On the use of embedded models and advanced analytics to model complex processes in the cement industry
Document Type
Article
Year
2022
Authors
A. Piladarinos, A. Kokossis, I. Marinos, T. Gentimis
Source
Computer Aided Chemical Engineering, Volume 51, pp. 565-570, 2022
Keywords
Embedded Model; ANN; Deep Learning; Cement Grinding; Ball Mill
Abstract
The paper explains a generic and systematic approach in the development of embedded models that could be further used for model reduction. The systems approach makes a structured and systematic use of data as they are produced at three distinct stages: simulation assignments by means of spatial differential equations, optimization runs that regress parameters for each simulation, and deep learning training that converts parameters into functions of system variables. Simulation models refer to steady-state operations of closed-circuit grinding models formulated as differential equations with parameters treated as degrees of freedom. Results from this implementation present a consistent accuracy improvement over the model used as basis.
