Smart monitoring system to diagnose faults in a greenhouse

Abstract

The greenhouse fault diagnosis is a challenging task due to the interaction between variables of different phenomena: biological, physical, electrical and mechanical. In this article we propose a novel intelligent monitoring system to detect faults in a greenhouse that mimics a human operator to a certain extent, using a method based on deep learning and artificial vision. We design our algorithm for four novel fault detection architectures: Micro Output Manifold Fault Detection (MOMFD), Output Manifold Fault Detection (OMFD), Attention Fault Long Short Term Memory (AFLSTM) and Fault Long Short Term Memory (FLSTM). Then, we compare them with the well known Long Short Term Memory (LSTM). Results obtained from simulations of the greenhouse show that MOMFD presents the shortest inference time about 1 second and the highest accuracy for training and validation 93.9 and 95.56 percent, respectively.