Most plants within process industries employ frequent low-fidelity (LF) online sensor data together with sparse high-fidelity (HF) laboratory measurements, e.g., for product quality monitoring. While LF data are used for real-time operation, HF data recalibrate LF sensors occasionally. It is though rare that historical HF data are used for long-term improvement of LF sensors. We present a multi-fidelity (MF) soft-sensor framework that combines these two data sources. In two studied use cases, the proposed MF model reduces the prediction error by 20–50% compared to LF sensors and reproduces HF trends with noticeable accuracy. The proposed method is general and transferable to other processes with similar data structure, providing interpretable results for improved monitoring and control.
Data-based multi-fidelity modeling for online sensors correction
Marco Vaccari;Riccardo Bacci di Capaci;Gabriele Pannocchia;
2026-01-01
Abstract
Most plants within process industries employ frequent low-fidelity (LF) online sensor data together with sparse high-fidelity (HF) laboratory measurements, e.g., for product quality monitoring. While LF data are used for real-time operation, HF data recalibrate LF sensors occasionally. It is though rare that historical HF data are used for long-term improvement of LF sensors. We present a multi-fidelity (MF) soft-sensor framework that combines these two data sources. In two studied use cases, the proposed MF model reduces the prediction error by 20–50% compared to LF sensors and reproduces HF trends with noticeable accuracy. The proposed method is general and transferable to other processes with similar data structure, providing interpretable results for improved monitoring and control.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
