The function of neural networks is to classify self-initiated emails from machines and/or plants from within the company to the maintainer, in order to perform predictive maintenance.

The Keras framework was used in Python.

The optimisation algorithm we have used is adamax, a variant of gradient descent which should give better results in the case where features are represented by a sparse matrix, not specifying a Learning Rate but simply using the value that Keras uses by default.

The dataset used to train our network is the IMDB Movie Review Dataset.

The network was optimised and tested using various open datasets provided by Kaggle.

A neural network can be used to refine the results of a linear regression in order to plan predictive machine maintenance. In the case of a printing press, the neural network can be used to plan predictive maintenance of the machine’s axes, motor status, bearing status and print roller unbalance.