Design of an embedded soft sensor for estimating the apparent calorific value of alternative solid fuels in a cement plant

Abstract

This thesis investigates the estimation of the apparent calorific value of alternative solid fuel in a cement calciner using only process signals available in the plant control system. Because this quantity is only available after daily laboratory analysis, the work first evaluates its observability, then develops a calibrated digital twin to generate realistic, labeled transient events. The results show that a recoverable signal exists, although it remains weak and partially masked by closed-loop control. A linear model with appropriate feature engineering achieves the best performance, while the energy-balance approach remains highly sensitive to uncertainty in fuel mass measurement.