Modelling of a Heat Conduction Calorimeter Used in Cement Plant Automation

Abstract: A model is used to describe a system physically or mathematically and to give information about changes in the system variables. Physical knowledge about the system and its quantities can be useful to build a physical model obeying the laws of nature. However, if the system is to be regarded as a black box with only input and output signals, system identification can be used to fit a mathematical model to describe the relation between the signals. The signals are usually physical quantities but the estimated model parameters do not necessarily reflect physical aspects of the system. The goal of estimating a model is either to calculate the effect given the cause, such models are called forward or direct models, or to calculate the cause given the effect, which is the objective of this thesis. These models are called inverse models and are widely used in heat conduction problems, e.g., estimating the heat flow rate at the source given measured temperature at a point. In this thesis the goal is to estimate the true heat flow rate from the hydration reaction in a cement sample when mixed with water from heat rate measurements in the isothermal calorimeter I-Cal Flex. The problem with such measurement is that it starts only after about 20-30 seconds, which is the time it takes for mixing the water and cement in an ampoule and charging it in the calorimeter. Under that time the cement sample temperature rises due to heat produced from the chemical reaction. The measurement process is automated in a robot called PolabCal and the measurement signal could possibly be used to control cement plants. The problem is that the measurement both includes the heat rate from the over-temperature that sample gained prior charging in the calorimeter and information about the dynamics of the instrument. There is therefore an interest in the pure heat rate from the cement hydration reaction as it is believed to be more informative than the measurement signal. Estimating the inverse model in this thesis was made by evaluating three methods; to estimate a forward model then to invert it, to estimate an inverse model directly from inputs and output measurements, and lastly to build a physical forward model and then use it to calculate the input.