Condition Monitoring Oriented to Predictive Maintenance Based on Autoencoders

In recent years, technological advances in industrial automation, such as IoT, artificial intelligence and advanced sensors, have led to important developments in production management and device maintenance. Therefore, machine condition monitoring provides essential data for predictive maintenance strategies since anomaly and fault detection allow prediction and prevention of possible failures, optimising costs and reducing machine downtime. The thesis aims to deepen the application of autoencoders, a class of deep learning models, in monitoring the operating conditions of industrial devices for anomaly detection. Therefore, after an overview of maintenance strategies and artificial neural networks, the work focuses on the practical implementation of an autoencoder for condition monitoring. The aim is to evaluate the model’s effectiveness in data reconstruction and analyse the data representation in the latent space to understand its structure.