We address the problem of real-time monitoring of products’ properties from spectroscopic measurements. Spectra are used as inputs for soft-sensors that estimate outputs difficult to measure on-line. To overcome the issues associated to calibrating such models from high-dimensional inputs, we propose to select the relevant inputs emerging from the topological structure of the data. The approach is independent on the estimation model to be embedded in the sensor. Being based on the original spectral features, the models retain the interpretability of the underlying system. The application of the method is illustrated on two cases from refining and pharmaceutical industry.
Measures of Topological Relevance for Soft Sensing Product Properties
BARATTI, ROBERTO
2007-01-01
Abstract
We address the problem of real-time monitoring of products’ properties from spectroscopic measurements. Spectra are used as inputs for soft-sensors that estimate outputs difficult to measure on-line. To overcome the issues associated to calibrating such models from high-dimensional inputs, we propose to select the relevant inputs emerging from the topological structure of the data. The approach is independent on the estimation model to be embedded in the sensor. Being based on the original spectral features, the models retain the interpretability of the underlying system. The application of the method is illustrated on two cases from refining and pharmaceutical industry.
I metadati presenti in IRIS UNICA sono rilasciati con licenza Creative Commons CC0 1.0 Universal, mentre i file delle pubblicazioni sono protetti da diritto d'autore, salvo diversa indicazione.
