Empirical modelling in non-linear predictive control : a coffee roaster application

Loading...
Thumbnail Image

Authors

Journal Title

Journal ISSN

Volume Title

Publisher

University of Pretoria

Abstract

This dissertation presents the development and implementation of a model predictive control (MPC) system for a coffee roasting process, to optimise roasting quality while minimising energy consumption. The study involved analysing historical temperature profile data and roaster inputs to develop a hybrid model, combining empirical and first principles techniques, which predicts the measured bean temperature as a function of the available roaster inputs. The combination of the first-principles model with empirical modelling techniques reduced validation data error by increasing measured temperature prediction accuracy. Subsequently, a nonlinear MPC was designed and tuned through a series of simulations, adjusting prediction and control horizons while limiting input changes relative to the real-time input value. The optimal configuration achieved a sig nificant reduction in the average usage of liquefied petroleum gas (LPG) while maintaining a wide input range. The impact of the intelligent modelling and control system on the reduction of raw material waste, the improvement of the quality of the final product, and the overall efficiency of the roasting process was evaluated, showing significant improvements in all three areas. The proposed system enables operators to perform simulations of roasts and reduce raw material wastage when developing roast profiles, providing a valuable contribution to the coffee roasting industry. Future work includes further investigation of hybrid modelling and nonlinear optimisation techniques.

Description

Dissertation (MEng (Control Engineering))--University of Pretoria, 2023.

Keywords

UCTD, Coffee roasting, Model predictive control, Process optimisation, Hybrid modelling, Machine learning

Sustainable Development Goals

Citation

*