Indirect adaptive neural control for precalcination in cement plants

Control of the precalcination degree in the precalciner of cement plants is a problem of great importance due to its effect to the quality of the clinker, the consumed energy and the byproducts of the whole cement pyroprocess. Divergence of the desired precalcination degree of the raw mix may cause increased carbon monoxide production which is a significant pollution factor that under certain circumstances may lead to explosive gas mixtures. The basic inputs of the precalciner are the temperature and pressure of the tertiary air, temperature and mass flow rate of the raw mix, and the mass flow rate of the fuel, namely pulverized coal. From these inputs, the coal mass flow rate is the only actuatable one. The outputs of the precalciner are the calcinated raw mix, characterized by its mass flow rate, temperature and degree of precalcination, and the abgasses consisting of oxygen, carbon monoxide, carbon dioxide, water and nitrogen, characterized by their temperature and the concentrations of the above elements. Some of the variables of the process, namely the raw mix and coal mass flow rates, the raw mix temperature, the temperature and pressure of the tertiary air and abgasses, and the oxygen concentration in them, can be measured by appropriate sensors. The rest variables can be computed from the measurable variables after appropriate approximations. In this paper, the precalcination degree is controlled via a dynamic controller that regulates the abgasses temperature. The parameters of the controller are computed on the basis of a polynomial neural network that identifies off-line the variations (around nominal values) of the real industrial measurements of the variables of the process. The results are illustrated via simulations of the closed loop system where it is shown that the abgasses temperature is set to a desired level while a desirable precalcination degree is derived.