AU772066B2 - Method and apparatus of manufacturing cement - Google Patents

Description

V t P/00/0 1 1 Regulation 3.2

AUSTRALIA

Patents Act 1990

ORIGINAL

9* G.e.

9 .9 99*9*9 999*** 9 COMPLETE SPECIFICATION STANDARD PATENT Invention Title: Method and Apparatus of Manufacturing Cement The following statement is a full description of this invention, including the best method of performing it known to me/us: FHPMELC699320001 .6 This invention relates to a method and apparatus of manufacturing cement. In its current embodiment, this invention relates to a method and apparatus for manufacturing cement, typically by way of a rotary kiln, in a wet or dry process. In a particular aspect, this invention relates to the operation and control of the grinding operation carried out by a grinding circuit in cement plants through an ordered-process sequence of process steps.

The conventional cement process comprises admixing predetermined proportions of calcareous(such as, limestone, marl, chalk and the like) argillaceous(sdch as clay, shale, slag, fly ash,(pozzolana) sand and the like) S and/or silicaceous materials, all being in slurry or dry form and fed into and burned within a cement forming kiln. The resulting clinker is mixed and pulverized with gypsum to form, generally, dry powdered modern day Portland Cement. The proportions of the ingredients in the mixture, their degrees of fineness, determines the chemical composition of the clinker and the final cement product. Additives are blended with one or more of the components to provide special cement characteristics. Essentially, the control is achieved purely by manual methods determined by the decisions of various operators with the result that a large portion of the process operates sub-optimally.

A critical operation in a cement plant is that of the grinding circuit. A grinding circuit in a cement plant, particularly a dry grinding circuit, consists of a ball mill, vertical roller mill (VRM) or roller press receiving a composite feed of several components, each of which discharges onto the mill feed conveyor through individual feed bins. Components (such as dry fly ash) which cannot be fed through a conveyor are pneumatically conveyed into the mill feed trunnion or discharge bucket elevator. Ball mills in cement grinding are typically, of the air swept type where material is transported through the mill by an air draft created by suction of a fan downstream. The mill contents discharge to a bucket elevator and are conveyed to a classifier, usually a dynamic air separator. In the separator, the material is classified into coarse material (grits) which is returned to the mill for regrinding, and fine material which is sent to a storage silo. Any fine dust in the mill draft line is also collected in an electrostatic precipitator (ESP) or bag filter and mixed with the fine material from the separator.

The grinding mill circuit is an important segment of the apparatus for manufacturing cement.

The grinding operation which is crucial to the quality and quantity of the final cement output is affected by regularly occurring periodic and cyclic disturbances which affect the throughput of the plant and the process. Each of the elements of the grinding mill circuit has its own control loop and monitoring i capabilities and adjustments for maintaining a given setpoint.

The object of the invention is to provide a grinding circuit for a cement mill in which the performance is controlled such that the total feed to the mill is maximized, compensating for any process drift while at the same time ensuring that the product (fine material) meets the fineness specifications for the process. Fineness is typically represented as a size distribution for raw mills, and the Blaine specific surface area for cement mills. The other process variables which are to be maintained within operating and/or safety limits are mill accumulation, mill motor power, bucket elevator power, separator power, and circulating load (total feed divided by fresh feed) or grits tonnage.

Hitherto, these operations of the grinding circuit in cement mills are primarily operator controlled. The invention provides a dry grinding circuit used in cement plants, which is provided with a multivariable model based supervisory control. The purpose of the method is to control the grinding circuit so as to achieve improved throughput, maintain product quality at the desired specifications, compensate for process-drift and ensure all equipment operating parameters is within prescribed safety limits. Control functionality is oo at a relatively higher supervisory level and is implemented by providing targets to lower level PID controllers through a Distributed Control System (DCS) or programmable Logic Controller (PLC) network to individually control the control loops associated with the grinding circuit.

Control of the grinding circuit is achieved by manipulating the tonnage of fresh feed to the mill, rpm of the separator, and dampers on various air lines in the circuit (typically mill suction line and separator air line). In the supervisory control methodology of this invention, the controller computes values for all the above-manipulated variables simultaneously, based on measurements of the variables to be controlled, and the performance targets, as described in the previous paragraph. Optimal control is ensured by the on-line solution of a constrained optimization problem., where the objective function to be minimized is the sum of the squares of the deviations from performance targets, with the manipulated variables as the decision variables and all operating and safety limits formulated as hard constraints in the optimization.

According to this invention there is provided a grinding circuit of a cement 4 plant, whose elements have a set of predetermined controlled and manipulated variables, which are controlled by a method comprising the steps of: providing a supervisory control computer with a data acquisition software interface [SuC] providing a graphical supervisory control interface [SCI]; inputting setpoints and constraints for the controlled and manipulated variables into the [SuC] via the [SCI] which in turn provides measurements to a Model Predictive Control software module [MPC] having multivariable state S* space models for the operative configuration of the elements of the grinding circuit; relating in the [MPC] the controlled variables [CV] to manipulated variables [MV] developed based on plant data received as feedback signals in the [SuC] via the data acquisition interface; providing a process control inference [PCI] software module which uses operating process data to estimate circuit product fineness as feedback for control computations in the [MPC]; providing a postprocessor rule based override software module [RBO] populated with a set of rules based on operator actions and adapted to receive control signals from the [MPC] and/or from the [PCI]; comparing signals received from the [MPC] and/or the [PCI] to generate operative or corrective signals; transmitting the operative or corrective signals to the [SuC] for conversion into instructions provided to operate or correct simultaneously all, or any of the elements of the grinding circuit, via distributed control systems[DCS]

I

programmable logic controllers

[PLC].

In accordance with one preferred embodiment of the invention, the method includes adaptation of models in the [MPC] to cover the entire operating region of the elements of the grinding circuit by identifying a large enough set of models, each valid over a subset of the operating region.

Typically, in accordance with this invention the MPC uses a weighted average e° of all model outputs for predictive control. Preferably, the model weightings are dynamically computed on-line depending on the operating point at any instant.

In accordance with a preferred embodiment of this invention the method b includes the step of updating the fineness model parameters using laboratory analysis data in the [PCI] if the model prediction is seen to drift from lab values.

Preferably, the method includes providing the [PCI] with inference rules currently used by operators and engineers to determine if disturbances (change in feed coarseness, moisture content, for example) are affecting the process.

Particularly the disturbance information is used in feedforward fashion by the [MPC] to provide rapid regulatory action by the [SuC].The invention will now be described with reference to the accompanying drawings, in which: Figure 1 represents the block diagram of the control methodology for the cement grinding mill circuit showing the various elements of the grinding mill circuit in accordance with this invention.

Figures 2 and 3 illustrate typical control implementation architecture, and typical supervisory control for the Portland Pozzolana Cement grinding mill circuit of Figure 1, in accordance with this invention; Figures 4 to 10 represent the graphical representation showing the throughput of the cement grinding mill circuit, product quality, input total feed with respect to time, accumulation percentage and operation of the Blaine estimation model in respect of an example of operation of the method in accordance with this invention.

Referring to the drawings, there is shown a method and apparatus for °-:Ccontrolling a cement grinding mill circuit in accordance with this invention.

A typical cement grinding mill circuit 10 is shown in Figure 1 of the drawings.

.The cement grinding mill circuit 10 in a cement plant, particularly a dry grinding circuit, consists of a ball mill, vertical roller mill or roller press GM, oo oi receiving a composite feed of several components, from hoppers 14, 16 and 18 wthich may typically be clinker, Gypsum and Pozzolana (fly ash) each of which discharges onto the mill feed conveyor 19 through individual feed bins 14,16 and 18.

Preferably, components (such as dry fly ash) which cannot be fed conveniently through a conveyor can be pneumatically conveyed into the mill feed trunnion or discharge bucket elevator. Ball mills in cement grinding are typically, of the air swept type where material is transported through the mill by an air draft created by suction of a fan 12 downstream.

The mill GM contents discharge to a bucket elevator 22 and are conveyed to a classifier, usually a dynamic air separator 20. In the separator 20, the material is classified into coarse material (grits) CV1 which is returned to the mill GM for regrinding, and fine material which is sent to a storage silo 28. Any fine dust in the mill draft line CV4 is also collected in an electrostatic precipitator (ESP) or bag filter 26 and mixed with the fine material from the separator drawn into the silo 28 line by means of a cyclone separator 24.

The object of the invention is to control the performance target of the grinding circuit 10 of the cement mill such that the total feed MV4 to the mill GM is maximized, while at the same time ensuring that the product (fine material) CV6 meets the fineness specifications for the process. Fineness is typically represented as a size distribution for raw mills, and the Blaine specific surface .*ooo i area for cement mills. The other process variables which are to be maintained within operating and/or safety limits are mill accumulation CV2, mill motor power CV3, bucket elevator power CV5, separator RPM MV3, and damper controls MV1 and MV2, and circulating load (total feed divided by fresh feed) or grits tonnage CV1. Of these MV1 to MV4 are manipulated variables whereas CV1 to CV6 are controlled/monitored variables.

The Controller Design is shown in Figures 2 and 3: The controller is based on a hybrid model predictive control methodology, and consists of three modules, described below.

Particularly referring to Figure 2, which illustrates the supervisory control architecture: Supervisory control is exercised through the supervisory control computer SuC to which an operator inputs setpoints and constraints via the SCI. In turn the data acquisition interface in the SuC provides measurement to the program blocks PCI, RBO and MPC and receives control signals from these blocks for transmitting corrective/operative instructions to different elements in the grinding mill circuit 10 via a DCS (distributed control systems) /PLC(programmable logic controllers) which are provided targets and acts as an interface between the controller SuC and the individual elements of the grinding mill circuit 10 as seen in Figure 1.

Model Predictive Control Module (MPC) The invention is based on model predictive control (MPC) technology, several variants of which have been patented (such as Prett et al. US Patent No: 4,349,869 (no Indian equivalent available) 1982; Lu, US patent No 5,572420 (No Indian equivalent available) 1996; Buescher et al. US Patent No: 5,659,667 (No Indian equivalent available, 1997). MPC has also been extensively researched and several developments have been published in the open literature {Garcia and Morshedi, 1986 Quadratic programming solution of dynamic matrix control (QDMC)", Garcia C. E. and Morshedi

A.

M.,Chem. Eng. Communn, 46, 73-87 (1986); Garcia et al., 1989- Garcia C.E.

SPrett D.M. Morar Model Predictive Control: theory and practice- a survey", Automatica, 23 335-348 (1989); Ricker, 1994- Ricker Model Predictive Control: state of the art", CPC IV proceedings of the 4m international conference of Chemical Process Control, Padre Island, Texas, 271-296 (1991)1. The MPC module in accordance with this invention uses multivariable state space models of the elements grinding circuit, relating controlled variables CV1 to CV6 to manipulated variables MV1 to MV4, developed based on plant data. Model adaptation to cover the entire operating region of the circuit is incorporated by identifying a large enough set of models, each valid over a subset of the operating region. The MPC algorithm uses a weighted average of all model outputs for predictive control. Model weightings are dynamically computed on-line depending on the operating point at any instant.

Rule Based Override module (RBO) a.

The RBO is the postprocessor module for MPC. Its function is to monitor •disturbance or alarming process conditions and take rapid corrective action for situations not adequately described by the process models. The RBO is an on-line expert system, populated with rules based on normal operator actions.

Process Condition Inference Module (PCI) There are several critical controlled variables in the cement grinding circuit which are not measured on-line. The fineness of the final product CV6 is the most significant of this type. The PCI module uses process data to estimate circuit product fineness as feedback for control computations in MPC. There is also a provision to update the fineness model parameters using laboratory analysis data, if the model prediction is seen to drift from lab values. In addition, the PCI module also contains several inference rules currently used by operators and engineers to determine if disturbances (change in feed coarseness, moisture content, for example) are affecting the process. This disturbance information is used in feedforward fashion by the MPC module to provide rapid regulatory action.

Figure 3 shows the hybrid control methodology in accordance with this invention, which is as follows. Process variable (PV) values for the controlled variables (CV1 to CV6) represented by block CVPV, and manipulated variables (MV1 to MV4), MVPV along with the current setpoint (SP) values for the MVs, represented by block MVSP are read in. These values are first monitored by the Abnormal Condition Trigger module ATC, which checks for excessive deviations between PV and SP for all MVs MV1 to MV4 (which would indicate that the control setpoints are not being utilized property) as also excessive rates of change for critical CVs CV1 to CV6. If there is no abnormal condition, then all the MV PV and SP are passed to a MV filter block MVF that filters the MV values to take into account the transient characteristics of the cement grinding mill circuit.

Filtered MV values and CV PVs are passed to the Blaine Estimator module BE which computes the Blaine estimates for the final ground cement product.

The Abnormal Condition Trigger ATC, the manipulated variable filter block MVF and Blaine Estimator BE comprise the Process condition inference PCI) module of Figure 2. All estimates, and PV values are then provided as process feedback block PFB to the Model Predictive Control (MPC) module.

The MPC module also takes as input the process setpoints and constraints provided by the plant operators. Normal MPC computation is carried out to yield the optimal setpoints [OSP]. If an abnormal condition is detected, an alarm AL is generated and the MPC module is bypassed. In this instance, all PV and SP values are supplied to the Rule Based Override (RBO) module which controls the cement mill circuit 10 with a rapid sampling frequency to quickly recover from the abnormal condition. Normal optimal setpoints from the MPC module are also passed through the RBO to check for constraint violations. The output from the RBO is the final MV SP value, which is downloaded to the field device lock DCS/PLC to effect control of the elements such as the air line damper 12, grinding mill elevator amps 22 and separator RPM 20, of the grinding mill circuit Benefits The benefits obtainable from the implementation of this supervisory control technology include: Rapid circuit stabilization Higher average throughput from the circuit Maintenance of desired fineness Reduction in specific energy consumption in grinding Significant reduction in variability in throughput and product fineness e Smoother operation of all important process units Novel Features Novel Features The novel feature of the invention described here is the use of a hybrid MPC technique, where a rule based override (RBO) is used to monitor and post process the control computations of the MPC module. The RBO consists of operating rules typical to dry grinding circuits used in cement plants.

Another novel feature is the process condition inference module, where a principal component analysis model is used to estimate the variations in the product fineness (size or Blaine specific surface area) based on measurements of other process variables in the circuit.

The novel features of this invention can then be summarized as follows: Hybrid MPC with expert system and inference module for control of cement mill circuits On-line product fineness estimator based on principal component analysis The core control methodology in the method in accordance with this invention is model predictive control (MPC). The controller model used in MPC is a discrete state space representation of the process fx(k) Fu(k) y(k) Cx(k) Du(k) (2) where x is the state variable vector, u is the vector of MVs, and y is the vector of CVs. k is the current instant in time.

During control computations, the predicted output at time k+ based on During control computations, the predicted output y at time k+1 based on current information is computed using: A A y(k+1/k) C<Dx(k/k-1) [CF+ D u(k)

A

y(klk-1)] (3) Once the predicted output is computed, an optimization problem is set up to compute the control moves required to achieve the desired target. The optimization step consists of finding the future set of optimal control moves that minimize the predicted error in a least-squares sense over the prediction horizon, subject to the satisfaction of the process constraints. The predicted error is computed from: e(k+j) yd(k+j) y(k+j/k) (4)

V

where yd is the desired trajectory (typically the setpoint at the current instant).

The objective function for the minimization is given by min X(Au) e T Q e AuT R Au subject to the specified constraints. Process constraints on manipulated and output variables can be written in terms of constraints on the future control moves. R and Q are weighting matrices for the control moves and predicted errors respectively. Determining the weighting matrices constitute the important tuning steps in MPC, and tuning rules for their selection have been discussed by a number of researchers (Maurath et al., 1988; Garcia et al., 1989).

To solve the above, a quadratic programming (QP) sub-problem is formulated with output and manipulated variable constraints stated as hard constraints on the decision variables: 1 min q(Au) uT G Au gTAu (6) 2 s.t. Cc T Au bc (7) OC* G is the Hessian of the objective function and g is the objective function gradient. represents the system of linear equality and inequality constraints. The constraint manifold is specified as follows.

Manipulated Variable Constraints For bound constraints on the manipulated variable, Cc T IL, where IL is a lower triangular matrix with all non-zero elements equal to For an upper bound umax, lower bound umin, and current value of manipulated variable the elements of bc are set equal to umin u(k) for lower bounds, and u(k)-umax for upper bounds. For velocity constraints, CcT I, where I is the identity matrix and the elements of bc are set equal to the magnitude of the maximum allowable control move.

Output Constraints Output constraints are imposed by requiring that the predicted output stay within the desired bounds over the entire prediction horizon, or a subset of the prediction horizon, termed the constraint window. The jth element of bc is given by .i N-1 N-1 Sbc(j) (y(kmin) (aj+i ai) u(k-i) (8) i=1 for a lower bound min on the output. a is the element of the step response vector for the CV-MV pair. The corresponding element of CcT is the jth row of the dynamic matrix A, a lower triangular Toeplitz matrix based on the step response coefficients, obtained by recording the transient response of the CV to a step change in the MV.

N-1 bc(j) Ymax) (aj+i- ai) u(k-i) (9) i=1 for an upper bound Ymax, on the output, a is the i th element of the step response vector for the CV-MV pair. The corresponding element of Cc T is the jth row of the dynamic matrix A, a lower triangular Toeplitz matrix based on the step response coefficients, obtained by recording the transient response of the CV to a step change in the MV.

The invention will now be described with reference to an actually worked example of the operation of the control system in accordance with this invention which was implemented at a 1 million TPY cement plant in India.

Some results from this implementation are described below and are graphically illustrated in Figures 4 to 10 of the accompanying drawings. Three cases are discussed.

Case 1. Normal Optimizing Control The figures 4, 5 and 6 are on-line data captured during operation of the supervisory controller SuC. The controller was brought on-line at 18:15, minutes after the cold start of the mill, and while the process was in transient.

In the graph shown in figure 4, line a shows the throughput of from start. This is much faster than the normal stabilizing time for the circuit .ooooi under manual control. In figure 5, the on-line fineness estimate is shown as Line b in the graph. The lower limit is 330 m2/kg, and simultaneous manipulation of the separator rpm Line c is seen to ensure that fineness is maintained at higher throughput. The figure 6 shows Line d the total feed setpoint requested by the controller and the actual feed rate (from belt conveyor measurement) Line e. There is significant offset in the Level 1 PID control action, and the supervisory control is able to compensate for this error as well.

Case 2. Abnormal Condition Recovery The functioning of the RBO module is illustrated here. In the figure 7, Line f 17 shows that the controller is increasing the throughput of the circuit, when there is a sudden ramp in the mill accumulation Line g (in figure 8 )at 14:05, due to a sudden increase in feed moisture. In the absence of control action to counter this, the mill would choke, resulting in significant downtime. The controller is seen to take smooth corrective action Line h, as opposed to Line i, to counter this disturbance, and resume optimizing control when the mill accumulation is normal. Under manual control the operator would typically cut feed drastically, resulting in throughput loss, and product quality variation.

During this run, the product fineness did not vary outside specified limits.

Case 3. Blaine Estimation Performance Figure 10 illustrates the performance of the on-line Blaine estimation model of the PCI module of this invention. The graph shows the actual Blaine specific surface area values of the product as obtained by laboratory analysis of samples taken over a 36-hour period (Line j and the estimated Blaine values from the PCI module at the time of sampling (Line k The close fit between these lines, both in value and trend establishes the accuracy and reliability of the Blaine estimation technique.

It will be understood that the term "comprises" or its grammatical variants as used herein is equivalent to the term "includes" and is not to be taken as excluding the presence of other elements or features.