LSTM-based Siamese Networks for Fault Detection in Meteorological Time Series Data

In this paper, we present a new method for fault detection in meteorological data collected by field sensors used in vineyards. In particular, a deep learning approach is applied to the analysis of temperature and humidity time series, which involves a pipeline based on a Long Short-Term Memory model for data reconstruction and a Siamese network for fault classification. A data-driven approach is used to create a new dataset of synthetic faults, generated by considering the most probable causes of failure of the system, ascribable either to the measurement process (hard or soft sensor fault) or to a data transmission failure. The obtained results prove that the proposed pipeline is capable of recognizing a persistent fault in the acquired signals with a mean accuracy of about 92% and 94% for the temperature and the humidity time series, respectively.