analyzing the surfaces that reflect differently depending on the orientation of the colored source, defects with three-dimensional characteristics, such as scratches, marks, porosity, exceeding material, are highlighted.
Through a work of annotation and classification of the available defects, possibly pre-processed with innovative techniques such as the one mentioned above to increase the resolution and therefore the detectability, it is possible to train specific neural networks that can learn effectively and then operate an automatic and reliable defect recognition. For these operations, Video Systems provides a suite of user interfaces designed to facilitate the various phases of annotation/classification, training, as well as analysis including pre-processing steps and use of trained models.