ETD

Archivio digitale delle tesi discusse presso l'Università di Pisa

Tesi etd-02022022-213759


Tipo di tesi
Tesi di laurea magistrale
Autore
GHERARDI, MARIO ALBERTO
URN
etd-02022022-213759
Titolo
Development of multi-modal deep feature learning architectures for degradation stage classification
Dipartimento
INGEGNERIA DELL'INFORMAZIONE
Corso di studi
COMPUTER ENGINEERING
Relatori
relatore Prof. Cimino, Mario Giovanni Cosimo Antonio
relatore Prof. Alfeo, Antonio Luca
relatore Prof. Vaglini, Gigliola
Parole chiave
  • Predictive Maintenance
  • Similarity
  • Contrastive Learning
  • Autoencoder
  • Multi Modal Data Fusion
  • Classification
  • Deep Learning
Data inizio appello
18/02/2022
Consultabilità
Non consultabile
Data di rilascio
18/02/2025
Riassunto
Predictive maintenance has acquired great popularity with the advent of the 4.0 industry. For this reason, many deep learning approaches has been used in predictive maintenance problems to learn relevant features from data. In this work we focus on exploiting the multimodality of the input data, proposing different architectures to improve the class separability of the features space. In particular, we focus on autoencoders that are known for their ability to distil important information from an input data, allowing the reconstruction of the input itself. However, it is difficult to force an ordered and semantically meaningful representation of the latent space: the information distilled in order to reconstruct the input may not be the information of interest for the analysis. To facilitate this process, it is possible to provide autoencoders with different inputs related to the same dynamics and force them to learn a shared representation in the latent space. We also implemented an architecture that exploit another recent deep learning concept already diffused in the image recognition field called contrastive learning, implementing an encoder based on similarity. These architectures will be used to train a classifier to classify the degradation stage of industrial bearings.
File