| KEYWORD |
Microelectronics
Edge AI: On Device Learning for Predictive Maintenance
Tesi esterna in azienda
Parole chiave MACHINE LEARNING, PATTERN RECOGNITION
Riferimenti LUCIANO LAVAGNO
Riferimenti esterni Marcello Babbi, Reply Torino
Gruppi di ricerca Microelectronics
Tipo tesi APPLIED RESEARCH
Descrizione Reply has developed a data-driven methodology for diagnostic (fault detection) and predictive
maintenance based on raw data from the field. Industrial needs require to integrate the diagnostic
system into low-power embedded platforms for both acoustic and vibrational signals. The thesis
project aims to develop ML/DL algorithms able to detect faults (i.e. anomalies) and perform ondevice-
learning to adapt to different environmental conditions, components, and sensor positions.
Conoscenze richieste Python programming, some knowledge of Machine Learning
Note The student will be involved in:
❑ SW requirements definition for edge deployment;
❑ Anomaly detection problem definition (supervised or semi-supervised);
❑ Data pre-processing and feature engineering;
❑ AI algorithms (e.g. Deep Neural Network, Encoder-Decoder) development for anomaly detection;
❑ On-edge performance optimization and compression methods development (e.g. pruning, quantization);
❑ On-device-Learning module development on MCU for model parameters learning;
❑ Hardware in the Loop validation and on-field testing.
Scadenza validita proposta 11/10/2024
PROPONI LA TUA CANDIDATURA