Virtual Sensing and Sensitivity Analysis of Sour compression technique (SCU) of a Cement Manufacturing Plant

Abstract

The cement industry is a prominent source of CO2 emission to the environment. To minimalize the pernicious influence of the CO2 emission on the environment, CO2 capturing has been the focus of research. Sour compression technique (SCU) is a reliant and commonly used method for CO2 absorption. For a stable and efficient operation of SCU, a robust sensing and control system is vital. The data-based model also termed as databased virtual sensors have been attracting attention in the process industry for enhancement and replacement of the conventional hardware sensors such as flow meter, pressure gauge, and composition analyzer. In this study, a databased virtual sensor is designed to relate process conditions such as pressure, temperature, and flow rate to the carbon-capturing capability of SCU. An Aspen Plus based model of the SCU comprising of CO2 capturing, desulfurization and denitrification processes was developed. The process model was converted to dynamic mode through the interfacing of MATLAB-Excel-Aspen to achieve the behavior of real-time cement plant operation. Five hundred fifty (550) datasets were generated that consisted of process conditions and their corresponding values of the CO2, SO2 and NO in the process outlet streams. The data was used to develop the virtual sensor through ensemble learning, i.e., boosting. Prediction performance of the virtual sensors for CO2, SO2 and NO was 98.86%, 99.63% and 99.7%, respectively. Moreover, a sensitivity analysis was done on datasets to checkout any influence of input or set of inputs on output. Variance based SOBOL and Fast Amplitude Sensitivity Analysis (FAST) are techniques to figure out the impact of inputs. The results demonstrated that the proposed framework could be used effectively for composition monitoring of CO2, SO2 and NO in the exhaust stream of a cement production plant.