Development of an engineering system for unburned carbon prediction

Within the computational methods used for the prediction of unburned carbon, coal combustion kinetics models, generally developed from the study of the real combustion process in experimental facilities, has the advantage to simulate the coal combustion process in a very realistic way. However, these models need the fluid and thermal behaviour in the boiler, which is usually obtained from simplified zonal approaches. The other group of models, namely CFD codes, present the opposite features. That is, they give a detailed description of the thermal and fluid dynamics behaviour in the boiler, but they use simple combustion models that cannot be used for a quantitative burnout determination. Moreover, the computing cost can be high and cannot be implemented in an on-line predictive system. edictive system developed in this work has the same structure as the so-called combustion kinetics models; however, it obtains the fluid and thermal description through CFD simulations. To solve the handicap of the high computational cost needed to run a CFD simulation, a neural network system is used to reproduce the solutions given by the CFD code. Moreover, a neural network system permits to interpolate in the range of variation used during the training stage, and thus, a predictive system covering the whole operational range of the plant can be obtained. s from the predictive system have been compared against those gathered at Lamarmora power plant (ASM Brescia, Italy), after carrying out a statistical study for validating and determining the prediction capability of the system. The comparison of both sets of data permits to conclude that the system predicts reasonably well over the whole range of operating conditions of the study plant.