The construction of a refiner is well known. The grinding of wood in the re- finer is normally done between two discs of which the first one (the rotor) rotates while the second (the stator) stays still. It is also possible that both the discs are rotated. A refining zone is formed between the two discs. So- called refining segments are fastened on the walls against each other of the refining discs. Normally the refiner is fed with wood chips near the midpoint of the refining discs. The wood is transported between the discs from the near midpoint to the periphery of the discs, while the wood is fibrillated at the same time. Ready refining pulp is taken from the periphery of the discs. The split between the refiner discs where the wood is refined and transported is called refining split. The distance between the discs i. e. the width of the re- fining split is normally adjustable.

The refining process has a long time been fairly unknown. First in 1985 knowledge on its base phenomenon was gained by filming the events in the refining split with a special camera. The chips turn into fibres immediately after their feeding in the refining split. The fibrillation takes place just before the outer periphery of the refining segments. The characteristics of the pulp, for example tear rises as a function of the distance from the midpoint of the refining discs.

Production of thermo mechanical pulp (TMP) has been thought to be a very stochastic process. Reliable measurements are difficult to carry out and the control of the process has up to now been more like art than science.

A refining process has active variables, which can be changed, and passive variables, which have their optimal values, which always are kept constant.

According to traditional control philosophies the active variables, which af- fect the refining, are, for example - production or mass stream wood through the refiner discs, - effect of the motor which rotate the rotor - refiner split width - refining consistence, and - effect distribution between separate refining steps, if the refining proc- ess has multiple steps.

The specific energy requirement SER (calculated as total refining effect in MW divided with the production in ton/h) has for a long time been an impor- tant control variable, because the energy consumption is relative high in comparison with the grinding process.

Passive variables are for example : -heating time of the chips, - pressure difference over the refiner and, - the quality of the chips.

In the control of a refining process there is measured how much a change of the value of a control variable affect a quality variable. After measuring of the change in the quality variable the control variable is adjusted. Several con- nections between separate control variables and quality criteria have been suggested.

According to Manner and Ryti/1/the production and SER are the most im- portant control variables. The quality criterion is freeness (CFS).

Bolin and Hill/2/assume that SER and the refining consistence are the best control variables. SER affects at the same time the long fibre fraction in the

pulp and the peak content at a constant SER. The refining consistence is controlled with mass-and energy balances. The quality criteria are CFS, long fibre fraction and peak content. CFS is measured on-line with a com- mercial meter. To reduce the quality variations by half, there has to be done at least three control steps in an hour.

Dumont, Legault and Rogers/3/use SER as control variable and it is fed back with CFS, which is measured on-line. The refiner effect is controlled with self-adjusting controller and the refining consistence has been forward connected with mass-and energy balances from the refining process. These balances are often very complicated.

Alsip/4/has similar strategy with the aforementioned one, but control of the refiner effects is done in a different way.

According to Johansson, Karlsson and Jung/5/it is not possible to use SER as control variable, because the refining split at constant SER can be re- duced so much that the strength properties for the pulp are considerable re- duced. According to them the refining split has to be used as control vari- able. As quality variables are used CFS and fibre distribution, which are measured optically. The refining consistence is calculated from mass-and energy balances and is controlled with forward coupling.

Dahlqvist and Ferrari/6/keep the refining effect and refining split constant and the production is controlled on the base of online measurements of the density of the chips. CFS is calculated after a model, which is updated by laboratory measurements.

With present methods can the refining production be controlled, and certain optimising can be used. The methods today do not give satisfying optimisa- tion result with mass quality and the refiner energy consumption.

The aim of this invention is to get a method for controlling the refining proc- ess where the aforementioned problems have been minimized.

A special aim for this invention is to accomplish a method for controlling the refining process with which an optimal pulp quality is reached, i. e. in such a way that an optimal value is reached for characteristics of the pulp quality.

One aim is also to optimise the energy consumption for the refining process.

These aims are reached with a method and features, which are shown in the characterising parts of the following patent claims.

From the above it is clear that it is known that the refining process can be controlled with the mass flow as control variable. In practice the speed of revolutions of the feeding screw, which feeds the wood to the refiner, is used as control variable. The control variables affect the quality i. e. measuring variables : CFS and possible fibre distribution, tear and tensile, to some ex- tent because the impact of the control variables on the measuring variables can be coupled.

No one has shown, which surprising effect for the optimisation, and with that for the pulp quality and energy savings, is accomplished in a refiner, when it is controlled according to this invention with so called multivariable control simultaneously with two or more of the following control variables : a) mass flow of wood or the speed of revolutions of the feeding screw, b) width of the refining split, c) speed of revolutions of the rotor or its peripheral speed, d) rotation force of the motor which rotates the rotor.

According to the present invention it is consequently for example possible to control with the new multivariable method CFS and fibre distribution with the

mass flow at the same time when the fibre strength properties (tear, tensile) are controlled with the refiner split. In that way the refiner is controlled with two input signals and two output signals.

A third degree of freedom can be used in the multivariable controlling. As such can function the rotation speed of the rotor or the rotation force of the motor, which rotates the rotor. With these it is possible also to control the quality (for example tear). By varying the rotation speed or rotation force of the motor it is also possible to better optimise the energy requirement (SER).

In other words it is possible to reduce the energy consumption if so desired.

With the present invention all combinations between a), b), c) and d) are possible. Of course also other variables can be used together with these.

Some examples of good couplings between control variables and measuring variables are: - mass flow to CFS, -mass flow to tear or some other quality variable for example fibre length distribution or fibre length, -refiner split to tear or some other quality variable for example fibre length distribution or fibre length, - the speed of revolutions of the rotor to tear or SER - rotation force of the motor to tear or SER.

With the inventive method it is possible to increase the production (or opti- mise it), and strength and many other properties can be changed for better.

Also a possibility to energy savings is given because the speed of revolu- tions of the rotor can be reduced in favour for better quality or the speed of revolutions can be set to a wanted value.

It is also possible to control the refining process with only the mass flow or the speed of revolutions of the feeding screw as control variable. This simple control method gives good result for the pulp quality.

The figures in this patent application show: Figure 1 shows schematically a usual refiner, Figure 2 shows a typical structure of a refining disc, and Figure 3 is an enlargement from one part of Figure 2.

The refiner 1 and refiner disc 2, which are shown in the figures 1,2 and 3 are well known and their structure or function is not discussed deeper here.

Figure 1 shows the refiner 1 in which only one of the refining discs, with other words the rotor 5 is rotated. It is of course possible to arrange the re- finer 1 of the invention so that both refining discs 4 and 5 are rotated. The rotor 5 is rotated with axle 6, which is rotated with a motor, which is not shown in the figures. The rotation force and/or the rotation speed of the mo- tor are advantageously controllable.

As is well seen from Figures 2 and 3, the texture on the refiner segment 8 of the refiner discs get thicker and thicker near the periphery 9 where the fibril- lation takes place.

The following reference numbers are used in the figures: 1 refiner 2 unrefined wood 3 refined mass 4 rotor 5 stator 6 the axle of the rotor 7 the refining segment of the stator

8 the refining segment of the rotor 9 the periphery of the refining disc.

An adaptive (self-adjusting) control algorithm is presented below. The con- troller is a generalisation of the multivariable control algorithm of Astrom and Wittenmark (1973).

The process can be described by the equation below :

y (t) + A1y(t-1) + ... + Any(t-n) = = Bou (t-k-1) +... + Bn1u (t-k-n) + e (t) + + Cle (t-1) +... Cne (t-n) (1) where u is the input vector and y is the output vector, and {e (t)} is a sequence of independent evenly distributed random vectors with a mean value of zero and the covariance E[e(t)eT(t)] \ R The dimension of all vectors u, y and e is p, and the dimension of all matrices Ai, Bi and Ci is pxp. The matrix Bo is non-singular.

Now we introduce the shift operator q~1 defined as q-1(t) = y(t-1) and the polynomial matrices A(z) = I + A1z + ... + Anzn B(z) = B0 + B1z + ... + Bn-1zn-1 C(z) = I + C1z + ... + Cnzn It is assumed-that all zeros of B (z) are outside the unit circle. Bo is non-singular. The system (1) can be written as A (q-1)y(t) = B(q-1) u (wok-1) + C (q'') e (t) (2) In each sampling interval the adaptive algorithm performs an identification based on the least squares method according to the model presented below.

The obtained parameters are used for calculation of the control strategy.

Estimation The algorithm estimates the parameters for the model y(t) + A(q-1)y(t) = B(q-1)y(t-k-1) + #(t) so that the error e (t) is minimised according to the least squares.

In the model (3) k is selected as the dead time for the process (2), and A (z) and B (z) are pxp polynomial matrices according to A(z) = A0 + A 1z + ... + AnAznA B(z) = B0 + B1z + ... + BnBznB First we assume that Bo = I and Bo = I where Bo is a matrix in the constant term of B (z) for the process (2).

Now we introduce the column vectors <BR> <BR> <BR> <BR> <BR> <BR> 0 0 nA nA<BR> <BR> #i = αil ...αip...αil ...αip ßil1...ßip1....ßilnB...ßipnB]T, i = 1,. where aim is the (i, j) element in the matrix Aki ßij is the (i, j) th element in the matrix Bk, and so on. Then the column vector Oi can be considered to contain the coefficients of the ith row in the model (3).

Further we introduce the row vector ¢ (t-k-1) = [-yT(t-k-1) ... -yT(t-k-l-nA) uT (t-k-2)... uT (t-k-1-nB) (5) The ith row in model (3) can be written as (t) Yi(t) - ui(t-k-1) - #(t-k-1)#i According to the least squares criterion the vector #i at each moment N is calculated so that is minimised. This results in a least squares estimation of each row in (2) based on data which is available at the moment N. When N is large, the initial values are of insignificant importance in (6). The criterion (6) can be written as / The value Ei which minimises (7) is given by the normal equations, see Astrom and Eykhoff (1971).

Control At each moment t the control strategy is calculated from

B(q-1) u (t) = A (q'') y (t) (9) where the polynomial matrices A (z) and B (z) are obtained from the current value of the estimated parameters.

The control strategy can be written as # ut (t) #i (t) i = 1, ..., p (10) The parameters for the controller are the same as the estimated parameters. When we use <BR> <BR> <BR> <BR> <BR> <BR> A » s<BR> <BR> # = [#1,#2, ... #p] ( the strategy (10) can be written as uT(t) = -#(t)#(t) ( The estimated parameter vector Oi in (8) can be recursively calculated from <BR> <BR> <BR> <BR> <BR> <BR> #i(t)=#i(t-1) + K(t-1)[yi(t)-ui(t-k-1)-#(t-k-1)#i(t-1)]<BR> K(t-1)=P(t-1)#T(t-k-1)[1+#(t-k-1)P(t-1)#T(t-k-1)]-1 (13)<BR> <BR> <BR> <BR> <BR> <BR> <BR> <BR> <BR> <BR> <BR> <BR> <BR> <BR> <BR> <BR> P(t)=P(t-1)-K(t-1)[1+#(t-k-1)P(t-1)#T(t-k-1)]KT(t-1) P (t) is a normalised covariance matrix of the estimated n parameters ei- The initial values of P (t) are assumed to be the same for s all parameter vectors Oi. The corresponding amplification vectors K (t-1) will also be the same for all estimators.

Sometimes it may be useful to introduce an exponential weighting for the parameter estimation. This can be done by modifying the criterion (6) to

The last equation in (13) changes to P (t) = l/i {P (t-l) -K (t-1) [1+#)t-k-1)P(t-1)#T(t-k-1)] x KT (t-1)] (15) Another possibility is to use Kalman filters. The covariance matrix P (t) is supplemented by adding to it a matrix Ri instead of the division by Then the equation (15) will be P (t) = P(t-1)-K(t-1)[1+#(t-k-1)P(t-1)#T(t-k-1)] x K (t-1) + Ri It should be noted that the algorithm can be construed as a union of a plurality (here 2) of simple self-adjusting controllers. For instance the controller 2 controls the output signal Y2 (t) by using the control variable u2 (t). y1(t-i) and u1(t-1-i) (i 0) can be used as feedforward signals. This means that the two simple self-adjusting controllers can operate in a cascade mode.

The possibilities for this feature strongly depend on the process properties regarding the model (2) and character of the minimum variance strategy. The multivariable self- adjusting control algorithm can in some circumstances result in the minimum variance, in other words when C (z) = I (the process interference is white noise).

Another possibility is an exclusively multivariable minimum variance control algorithm, which is not adaptive.

It is eventually possible to use another type of controlling than stochastic control.

Even if the invention has here been described with reference to what in the moment can be kept as the most practical and preferred embodiments, it is understood that the invention is not to be limited to the above described em- bodiments, but it is meant to also enclose different modifications and equiva- lent technical solutions in the scope of the enclosed patent claims.

References: 1. Ryti N. , Manner H. , A control strategy for a termomechanical pulping process. Paperi Puu 59,10 : 640-643,645-646, 648-651,639 (Oct. 1977).

[Eng.; Finn. sum.] 2. Bolin G. , Hill J. , Automatic quality control of TMP-manufacture. Appita Vol.

32 5 : 359-365 (1979) 3. Dumont G. A., Legault N. D. , Jack J. S. , Rogers J. H. , Computer control of a TMP plant., Pulp & Paper Canada 83 : 54-59 (1982) 4. Alsip W. P. , Digital control of a TMP operation, Pulp & Paper Canada Vol.

82 no. 3 37-42 (1981) 5. Johansson B., Karlsson H. , Jung E. , Experiences with computer control, based on optical sensors for pulp quality, of a two-stage TMP-plant, Process Control Conference (Halifax).

6. Dahlqvist G. , Ferrari B., Mill operation experience with a TMP refiner con- trol system based on a true disk clearance measurement, EUCEPA Intern. Mech. Pulping Conf. (Oslo) Preprints of Papers Session 3. no. 6: 14 p. (June 16-19, 1981). [Engl.]