• data science
• industry 4.0
• LSTM
• predictive maintenance
• process optimization
• RNN

This thesis is about the analysis of a dataset coming from sensors in industrial context, in particular from machineries for leak testing. From more than one year of data collecting, including cases of past failures and normal activity values, we are able to model the conditions of both situations and we want to predict when the next downtime needs to be scheduled, for performing the relative maintenance (predictive maintenance). Since the nature of data, that are sequential and with long term dependencies, we deploy Recurrent Neural Network (RNNs) for modeling such dependencies, in particular the Long Short Term Memories (LSTM). We also propose a comparison with non-recursive models.