WO2007110466A1

WO2007110466A1 - Method for estimating the ball charge of a grinding mill

Info

Publication number: WO2007110466A1
Application number: PCT/FI2007/000069
Authority: WO
Inventors: Jussi JÄRVINEN; Peter Blanz
Original assignee: Outotec Oyj.
Priority date: 2006-03-29
Filing date: 2007-03-19
Publication date: 2007-10-04
Also published as: FI20060302A0; FI20060302L

Abstract

In the present invention a method, apparatus and computer program are presented, with which the amount of balls among ore material contained in a grinding mill is estimated as a percentage by volume of the total volume of the mill. Preferably, the invention relates to semi-autogenous mills. In the invention, an expanded Kalman filter is used to estimate the ball charge such that measurements obtained from the process and process models are utilized. The invention combines the exact measurement of the fill level of the mill and some other measurement which is dependent upon the fill level and ball charge. By taking this information along into the Kalman filter algorithm an exact estimate for the ball charge can be calculated.

Description

METHOD FOR ESTIMATING THE BALL CHARGE OF A GRINDING MILL

FIELD OF THE INVENTION

The invention relates to estimating the size of the ball charge of grinding mills used in grinding of minerals .

BACKGROUND OF THE INVENTION

Different types of minerals are obtained from nature by searching for ore deposits and by enriching the ore thus obtained such that the valuable minerals are separated from the useless rock matter. In mines, the rock material is first mined from the bedrock. The ore blocks obtained from mining are guided into an enriching process, the parts of which are typically crushing, grinding, enrichment, drainage and enrichment handling. Crushing can be done, for example, using jaw, gyratory or cone crushers. The crushed rock thus obtained is guided into grinding, to which the present invention particularly relates . Grinding can be further divided into primary and secondary grinding, which can be implemented using mills tailored for these purposes . Into primary grinding are guided crushed rock and blocks, and the grinding result achieved is further guided into secondary grinding, the final result of which is a fine ore dust. The purpose of grinding is naturally to ease the next stage of the process as much as possible. The ground ore is further guided into enrichment. Enrichment can be performed, for example, as flotation, in which to the dust are added water and flotation chemicals. Each desired mineral has its own chemical, to which the desired valuable mineral attaches on the surface of the bubbles that are forming. Extraneous water is removed from the enriched product by concentrating and filtering. Mills used in grinding are typically cylindrical in shape, in which, in addition to the ore itself, are used water and the desired grinder pieces to aid in grinding. The 'grinder pieces can be rods, balls, or large ore blocks themselves can function as pieces that grind material.

In the crushing and grinding process, the average size of blocks reduces then as a function of time, until a dust-like rock material of the desired size is achieved. A portion of the ground blocks can be defined as of so-called critical size. Rock material of this size grinds quite poorly, because the particles are too big to be crushed efficiently, but, on the other hand, are also too small to efficiently crush other material. Material of critical size must often be ground separately in a crusher.

The functioning principles of grinding mills are of several different types. Mills can be divided into four different types on the basis of the use of the grinders used: autogenous mills (AG) , semi- autogenous mills (SAG), ball mills and rod mills.

In autogenous mills, the mill is fed for grinding only with the ore itself, which can be of greatly varying size. When large stone blocks and smaller pieces of crushed stone roll around in the mill and strike one another, they crush the material of one another, and thus, as the mill rotates, smaller rock material is created. When the material has been ground to an adequately small size, it is removed through the discharge opening of the mill and, on the other hand, new ore can constantly be added to the mill. The disadvantage of an autogenous mill is that it creates a relatively large amount of critical size rock matter.

In semi-autogenous mills, the mill is fed, in addition to the ore material to be ground and water, with separate balls, the purpose of which is to speed up the grinding process. In addition to the balls, the ore blocks themselves grind the ore as in an autogenous mill. The specific gravity of the balls used is typically 2-3 times greater than the average specific gravity of the rest of the ore mass. The material of the balls is typically, for example, iron. Due to these facts, the balls form a particularly large portion of the costs in ore grinding processes, so on the part of a well-functioning process it is not worthwhile to add too great an amount of balls to the mill. An additional consequence of the specific gravity is that if there are too many balls on the part of a well-functioning process, they increase unnecessarily the amount of power the mill needs from the motor. A semi-autogenous mill is typically used in primary grinding. It is characteristic of semi-autogenous mills that, due to the balls, the mill grinds above- mentioned rock matter of critical size more efficiently than autogenous mills .

A third type of mill is a so-called ball mill. In this kind of mill, the portion of the balls can even be about 40%. In this case, grinding is achieved by the balls. The input for a ball mill can be, for example, grains about 1 millimetre in size and as the result of grinding a powder-like ore is obtained. Ball mills are typically used for secondary grinding. For this reason, large blocks are not originally fed into the ball mill nor does rock matter of critical size have an opportunity to form there.

Rod mills represent a fourth type of mill. The grinding elements of the mill are rods located horizontally, wherein, as the mill rotates, the rods strike the crushed rock.

Fill level of the mill generally means the portion the portion of the material within the mill as a volume percentage of the total volume of the mill . The material is composed then of all the matter con- tained within the mill, in other words, the ore itself, added grinding pieces and water.

Fill level of the mill can, in principle, be measured in several different ways. However, in practical application problems have arisen with these methods, due to which the measurement of fill level is used quite little in the known art. One means for giving a particularly exact estimate is presented in FI 115854. The method described in this patent utilizes longitudinal lifters around the perimeter of the mill which strike the material contained within the mill to be ground. The toe of the load of the mill on the perimeter of the mill (the striking point on the longitudinal lifters) depends i.a. on the fill level and rotation speed of the mill. Due to the striking of the material against the longitudinal lifters, oscillation is directed to the power or torque required to rotate the mill. When the toe shifts, the oscillation stage changes. The power oscillation in the time domain is converted to a frequency domain by a Fourier transformation. Additionally, in the method of the patent, compensation is made for variations in the rotation speed of the mill . The meeting angle of the load of the mill is defined and additionally, by utilizing the information regarding rotation speed, the fill level can be calculated, for example, using some suitable mathematical model. Thus, as the result of FI 115854 an exactly measured mill fill level is obtained by utilizing the power information received from the motor.

To rotate the mill power is required from the motor. The power required is to a great extent defined by the mill and the parameters of the material within it, such as fill level and ball charge. It is logical that the more material rolled around within the mill, the more power is required from the motor. In mills of the known art, the measurement of power is a standard measurement .

Additionally, in the known art, the total mass of the mill is measured either using a scale and/or by monitoring the oil pressure of the bearings . Weighing with a scale can be implemented such that the mill and its contents cause a compression of a given size which can be measured.

One essential function-related parameter of semi-autogenous mills is the portion of the balls within the mill as a volume percentage of the total volume of the mill . The same parameter can also be called the ball charge of the mill. Balls can be added to the grinding process either at desired time intervals a larger amount at a time or constantly adding a desired number per unit of time. In one existing relatively small-sized mill example, the size relationships are divided such that the diameter of the mill is 3 metres and the diameter of the balls to be added is about 10 cm. From this type of mill is obtained as the milling result ore matter going through at a speed of about 140 t/h and the size class of the amount of balls to be added can be about one ton per day. On the other hand, the diameter of a large mill can be in the size class of 10-12 metres. In semi-autogenous mills of the known art, the size of the ball charge is typically in the size class of 6-18 percent. For semi- autogenous mills it can be concluded, when using a large ball charge, that the masses of ore and balls are of nearly the same size class.

The ball charge amount in a mill is important information, which in the known art has not, however, been possible to accurately measure or estimate. The problem is created in that to make a visual observation of the contents of the mill, the mill must first be stopped. In practise, during maintenance breaks or shut-downs of the mill, a visual estimate can be made on the basis of the visible surface area of the mill contents, and the percentage portion of ore and the percentage portion of balls of the contents can be estimated. After this, an estimate can be made based on previous experience of the suitable addition of balls such that the grinding process would function optimally. This type of estimation can be based even on that the capacity of the mill has been monitored, in other words, the amount of ground mass obtained from the mill per unit of time as a function of the different amounts of balls added and by selecting such an addition amount of balls that gives the greatest capacity. This manner is then empirical and requires experience on the part of the persons using the mill. Additionally, the method must be repeated separately for each mill, as the result for one mill cannot be generalized for other mills or for other types of ore to be handled.

Another manner of estimating the size of the ball charge in the mill has been to monitor the rate of wear of the balls in the mill in relation to time or tons of ore processed. However, the rate of wear varies according to i.a. ore hardness, grain size, the amount of ore in the mill and the amount of water.

On the other hand, in principle, the amount of material going through the mill could be monitored and an estimate obtained of the amount of balls required per ton processed. In the known art, it has, however, been difficult to monitor the composition of the ore mass ground by the mill. The total mass coming from the mill in a given unit of time could be monitored using a suitable weighing solution, but because in the material both ore and balls are ground, it is impossible to conclude from the mass the portion of balls of the total drain.

In the known art, the size of the ball charge is also estimated on the basis of experimentally ob- tained curves. Curves have been created based on empirical observations, wherein on the x-axis can be the size of the ball charge, on the y-axis the power intake of the mill and the variable parameter of the curves of the curves set is the fill level of the mill. However, the accuracy of this method is not the best possible, because the method is based on approximate inspection of curves obtained empirically. Additionally, the set of curves always relates to a specific ore and mill.

In the known art, is also presented a calculation principle known by the name ^λSoft Sensor' . Soft Sensor is a software-based way of handling information obtainable from different processes such that as an good estimate as possible can be estimated for a desired unknown variable. In this connection, the actual process can be measured concretely using so-called hard sensors . Models of the process and measurements can be created which together with actual measurement results are guided into the estimator. The estimator can conclude the output signal of the Soft Sensor, for example, using the smallest square sum method, wherein the state estimate sought is figured out.

The disadvantage of the known art is that the ball charge contained within the grinding mill cannot easily be figured out or measured by any sensor. Another disadvantage is that the methods that require stopping the mill are impractical and expensive, as stopping the mill interrupts the entire enrichment process .

PURPOSE OF THE INVENTION

The purpose of the invention is to present a new type and more accurate manner of defining the size of the ball charge of a grinding mill without stopping the grinding mill . SUMMARY OF THE INVENTION

The present invention presents a method for estimating the ball charge in an ore enrichment process that uses a mill that grinds the ore and contains at least ore and balls. A motor functions as the power source that rotates the mill. According to a method according to the invention, the fill level of the mill is measured and additionally is measured at least one other value which are dependent upon the ball charge and the fill level of the mill. Characteristic of the method is that in it the ball charge estimate is calculated using a Kalman filter using as its measurements the fill level and at least one said value.

In one embodiment of the present invention, the ball charge estimate is calculated from the Kalman filter equations:

_and

_{in which} x _{is state,} u _{is input}, J is measurement, k is discrete time, * and h are functions and both ^w and ^v are noise terms .

In one embodiment of the present invention, Yi = Xi is set as a correlation between the Kalman filter fill level measurement yi and fill level state xi.

In one embodiment of the present invention, the power obtained from the motor is defined as one value to be measured yi and the estimated power ^i is calculated by Kalman filter states x using the model of Morrell.

In one embodiment of the present invention, the total mass of the mill is defined as one value to be measured yi and the estimated total mass ^i is calculated by Kalman filter states x using the model of Morrell.

In one embodiment of the present invention, the oil pressure of the bearings of the mill is defined as one value to be measured yi and the estimated oil pressure of the bearings ^± is calculated by KaI- man filter states x using the model of Morrell.

In one embodiment of the present invention, the fill level of the mill xi and the ball charge of the mill X₂ are used as Kalman filter states.

In one embodiment of the present invention, to the Kalman filter states xi is added at least one state from the group including grinding rate of the ore, wear rate of the balls and amount of water.

In one embodiment of the present invention, the Kalman filter is set as * ^~ *^■ * , when ^{M =}0. when the inputs are taken into consideration (^{u ≠}®), a more complicated function can be used which can be generally described f = f{x\^k),u{k),k) _{^ A more} complicated function / can also be used in a situation where as states are used others than the fill level and ball charge .

In one embodiment of the present invention, as the Kalman filter input ^u is used at least one input from the group including addition of ore, addition of balls and addition of water. In a simplified embodiment of the invention, the inputs are not taken into consideration (^u=®).

In one embodiment of the present invention, in the Kalman filter the variances of the noise terms ^w , ^v relating to the measurements ^ and states ^x are varied on the basis of the estimation errors ^-y,--

In one embodiment of the present invention, balls are added to the mill a desired amount per unit of time on the basis of the ball charge estimate.

The inventive idea of the present invention also comprises the apparatus corresponding to a method for estimating a ball charge in an ore enrichment process . The apparatus comprises a mill for grinding ore and containing at least ore and balls. Additionally, the apparatus includes a motor that rotates the mill. Further, first measurement means are required which are arranged to measure the fill level of the mill as well as second measurement means which are arranged to measure at least one other value which are dependent upon the fill level and ball charge of the mill. Characteristic of the apparatus is that it comprises a processor which is arranged to calculate a ball charge estimate using a Kalman filter which uses as its measurements the fill level and at least one said value.

The apparatus is arranged to perform the steps above described in connection with the method such that the first and second measurement means perform the apparatus-related measurements, and the calculation relating to the Kalman filter is performed in the processor. In one embodiment of the invention, means for adding balls additionally belongs to the apparatus .

The inventive idea of the present invention also comprises a computer program such that the program code is arranged to perform the above-mentioned steps of the method.

The advantage of the invention is that a good estimate is obtained of the amount of balls contained in the grinding mill. This information can be utilized such that required additions in the amount of balls can be guided more exactly and more efficiently on the parts of both the process and costs. Using this method, the capacity of the mill is then increased and, at the same time, the costs are nonetheless kept as optimal .

The advantages of the invention principally result from the fact that in the method, the state of the mill is measured directly, in other words, the principle of the Kalman filter is connected to the exact measurement result of the fill level which is obtained directly as a corresponding state information. This enables calculation of an exact estimate for the size of the ball charge as well.

LIST OF FIGURES

Fig. 1 shows a functional principle of an expanded Kalman filter according to the known art,

Fig. 2 shows a preferred embodiment of a method according to the invention for the calculation of the ball charge of a grinding mill, and

Figs. 3a and 3b show practical simulation results utilizing a Kalman filter according to the invention.

DETAILED DESCRIPTION OF THE INVENTION

For the characteristic features of the present invention reference is made to the claims .

In the invention, to the measurement results obtained from the mill is attached the idea of an estimation of ball charge done using an expanded Kalman filter. The Kalman filter is a per se known art, but using it in estimating the ball charge of a grinding mill requires an exact estimate of the fill level of the mill and exact process models, and nowhere in the known art is a Kalman filter presented in this connection. The Kalman filter, in principle, gives more stabile estimates for the desired state value, although the values to be measured appear in a distribution due to different types of imperfections (which can generally be called "noise") . The process relating to use of an expanded Kalman filter can be described in general form as follows :

y{k)=h(x{k),k)+v{k)

where vector x represents the state of the process, vector u input, vector y measurement, k is discrete time, / and h are non-linear (can also be linear) functions and both w and v represent noise and other unknown factors .

The function / describes then the formation of the next state and it is affected by the previous state, the inputs entering to the process and, on the other hand, also the moment in time k. The function h , on the other hand, describes the state of the process at a given moment in time and the correlation between the measurement results. The noise term w^'• expresses the unknown portion of the formation of state, which is not modelled by the function / . The term v describes the noise related to measurement. The co- variance matrices of the noise terms are presumed known.

By using a Kalman filter, the internal state of the grinding process is then estimated. The Kalman filter algorithm can be described using a signal diagram which is presented in Fig. 1. Into the block 11, which calculates the a-priori-estimate x (k/k-1) , are guided the process input u(k-l) and the a-posteriori- estimate of the previous moment in time x (k-ljk-1) . This anticipating estimate x (k/k-l) is fed into the measuring function h , wherein is obtained the measurement estimate y (k) . This is compared to the actual measurement y(k) received from the process 10, wherein the error term is obtained which tells the deviation of the estimate from the actual measurement of the process. The measurement y(k) contains noise, consequently the error term also contains noise. The a- posteriori-estimate 12 x (k/k) corrects a-priori- estimate 11 by emphasizing the difference in the amplification matrix K (not in the figure, is contained within the correction block 12) . To calculate the amplification matrix the filter estimates the covari- ances of the errors of the a-priori- and a-posteriori- estimates. The covariance of the A-priori-estimation error can be written using P(kjk-l) and the a- posteriori-estimation error using P(kjk) . For the calculation of K and P, a linearized state model of matrices A and C is needed:

x(k+l)=Ax{k)+Bu{k) (2) y{k)=Cx{k)

To calculate these matrices, i.e. to linearize the non-linear functions / and h , a so-called

Jacobian matrix must be calculated as follows. The matrix A is formed by partial derivation of the components of function / in relation to each state variable. Correspondingly, the matrix C is formed by partial derivation of the components of function h in relation to the state variables. In other words, if there are two elements in the state vector, then / is also two-rowed and the linearized matrix A or the Jacobian matrix is in size a 2*2 matrix and it is obtained by:

The matrix C is obtained in a corresponding manner. As an example it can be presumed that the measurement y contains two components, and consequently the function relating to the measurement h contains two rows. In this case, C is then in size a 2*2 matrix and it is obtained by:

Bh₁ Bh₁

C = Bx₁ dx₂ (4 ) dh₂ Bh₂

Bx₁ dx₂ In other words, the functions / and h are linearized at a given function point 3c . If / is nonlinear, the matrices will have terms that contain state variables (xχ_t x∑, •••), components of the input vector u and/or the time variable k. For this reason, the matrices A and C must be calculated again on each round by placing the current known values x , u and k into equations (3) and (4) and thus the number values for the elements of the matrices A and C are obtained. When calculating matrix A, the a-posteriori-estimate x (k-l/k-1) of the previous round is used as a state vector x , whereas when calculating matrix C, the a- priori-estimate x (k/k-1) of the current round is used. The number values of the elements of the matrices K and P can further be calculated from A and C.

In a situation according to the present invention, in which the size of the ball charge located within the mill is estimated, the Kalman filter can in the preferred embodiment of the invention be used such that the states and measurements are selected as follows. As the states x the fill level of the mill and the ball charge of the mill can be selected, or expressed mathematically:

x = (5) xb>all_charge

Other states could also be chosen for the vector x . Such could be, for example, the wear rate of the balls, the grinding properties of the ore and the amount of water.

In the preferred embodiment of the invention, for the vector y or the measurement vector can be selected the measurement of fill level and at least one additional measurement which changes as a function of fill level and ball charge. One alternative for the latter measurement is a measurement of the power obtained from the motor. Mathematically y can be expressed as follows:

yfilljevel y = (6) ypower

More measurements can also be taken into the vector as elements yβ, Y_A, etc. One example is a measurement of the oil pressure of the bearings, with which the examination of the total mass is brought along into the calculation. The total mass can also be measured directly by means of weighing cells.

In the vector u can be included different types of process inputs, such as ore input, water input, and/or input of balls. In an advantageous example of the invention, inputs are not taken into consideration for purposes of simplification, i.e. u is a null vector.

Because the function h in this case connects together the measured power required from the motor {y₂ in the above-mentioned notation) and the state vector x, the correlation between the power required and the amount of material contained within the mill must be clarified, preferably as exactly as possible.

Known art publication S. Morrell: "Power draw of wet tumbling mills and its relationship to charge dynamics - Part 1: a continuum approach to mathematical modelling of mill power draw" (Transactions of the Institution of Mining and Metallurgy, Section C, 105, January-April 1996) describes a power model, in which the state information about the ball charge and fill level in addition to other parameters are included in the measurement of power. The original idea of Morrell has been to define the motor required by the mill (for example the available maximum power) when the mill (for example the diameter of it) and parameters relat- ing to the state of it are known. In the present invention, the information of Morrell is used to calculate the power at a given state of the mill x . The idea of Morrell is to calculate the total power required from the motor as follows :

* total ⁼ K-I* no-load ~^ for charge-motion ( ' )

Where the 'k's represent so-called calibration multipliers which can be considered to contain heat losses, rotation of the material to be ground and scraping against one another, which expends energy, and, on the other hand, the placing of the material inside the mill. The power model of Morrell is particularly complicated, but the essential thing on the part of this invention is that from it the function h can be figured out on the part of the measurement of power. In other words, the actual fill level and ball charge at the desired moment in time can be attached to the power to be obtained from the motor and to be measured.

Characteristic of the invention is that relating to the fill level the function h = 1, or in other words, because an exact measurement of fill level is obtained in the manner presented in FI 115854, from it is obtained directly the state xi itself. In this manner, more exact results are obtained from the Kalman filter, when measurements and the models relating to them are exact.

In the preferred embodiment of the present invention as f is selected:

which means that the next state x_k+i is obtained by: ^Xflll_level ^~ ^W fill Jevel ^xk+ι ~ (9) ^Xball _ ch arg e ^"*^" ^ ball _ ch arg e _

The change in state is then seen to be occurring only under the influence of noise in the example in question.

If it is desired to take into consideration additional states, inputs and/or time, f can be selected as the desired function from the states, inputs and time:

f = f{x,u,Jc) (10)

In another example, the ore tons added to the mill can be included in the state vector and, on the other hand, the ore tons ground in the mill can also be taken into account. In this case, the grinding rate can be included in the state vector as a new state.

The present invention is preferably applicable to semi-autogenous mills, whose process is relatively complex and also less stabile in comparison to ball mills. The method, however, works for ball mills as well. Relating to ball mills, the estimation of ball charge can be done even more exactly, because a great portion of the mass of the mill contents is made up of balls. In this case, from the power taken by the mill an estimate of the amount of balls can be obtained exceptionally well even directly. Using a method according to the invention, this estimate can be refined further yet .

The preferred embodiment of the invention can be clarified on the basis of that presented above into the flowchart presented in the following. Fig. 2 shows the essential components of a preferred example of this invention. The essential parts of the invention are different measurements and models of a process in a dynamic state, and parameters that otherwise relate to the process. These are represented in Fig. 2 by blocks 20-24, 26 and 27. According to the principle of patent publication FI 115854, the defined exact estimate 20 for the fill level of the mill is obtained, which is essential for the functionality of a method according to the invention. The mills have a measurement of power to be obtained from the motor 21 implemented as a standard measurement. Power is defined i.a. from the measurements of the mill, construction, rotation speed and the quality and amount of material to be ground contained within the mill. Additionally, if there is a desire to consider the input material added to the mill, they must be considered as parameters 22. The input material can be the ore mass itself, balls and/or water.

The total mass of the mill 23 is a useful measure and it can be implemented using a solution like a scale, with which the pressure created by the mill and its contents can be measured. One alternative to measurement of total mass is to measure the oil pressure predominant in the bearings of the mill, which is comparable to mass. Other states, measurements and inputs can, in a method according to the invention, be freely added 26, 27, 29 as input parameters. However, it must be noted that the more states the Kalman filter has to handle, the more easily the calculation process disperses and the desired size of the ball charge is not estimated well . On the other hand, if the method has exact additional measurements in use, the ball charge estimate can be refined in this manner.

Additionally, the uncertainties relating to the process, the physical portions that are a part of it and measurements can be represented as noise terms 24 which are also visible in the calculation formulas of the Kalman filter.

In the invention, to the exact estimate 20 of the fill level of the mill is then added other measurements obtainable from the mill and these are added to the model describing the dynamic grinding process in an expanded Kalman filter 25, in which an exact estimate of ball charge is obtained as a result of recursive calculation. According to the Kalman filter principle described earlier, the algorithm of the filter comprises a first function, according to which the next state is concluded from the first state as a function of time 25a, and additionally a second function, which adds together the actual state and the measurement at a given moment in time 25b. In other words, these functions are in Fig. 2 named as the process logic between the states 25a and, on the other hand, as the logic of correlation between the measurements and states 25b. Above-mentioned functions are correspondingly f and h which are of the form of vectors. The functions can be non-linear or linear, but a function describing a practical process is almost always non-linear. By using the fundamental principle of Fig. 1 and the measurements and states of Fig. 2, the recursive calculation of the Kalman filter ultimately gives an estimate of the amount of balls 28 as a volume percent of the total volume of the mill .

The principle of an expanded Kalman filter according to the present invention applied to defining ball charge is particularly well suited for computerized seeking for optimal state estimates. The filter functions on a recursive principle, wherein measurement results can be handled in accordance with when information is obtained from the process, and the measurement history does not then necessarily need to be stored. In a preferred embodiment, the calculation is performed by the control logic, in which the steps of the method are, in practise, implemented as a computer program.

One example of a practical test is presented in the following with simultaneous reference to Figs. 3a and 3b. Normally, at the enrichment plant of the example, balls are added to the mill as a lot of one ton per day. In the ball test of the example, balls were not added during the time period 15.1. - 18.1., but after this balls were added in two lots such that on both the afternoon of 19.1 and the morning of 20.1 balls were added to the mill. After the additions, the ball deficiency that had earlier formed in the mill was fixed by an addition of a total of six tons of balls (during the time period 15.1. - 20.1.).

In the Kalman filter of the invention, the selection of noise parameters for different values is emphasized, i.e. the defining of dispersion or variance of noise (vectors w and v above) for different states and measurements.

In the first simulation, the power model relating to the Kalman filter was calibrated to 5000 data points and using 12 % as the value of the ball charge. Variance (the square of standard deviation) was set as 10000 for measurement of power, the variance 1 for the fill level and the variance 0,1 for the state of fill level. When variance was set as 10^"6 for the state of the ball charge, a graph according to Fig. 3a was obtained for the estimate of ball charge.

From Fig. 3a it can be concluded that the estimate behaves correctly. The estimated size of the ball charge declines over the time period 16.1. 19.1. and the ball charge ultimately turns to an incline on 19.1. This coincides well with the actual time of adding balls of the ball test.

In another example, the variance of the ball charge was increased to the value 10^"5. While the other parameters remained the same as in the preceding situation, a somewhat different type of graph for the ball charge was obtained, which is shown in Fig. 3b. From the graph it can be seen that the change in variance influences the functioning of the Kalman filter or the variance of the state of the ball charge defines the rate of change of the ball charge estimate seen in the figures. In other words, the larger a variance is selected, the more radically the graph behaves and describes the addition of balls at this time better. The ball charge variance to be used can be estimated, for example, according to the wear rate of the balls, in order that the estimate of the filter is made as exact as possible. In the example, it was also noted that the variance of the power measurement should not be set very low, as in that case the Kalman filter compensates local variation in power by changing the ball charge. The same phenomenon is noted from Fig. 3b at date 16.1, wherein the power drops precipitously in comparison to the fill level probably due to a reduction in block input occurred at this time. At this time, the modelled value of power is so far above the actual measured power that the filter compensates the difference using a substantial reduction in the ball charge estimate. As is seen from the graph, as the situation normalizes (over time and as the input of blocks normalizes) the ball charge estimate returns to "the more proper level". On the basis of Figs. 3a and 3b, the filter seems to function in the manner expected giving a particularly good ball charge estimate.

In one embodiment of the invention, the above-mentioned variance of ball charge state can be varied. In this case, the residual signal y(k)—y(k), which is in the signal diagram of Fig. 1 one of the inputs going into the correction block 12, can be monitored. The residual signal represents then estimation error of the measurement. For residual signal can also be made, for example, statistical analysis and on the basis of the analysis it can be concluded whether the addition of balls has occurred, wherein variance can be increased. Using a varying variance, the ball charge estimate would be found yet more accurately than in Figs. 3a and 3b. In this case, the wavelike behavior of the ball charge estimate seen in Figs. 3a and 3b could be decreased and the estimate of ball amount would decrease quite evenly over the time period 16.1. - 19.1, afterwards rising in two step-like jumps on 19.1. and 20.1. as described above.

The invention is not limited to only the embodiment examples presented above, rather many variations are possible within the scope of the inventive idea defined by the claims .

Claims

1. A method for estimating the ball charge in an ore enrichment process, in which is used a mill that grinds ore containing at least ore and balls, which mill is being rotated by a motor, and which method contains the steps of: measuring the fill level of the mill; measuring at least one other value which are dependent upon the fill level and ball charge of the mill; charac t eri z ed in that the method further comprises the step of: calculating an estimate for the ball charge using a Kalman filter which uses as its measurements the fill level and at least one said value.

2. A method according to claim 1, charac teri z ed in that the method further comprises the step of: calculating an estimate for the ball charge from the Kalman filter equations: x(k+l) = f(x(k),u(k),k)+w(k) and y{k)= h(x(k),k)+v(k) , in which x is state, u is input, y is measurement, k is discrete time, / and h are functions and both w and v are noise terms .

3. A method according to claim 2 , charac teri z ed in that the method further comprises the step of: setting y_x = X₁ as a correlation between the Kalman filter fill level measurement y_% and fill level state

X₁.

4. A method according to. claim 2 , charac teri z ed in that the method further comprises the steps of: defining the power obtained from the motor as one value to be measured y±; and calculating the estimated power y± by Kalman filter states x using the model of Morrell.

5. A method according to claim 2 , characteri z ed in that the method further comprises the steps of: defining the total mass of the mill as one value to be measured y±; and calculating the estimated total mass Ji by Kalman filter states x using the model of Morrell.

6. A method according to claim 2 , charac teri z ed in that the method further comprises the steps of: defining the oil pressure of the bearings of the mill as one value to be measured yi; and calculating the estimated oil pressure of the bearing y± by Kalman filter states x using the model of

Morrell .

7. A method according to claim 2 , characteri z ed in that the method further comprises the step of: using as the states of the Kalman filter the fill level of the mill x± and the ball charge of the mill x₂.

8. A method according to claim 7 , char acteri z ed in that the method further comprises the step of: adding to the states x± of the Kalman filter at least one state from the group including grinding rate of the ore, wear rate of the balls and amount of water.

9. A method according to claim 7 , char ac teri zed in that the method further comprises the step of: setting in the Kalman filter / = [l l]^r , when u-0; or / = f(x(k),u(k),k) , when u ≠ 0.

10. A method according to claim 8, characteri zed in that the method further comprises the step of: setting in the Kalman filter f

.

11. A method according to claim 2, characteri z ed in that the method further comprises the step of: using as the Kalman filter input u at least one input from the group including addition of ore, addition of balls and addition of water.

12. A method according to claim 11, characteri z ed in that the method further comprises the step of: using as the Kalman filter input «^" = 0.

13. A method according to claim 2 , characteri z ed in that the method further comprises the step of: varying in the Kalman filter the variances of the noise terms ^w , ^v relating to the measurements ^ and states ^x on the basis of the measurement estimation errors y± - y ι.

14. A method according to claim 1, charact eri z ed in that the method further comprises the step of: adding balls to the mill a desired amount per unit of time on the basis of the ball charge estimate.

15. An apparatus for estimating the ball charge in an ore enrichment process, which apparatus comprises : a mill that grinds ore and contains at least ore and balls; a motor that rotates the mill; first measurement means which are arranged to measure the fill level of the mill; second measurement means which are arranged to measure at least one other value which are dependent upon the fill level and ball charge of the mill; characteri z ed in that the apparatus further comprises: a processor which is arranged to calculate the ball charge estimate using a Kalman filter which uses as its measurements the fill level and at least one said value.

16. An apparatus according to claim 15, characteri z ed in that the apparatus further comprises : said processor which is arranged to calculate the ball charge estimate from the Kalman filter equations: x(k+l)= f(x(k),u(k),k)+w(k) and y(k)= h(x(k),k)+v(k) , in which ^x is state, ^u is input, ^ is measurement, k is discrete time, * and ^ are functions and both ^w and ^v are noise terms .

17. An apparatus according to claim 16, charac teri z ed in that the apparatus further comprises : said processor which is arranged to set yx = Xx as a correlation between the Kalman filter fill level measurement V₁ and fill level state Xj_..

18. An apparatus according to claim 16, charac teri z ed in that the apparatus further comprises : said second measurement means which are arranged to measure the power obtained from the motor as one value to be measured yi; and said processor which is arranged to calculate the estimated power -^i by Kalman filter states x using the model of Morrell .

19. An apparatus according to claim 16, charac teri zed in that the apparatus further comprises : said second measurement means which are arranged to measure the total mass of the mill as one value to be measured y±; and said processor which is arranged to calculate the estimated total mass ^i by Kalman filter states x using the model of Morrell.

20. An apparatus according to claim 16, charac teri z ed in that the apparatus further comprises: said second measurement means which are arranged to measure the oil pressure of the bearings of the mill as one value to be measured y±; and said processor which is arranged to calculate the estimated oil pressure of the bearings ^i by Kalman filter states x using the model of Morrell.

21. An apparatus according to claim 16, charac teri z ed in that the apparatus further comprises : said processor which is arranged to set as the Kalman filter states the fill level of the mill X₁ and ball charge of the mill x₂.

22. An apparatus according to claim 21, charac teri z ed in that the apparatus further comprises : said processor which is arranged to add to Kalman filter states x± at least one state from the group including grinding rate of the ore, wear rate of the balls and amount of water.

23. An apparatus according to claim 21, charac teri z ed in that the apparatus further comprises : said processor which is arranged to set in the Kalman filter / = [l l]^r , when w^" = 0; or / = f(x(k),u(k),k) , when u≠O .

24. An apparatus according to claim 22, charac t eri z ed in that the apparatus further comprises : said processor which is arranged to set in the Kalman filter f = f(x(k),ϊϊ(k),k) .

25. An apparatus according to claim 16, charac teri z ed in that the apparatus further comprises : said processor which is arranged to use as Kalman filter input u at least one input from the group including addition of ore, addition of balls and addition of water.

26. An apparatus according to claim 25, characteri z ed in that said processor is arranged to use as the Kalman filter input M=O.

27. An apparatus according to claim 16, characteri z ed in that the apparatus further comprises : said processor which is arranged to vary in the Kalman filter the variances of the noise terms ^w , ^v relating to the measurements ^ and states ^x on the basis of the measurement estimation errors y± - y±.

28. An apparatus according to claim 15, characteri zed in that the apparatus further comprises : means for adding balls which are arranged to add balls to the mill a desired amount per unit of time on the basis of the ball charge estimate.

29. A computer program for estimating the ball charge in an ore enrichment process, in which is used a mill that grinds ore and contains at least ore and balls, which mill is being rotated by a motor, and which computer program comprises a program code which, when run in a data processing device, is arranged to perform the following steps of: measuring the fill level of the mill; measuring at least one other value^" which are dependent upon the fill level and ball charge of the mill; characteri zed in that the program code, when run in a data processing device, is further arranged to perform the step of: calculating an estimate for the ball charge using a Kalman filter which uses as its measurements the fill level and at least one said value.

30. A computer program according to claim 29, characteri z ed in that the program code, when run in a data processing device, is arranged to perform the step of: calculating an estimate for the ball charge from the Kalman filter equations: x(k+l)= f(x(k),u(k),k)+w(k) and y(k)= h(x(k),k)+v(k) , in which x is state, u is input, y is measurement, k is discrete time, / and h are functions and both w and v are noise terms.

31. A computer program according to claim 30, charac teri z ed in that the program code, when run in a data processing device, is further arranged to perform the step of: setting y₁ = X₁ as a correlation between the Kalman filter fill level measurement y_% and fill level state

X₁.

32. A computer program according to claim 30, charac teri z ed in that the program code, when run in a data processing device, is arranged to perform the steps of: defining the power obtained from the motor as one value to be measured yι; and calculating the estimated power y ± by Kalman filter states x using the model of Morrell.

33. A computer program according to claim 30, charac t eri z ed in that the program code, when run in a data processing device, is further arranged to perform the steps of: defining the total mass of the mill as one value to be measured yu and calculating the estimated total mass y± by Kalman filter states x using the model of Morrell.

34. A computer program according to claim 30, charac t eri z ed in that the program code, when run in a data processing device, is further arranged to perform the steps of: defining the oil pressure of the bearings of the mill as one value to be measured Vj_.; and calculating the estimated oil pressure of the bearing y± by Kalman filter states x using the model of

Morrell.

35. A computer program according to claim 30, characteri z ed in that the program code, when run in a data processing device, is further arranged to perform the step of: using as the states of the Kalman filter the fill level of the mill X₁ and the ball charge of the mill x∑.

36. A computer program according to claim 35, charac t eri z ed in that the program code, when run in a data processing device, is further arranged to perform the step of: adding to the states x± of the Kalman filter at least one state from the group including grinding rate of the ore, wear rate of the balls and amount of water.

37. A computer program according to claim 35, charac teri zed in that the program code, when run in a data processing device, is further arranged to perform the step of: setting in the Kalman filter / = [l l]^τ , when u=0; or / = f(x(k),ϊϊ(k),k) , when ϊϊ≠O .

38. A computer program according to claim 36, charac teri z ed in that the program code, when run in a data processing device, is further arranged to perform the step of: setting in the Kalman filter f = f(x(k),u(k),k) .

39. A computer program according to claim 30, charac t eri z ed in that the program code, when run in a data processing device, is further arranged to perform the step of: using as the Kalman filter input u at least one input from the group including addition of ore, addition of balls and addition of water.

40. A computer program according to claim 30, characteri zed in that the program code, when run in a data processing device, is further arranged to perform the step of: using as the Kalman filter input α=0.

41. A computer program according to claim 30, characteri z ed in that the program code, when run in a data processing device, is further arranged to perform the step of: varying in the Kalman filter the variances of the noise terms ^w , ^v relating to the measurements ^ and states * on the basis of the measurement estimation errors y± - y ±.

42. A computer program according to claim 29, charac teri z ed in that the program code, when run in a data processing device, is further arranged to perform the step of: adding balls to the mill a desired amount per unit of time on the basis of the ball charge estimate.