Efficient learning in Artificial Intelligence applied to Industry

The overall objective of the ArgIA project is to generate efficient learning methods in artificial intelligence applications carried out in industry, in such a way as to maintain or increase the performance of algorithms, drastically reducing both the volume of data required and the development time of such projects. 

ArgIA is therefore a basic research project within the field of artificial intelligence, which aims to address the challenge of learning from a small amount of annotated data, turning this learning into efficient learning. 

PROJECT RESULTS AND IMPACT

The solution proposed by ArgIA to address the situations outlined in the opportunity section is to apply efficient learning techniques based on transfer learning to transfer the knowledge generated in the first modelling exercise and, based on a few annotated samples, generate the new model. Based on these specific results, the technological advances that will be achieved in the project are as follows: 

- Learning models for predicting the hardness and composition of steel based on its microscopic metallography, and knowledge transfer applied to predicting its hardness based on sound waves.

- Optimisation of the actual use (and life cycle) of refractory materials based on deep models and their transfer to different types of facilities.

- Generation of few-shot techniques for learning and detecting anomalies in production defects, and domain transfer for surface inspection of steel in different types of parts and references, as well as adaptation of weak supervision techniques for defect classification. 

- Data collection based on advanced simulation and generation of synthetic data in component manufacturing, for reuse and adaptation to other contexts and environments.

- Incremental and adaptive learning models in manufacturing processes to characterise the process and transfer the models to other contextual cases (different materials, different process conditions).