Abstract
Reducing CO₂ emissions in the atmosphere is a critical task, and carbon capture and storage (CCS) plays a crucial role in various actions for mitigating climate change. Safety issues and fiscal metering in CCS systems require reliable sensor measurements. In this work, we propose an architecture for sensor validation, i.e., performing sensor-fault detection, isolation, and accommodation (SFDIA). The architecture is layered and based on soft sensors and a soft classifier. Both the soft sensors and the classifier are built as neural networks. The performances of the SFDIA architecture have been assessed using real-world measurements from a real-scale research facility at SINTEF Energy Research, namely DeFACTO, and synthetically generated faults. Several experiments have been conducted to explore the transferability of the trained models across time and space.