Research database

The major challenges are represented by: 1. Real-time response Computational costs of the artificial intelligence algorithms are constrained by the necessity of detecting defects on a layer-by-layer basis. To tackle this problem, we will consider software/hardware acceleration strategies. 2. Need for large number of training images and datasets ML techniques are built on top of a pre-annotated training sets. This set should be: Massive ? in case of deep learning, huge amount of experimental data Balanced ? same number of examples of different experimental conditions (faulty/not faulty, different categories of defects, different geometries of manufacts, different materials, etc.) 3. Need for heterogeneous data integration The algorithms need to integrate data from different sources (cameras, sensors, CAD, post-processing tests, user manuals, etc.), which may be structured (e.g. images) or unstructured (e.g. manuals) and have different formats and granularities