November 2022

Excerpt from the CTCV publication: Technical Journal of Ceramics and Glass (November/December 2022)

Authors:

Margarida Felício [1], Rui Carreira [1] Idalina Eusébio [1], Filipe Soares [2], João Gonçalves [2], Luís Rosado [2], Alberto Gil [3], André Santos [3], Victor Francisco [4], Rui Neves [4]

[1] MATCERÂMICA - Fabrico de louça SA, [2] Associação Fraunhofer Portugal Research (FhP), [3] AGIX, Lda, [4] CTCV – Centro Tecnológico da Cerâmica e do Vidro

Ongoing developments

The project began with a survey, selection and testing of existing technologies on the market for detecting surface defects and structural integrity. This work included a preliminary analysis of the technologies applicable to the context of the project in order to validate which existing functionalities on the market were the most viable to acquire and to confirm which functionalities would be developed in the context of the project.

This was followed by a study of acoustic inspection on samples, comprising tests in a controlled environment, using samples of ceramic products, in order to identify specific and differentiating properties of each type of tile from specific lines selected for this purpose. As part of the task, a search was carried out for studies on acoustic inspection of ceramic tiles. Studies were selected that were based on the extraction of characteristics from the audio signal emitted by ceramic tiles and, in particular, that were aimed at quality control. The response to the impact impressed on different types of ceramic tiles was evaluated in the temporal, spectral and statistical domains. This task made it possible to deepen the team's knowledge of the problem addressed, thus enabling better assessment and mitigation of risks in prototyping tasks.

An exploratory study into visual inspection on samples was carried out and the applicability of existing solutions to the VAICeramics project was analysed. The need for a high-priced software licence for the solution with the greatest potential, the difficulty of customising computer vision software libraries (for example, not compatible with the need to explore more segmentation, object detection and classification methods), the limitation of using annotation tools specific to the use case, and the difficulty of acquiring specific hardware that was not available for delivery (for example, smart cameras), reinforced the need for the VAICeramics consortium to develop its own software with the acquisition of cameras with image sensors and communication protocols widely used in the artificial vision industry.

A report on the characterisation of defects occurring in ceramic products was also drawn up, with the aim of characterising and selecting the main and most common defects in all production that occur in utilitarian ceramic products, based on the production process of the user company.

This characterisation of defects was carried out by processing the information available in terms of the occurrence of each type of defect, namely: 

Winding / Pores

Cracked

Rubbish Finish

Spots

Lack of Glass

Air in the pulp

Oven Waste

Broken freight

Since one of the project's objectives is to create a system for automatically detecting and identifying defects in ceramic pieces using artificial intelligence methods, it was necessary to create a platform for manually annotating defects and areas of the pieces so that the system can learn to identify them correctly. Usability tests were carried out, in which some participants interacted with the annotation platform, in order to assess its usability and identify some aspects for improvement.  The aim of the tests was to assess user interaction with the interface and to understand the state of the annotator's general usability.

The requirements for the general system architecture and back-end and integration services, the data acquisition and machine learning specification and the front-end specification have also been developed.