SIGURD KARIN (CH)
| WO2002022246A1 | 2002-03-21 |
| EP1382905A1 | 2004-01-21 | |||
| US5687077A | 1997-11-11 | |||
| US6790034B1 | 2004-09-14 | |||
| US5707444A | 1998-01-13 |
| P A T E N T C L A I M S
1. Method for predicting the evolution of concentration of volatiles in the clinker over time and for determining model parameters for model predictive control of the transforming process in a kiln System, comprising internally sloping cylinders for transforming raw meal made of gypsum and lime stone into clinker by pre-heating said raw meal in a pre-heater and adding fuel and burning said raw meal in a burner, some of said said system further comprising means for predicting the evolution of concentration of volatiles in the clinker over time and for determining model parameters for model predictive control of the transforming process, said predicting means comprising real-time input parameters from the transforming process, namely the concentration of the volatiles in the pre-heater (x(t)) and the concentration of the volatiles as a residue in the clinker (r(t)), said model parameters to be determined being a primary volatility (E- t ), describing the degree to which the volatiles are evaporated in the burner, a secondary volatility (E 2 ), describing the degree to which the volatiles are again evaporated from a portion of the solid phase in the pre-heater that has already been recycled at least once, and an adsorption degree (A), describing the degree to which the evaporated volatiles are adsorbed by the raw meal in the pre-heater, said evolution of concentration of volatiles in the clinker being determined in said predicting means by applying the mass balance equations x(t) = ABiCp C cU 1 (Y) + E 1 c mcc u 2 (t -τ) + E 2 x(t - τ)) - x(t) and r(t) = (I - E 1 )Cj WC cU 2 (t - T) + (1 - E 2 )x{t - τ) - r(t) , where χ(ή is the concentration of the volatile component in the pre-heater and r{t) is the concentration of volatile as a residue in the clinker, U 1 (Z) is the fuel input, u 2 {t) is the raw meal input and τ \s a variable time delay, B is the fraction of volatile that is not let out of the system as emissions having a constant value in the range of 0 to 1 , cpcc and c mC c are vectors where the i th element denotes the concentration of volatile in the i th component of the fuel or feed, respectively; said method comprising the steps of - varying the secondary volatility parameter (E 2 ) at b+i * c, with i=0,...,n and c=(1-b)/n, - implementing a set of n+1 converging extended Kalman filters for determining the primary volatility and the adsorption degree, and
- estimating the second volatility parameter (E 2 ) with a time delay τ by comparing the changes in volatiles estimated by the Kalman filters to actually measured values.
2. Method of claim 1 , characterized in, that the variation of the second volatility parameter (E 2 ) depends on the type of volatiles and the variation parameter b is chosen accordingly.
3. Method of any of claims 1 to 2, characterized in, that the Kalman filters are based on the following mass equations: x(t) = AB(Cp C cU 1 (t) + E x c mcc u 2 (t ~τ) + E 2 x(t - τ)) - x(t) , and r(t) = (l-E 1 )c mcc u 2 (t - T) + (1 - E 2 )x(t - τ)- r(t) ; x(t) denoting the concentrations of volatiles in the pre-heater, r(t) denoting the concentrations of volatiles as a residue in the clinker, B, c FCC and c mcc being constants, u x (t) being the fuel input, u 2 (t) being the raw meal input and τ being a variable time delay.
4. Method for predicting the evolution of concentration of volatiles in the clinker over time and for determining model parameters for model predictive control of the transforming process in a kiln System, comprising internally sloping cylinders for transforming raw meal made of gypsum and lime stone into clinker by pre-heating said raw meal in a pre-heater and adding fuel and burning said raw meal in a burner, some of said said system further comprising means for predicting the evolution of concentration of volatiles in the clinker over time and for determining model parameters for model predictive control of the transforming process, said predicting means comprising real-time input parameters from the transforming process, namely the concentration of the volatiles in the pre-heater (x(t)) and the concentration of the volatiles as a residue in the clinker (r(t)), said model parameters to be determined being a primary volatility (E-i), describing the degree to which the volatiles are evaporated in the burner, a secondary volatility (E 2 ), describing the degree to which the volatiles are again evaporated from a portion of the solid phase in the pre-heater that has already been recycled at least once, and an adsorption degree (A), describing the degree to which the evaporated volatiies are adsorbed by the raw meal in the pre-heater, said evolution of concentration of volatiies in the clinker being determined in said predicting means by applying a set of mass balance equations defining the changes in concentrations of volatiies over time in the pre-heater and as a residue in the clinker on a basis of said model parameters said method comprising the steps of
- varying the secondary volatility parameter (E 2 ) at b+i*c, with i=0,...,n and c=(1-b)/n,
- implementing a set of n+1 converging extended Kalman filters for determining the primary volatility and the adsorption degree, and
- estimating the second volatility parameter (E 2 ) with a time delay by comparing the changes in volatiies estimated by the Kalman filters to actually measured values.
5. Method of claim 4, characterized in, that mass balance equations are equations comprising explicit nonlinearities and/ or differential algebraic equations and/ or partial differential equations and/ or finite elements.
6. Method of any of claims 4 or 5, characterized in, that parameter estimation is based on sensitivity equations and/ or direct optimization and/ or genetic algorithms. |
Model for a Cement Kiln Process
D E S C R I P T I O N
Field of the Invention
The invention relates to the field of cement kiln. It relates to a method for determining model parameters for model predictive control of the transforming process in a kiln System
Background of the Invention
Cement is produced in large, hollow, internally sloping cylinders known as kilns where raw meal is heated with fuel to 1500°C and transformed into clinker. Gypsum and lime stone are the main ingredients in the raw meal fed to the kiln, while the fuels include both conventional fuels such as coal and oil and alternative fuels like tires, plastics and solvents. Some components in the raw meal and the fuel are volatile and evaporate to some extent as the raw meal moves from the pre-heater down towards the burning zone of the kiln, where the fuel is added - sulfur, chlorine and alkalis are important examples of such components. The evaporated components travel in the opposite direction up the kiln, until eventually the volatile components condense back and close the circle, as is schematically depicted in the figure below.
Volatile components play an important part in the mass and energy balances that shape the cement production process. The volatile components are as a rule viewed as contaminations in the clinker - the concentration of volatiles in the clinker should therefore be kept below an upper limit and as low as possible for quality purposes. To reduce the overall volatile concentration in the system, volatiles can be let out as emissions in the gaseous phase, but environmental laws regulate the upper bounds for these emissions and, furthermore, energy is lost from the system which could otherwise be used to heat the raw meal. Recently, the negative cost associated with alternative fuels has prompted cement manufacturers to switch part of their fuel consumption to alternative fuels such as tires and solvents; the drawback is that many of these fuels contain high levels of volatiles like sulfur.
BESTATIGUNGSKOPIE
Model predictive control and mixed-integer programming are used to optimize cement production. Since constraints concerning clinker quality and emissions control also need to be satisfied, the concentration of volatile components naturally becomes part of the optimization constraints. For optimal results and margins as narrow as possible, the concentration of volatiles needs to be estimated and predicted as accurately as possible.
The circulation of volatiles in kiln operation has been extensively studied in experiments and simple mathematical models have been proposed based on mass balances that capture the essential features of the process; two or three volatility parameters indicating the degrees of adsorption and evaporation, respectively, are key parts of the models.
Both P. Weber (,,Alkaliprobleme und Alkalibeseitigung bei warmβsparenden Trockendrehόfen", Zement-Kalk-Gips, 17(8):335-344, 1964) and H. Ritzmann (,,Kreislaufθ in Drehδfensystemen", Zement-Kalk-Gips, (8):338-343, 1971) formulated equations describing the circulation of volatiles where three volatility parameters were introduced: an adsorption degree, describing the extent to which the volatile component is adsorbed to the solid phase in the pre-heater at condensation, a primary volatility, indicating to what degree the volatile component is evaporated in the burning zone from the raw meal, and a secondary volatility, indicating to what degree the volatile component is again evaporated from the portion of the solid phase in the pre-heater that has already been recycled at least once (recycled solid phase).
While existing mathematical models of the cement kiln assume the volatility parameters to be known and constant, these parameters are in fact time-varying functions of a number of chemical and mechanical factors; in a dynamic kiln environment, the assumption of known and constant volatility parameters is likely to give rise to an inaccurate prediction of the evolution over time of the volatile concentration, which in turn will lead to inaccuracy in the optimization. If the fuel quality and composition vary considerably, the volatility parameters may change drastically over time and the assumption of constant parameters is likely to give rise to an erroneous projection of the volatile concentration forward in time.
Description of the Invention
it is an object of the invention to provide a method to determine model parameters for model predictive control of the transforming process in a kiln System and thereby determine the concentration over time of volatile components in a kiln with the adsorption degree of volatile components to the solid phase at condensation in the pre-heater of the kiln and the evaporation degrees of volatile components from the raw meal and from the recycled solid phase presumed unknown and varying.
This object is achieved by a method according to claim 1.
Changing volatility parameters are determined and the concentration of the volatiles is estimated with a parameterized model of the circulation of volatiles. The estimates of the volatility parameters are then used to predict the evolution of the volatile concentration over time.
Brief Description of the Drawings
The invention will be explained in more detail in the following text with reference to the attached drawings, in which:
Fig. 1 schematically shows a cement kiln, and
Fig. 2 shows a volatiles evaporation-adsorption diagram.
Detailed Description of Preferred Embodiments
When volatility parameters of a cement kiln model are assumed constant and known, one mass balance equation is sufficient to predict the trajectory of the volatile concentration. With unknown and varying volatility parameters however, the system based on one mass balance equation turns unobservable.
Several different mass balance equations can be taken from cement literature, forming together with the original mass balance equation a system of equations with quite some redundancy. Exemplary equations represent mass balances for the volatile in the pre- heater, for the clinker residue, for volatile emissions in the chimney and for dust recirculation.
When introducing the volatility parameters as noise-driven augmented states, identifying the volatility parameters with dual estimation would be preferable. The information redundancy makes the augmented system unobservable for the three-parameter case; however, by combining the original equation with one more mass balance equation, two out of the three volatility parameters, the adsorption factor and one of the volatilities, can be estimated with an extended Kalman filter.
The two mass balance equations used, obtained from the cement literature, are given below as equations (1) and (2), where x(t) denotes the concentration of the volatile component in the pre-heater and r(t) is the concentration of volatile as a residue in the clinker. B, c FCC and c mcc are constants while u x {t) is the fuel input, w 2 (/) is the raw meal input and τ is a variable time delay. B, varying from 0 to 1, is the fraction of volatile that is not let out of the system as emissions; c FCC and c mcc are vectors where the i th element denotes the concentration of volatile in the i th component of the fuel or feed, respectively. x(t) = AB(Cp C cU 1 (0 + E 1 C KMC cU 2 (t ~τ) + E 2 x{t - T)) - x(t) ( 1 ) r(t) = Q.- E ι )c mcc u 2 (t-τ) + (l- E 2 )x(t~r) - r(t) (2)
The third volatility parameter has to be assumed constant and known - since no such information was available, this filter was not sufficient to predict the trajectory of the volatile concentration.
Experimental data show the range of the secondary volatility to be much reduced compared to those of the other two volatility parameters; while adsorption degree A and primary volatility Ei vary anywhere in the full range from 0 to 1 , secondary volatility E 2 is somewhat restricted to E 2 e [b,l] with bs [o.68,l], the value of b depending on the type of volatile. For sulfur, b varies in the interval [o.68,l] whereas for chlorine&=l.O, making the value of £ 2 =1.0.
By choosing the secondary volatility to be the parameter assumed constant and known in the filter, the additional information provided with the reduced range can be of help.
The prediction of the volatile concentration has rather low sensitivity to the value of the secondary volatility within the reduced range. The prediction of the volatile concentration obtained with the median E 2 = can be considered accurate enough. The low sensitivity is related to the fact that seen as a function of E 2 , the system function as
expressed above in equations (1) and (2) is linear. When E 2 varies in the interval of υnphysical values [0,0.68], the results were quite different with much higher sensitivity to the value of E 2 and qualitatively different trajectories.
Indirectly, the estimated secondary volatility can be validated with a time delay, by comparing the estimation and prediction errors obtained with different values of E 2 ; that way, the unobservability associated with the simultaneous estimation of all three volatility parameters can be eluded and an estimate also of the third parameter can be obtained.
The inventive method can be applied in a computer based simulation of the cement kiln. Model predictive control and mixed-integer programming are used to optimize cement production. The module predicts the evolution over time of the concentration of volatile components, a quantity which is used in the mixed-integer program solved in each optimization.
The inventive method can be extended to other configurations of a cement plant, comprising exhaust gas bypasses or inputs of alternative fuels at different points of the kiln. The inventive method is also not restricted to the application for cement kiln but can rather be extended to other industrial processes where volatile components appear or even to mathematically analogous problems not necessarily related to volatiles.
In the same way the inventive method is not restricted to the particular form of the equations displayed in (1) and (2) or to the usage of Kalman Filters. It can be extended to other versions of the equations, e.g. equations comprising explicit nonlinearities, differential algebraic equations, partial differential equations, finite element analysis, or others. The inventive method can further be extended to other methods for parameter estimation, like those based on "sensitivity equations" (see "Maximum Likelihood Parameter Estimation from Incomplete Data via the Sensitivity Equations"; Charalambous and Logothetis, 1999 American Control Conference, San Diego, CA, USA, which is herewith included by reference), direct (nonlinear) optimization, genetic algorithms or others.