Séminaire

Mardi 7 Mars 2023 à 11h00.

Seismic events prediction in knitted fabric through deep learning.

''Knitted fabric, a meta-material created from a single thread, deforms by discontinuous and discrete slips of the yarn's contact points, that can then propagate to neighboring stitches, resulting in an avalanche phenomenon. Those intermittent and scale invariant events are displayed both in the force signal of the system and in the stitches' displacement field, and their systematic study define the limits within which the knitted fabric can be considered an earthquake-like system. As in most analogous seismic models, the peculiar statistics of the corresponding time-series severely jeopardize the accurate occurrence's prediction of extreme events. We therefore introduce several machine learning techniques that allow us to design, evaluate and compare not only the prediction of these “knitquakes”, but also decision-making policies - as for evacuation - in a risk management context in an imaginary city, seismically active based on the crackling dynamics of the mechanical response of the knit.''