On Multi-modal Anomaly Detection Techniques Applied to Industrial IOT

Abstract

In this work, anomaly detection in an Industrial Internet of Things (IIoT) multimodal context is evaluated from multiple perspectives. Particularly, a significant literature review was conducted to ascertain the highest value avenues for research. Ultimately, the entire thesis can be divided into three main outcomes. The first is the introduction of a high functioning Variational Autoencoder (VAE) model that solves the issue of absent or missing modes. Next, the theoretical construct of Functional Data Analysis (FDA) is applied to address open areas of research in high noise and high dimensional contexts. Finally, Quantum-based methods are utilized to better identify difficult, nascent faults in the IIoT. Each of these works are evaluated on high quality real-world benchmarks; proving their efficacy. Collectively, these works address some of the most challenging current issues in IIoT anomaly detection