Machine learning based anomaly detection for industry 4.0 systems.

Abstract

This thesis studies anomaly detection in industrial systems using technologies from the Fourth Industrial Revolution (4IR), such as the Internet of Things, Artificial Intelligence, 3D Printing, and Augmented Reality. The goal is to provide tools that can be used in real-world scenarios to detect system anomalies, intending to improve production and maintenance processes. The thesis investigates the applicability and implementation of 4IR technology architectures, AI-driven machine learning systems, and advanced visualization tools to support decision-making based on the detection of anomalies. The work covers a range of topics, including the conception of a 4IR system based on a generic architecture, the design of a data acquisition system for analysis and modelling, the creation of ensemble supervised and semi-supervised models for anomaly detection, the detection of anomalies through frequency analysis, and the visualization of associated data using Visual Analytics. The results show that the proposed methodology for integrating anomaly detection systems in new or existing industries is valid and that combining 4IR architectures, ensemble machine learning models, and Visual Analytics tools significantly enhances theanomaly detection processes for industrial systems. Furthermore, the thesis presents a guiding framework for data engineers and end-users.