Hong-Cheol Ko - Academia.edu (original) (raw)
Uploads
Papers by Hong-Cheol Ko
Journal of Institute of Control, Robotics and Systems
Combustion optimization, determining the optimal combustion air flow by measuring O2 and CO conce... more Combustion optimization, determining the optimal combustion air flow by measuring O2 and CO concentrations of exhaust gas, is essential for reducing greenhouse gas emissions in steel plants. Recently, the tunable diode laser spectroscopy(TDLS) analyzer is used to measure O2 and CO concentrations. However it is expensive and sometimes out of order even less than the sampling type analyzer. Virtual sensors were developed to predict O2 and CO concentrations of the exhaust gas and used for detecting analyzer faults and backing them up. Four machine learning techniques, PLS, ANN, A-ANN, and XGBoost, were compared and evaluated for virtual sensor modeling. Among them, XGBoost had the highest coefficient of determination (R 2) of 80% or more and relative accuracy (RA) of 89% or more. Additional machine learning was performed to minimize errors caused by faults of the measurement sensors and increase the reliability of the virtual sensors. PCA and A-ANN were applied to 40 measurement sensors to build prediction models, and A-ANN had higher R 2 and RA of more than 90%. The study found that virtual sensors helped maintain combustion optimization without interruption from fault TDLS readings.
Korean Journal of Chemical Engineering, 1989
A computational procedure based on the arclength continuation method is designed for finding a se... more A computational procedure based on the arclength continuation method is designed for finding a sequence of solutions on variations of specifications in stagewise separation processes. The arclength continuation method can be implemented simply with an existing code that solves the modelling equations by Newton-Raphson method. The procedure is capable of direct handling of functional form specifications as varying variables. In order to promote the computational efficiency, emphasises are given to generation and handling of Jacobian matrices and calculation of integration step-sizes. The applicability of the designed procedure to the problems with multiple solutions is shown on numerical examples.
Journal of Institute of Control, Robotics and Systems
Journal of Institute of Control, Robotics and Systems
Combustion optimization, determining the optimal combustion air flow by measuring O2 and CO conce... more Combustion optimization, determining the optimal combustion air flow by measuring O2 and CO concentrations of exhaust gas, is essential for reducing greenhouse gas emissions in steel plants. Recently, the tunable diode laser spectroscopy(TDLS) analyzer is used to measure O2 and CO concentrations. However it is expensive and sometimes out of order even less than the sampling type analyzer. Virtual sensors were developed to predict O2 and CO concentrations of the exhaust gas and used for detecting analyzer faults and backing them up. Four machine learning techniques, PLS, ANN, A-ANN, and XGBoost, were compared and evaluated for virtual sensor modeling. Among them, XGBoost had the highest coefficient of determination (R 2) of 80% or more and relative accuracy (RA) of 89% or more. Additional machine learning was performed to minimize errors caused by faults of the measurement sensors and increase the reliability of the virtual sensors. PCA and A-ANN were applied to 40 measurement sensors to build prediction models, and A-ANN had higher R 2 and RA of more than 90%. The study found that virtual sensors helped maintain combustion optimization without interruption from fault TDLS readings.
Korean Journal of Chemical Engineering, 1989
A computational procedure based on the arclength continuation method is designed for finding a se... more A computational procedure based on the arclength continuation method is designed for finding a sequence of solutions on variations of specifications in stagewise separation processes. The arclength continuation method can be implemented simply with an existing code that solves the modelling equations by Newton-Raphson method. The procedure is capable of direct handling of functional form specifications as varying variables. In order to promote the computational efficiency, emphasises are given to generation and handling of Jacobian matrices and calculation of integration step-sizes. The applicability of the designed procedure to the problems with multiple solutions is shown on numerical examples.
Journal of Institute of Control, Robotics and Systems