Industry 4.0 has played an important role in the digitalization of modern industrial processes, where multiple sensors are typically used to monitor the industrial systems. This means that large volumes of data regarding the performance of the systems are now available, which are then used for various industrial applications that range from process control and optimization to maintenance operations. In this context, processing these volumes of data to detect and diagnose anomalies has become a necessity as industrial systems are prone to faults and an undetected fault can lead to critical damage, besides reducing the productivity and increasing the maintenance operation costs.
This thesis presents a large-scale industrial monitoring system to detect and diagnose anomalies in multi-sensor systems. The monitoring system is composed of three main modules. First, a supervised Deep Learning (DL) based anomaly detection system that identifies anomalies within multi-sensor systems. Second, an anomaly diagnosis system that relies on interpretability methods to explain how the DL based anomaly detection model reaches a decision, thus identifying in which sensors and time span the anomalies occur. Third, a large-scale monitoring system based on Big Data and cloud computing technologies to process the large volumes of data in a fast, scalable, and fault-tolerant way. Therefore, the anomaly detection and diagnosis systems can be deployed in the cloud to support the monitoring of multiple industrial systems. To validate the proposal, each module has been validated in an industrial case study.
The algorithms and methods resulting from this thesis have demonstrated to be suitable for detecting and diagnosing anomalies in industrial multi-sensor systems since they improve state of the art results.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados