CA2234221A1 - Process for determining the final point of pulp cooking and an arrangement for controlling the pulp cooking time in a reactor - Google Patents
Process for determining the final point of pulp cooking and an arrangement for controlling the pulp cooking time in a reactor Download PDFInfo
- Publication number
- CA2234221A1 CA2234221A1 CA002234221A CA2234221A CA2234221A1 CA 2234221 A1 CA2234221 A1 CA 2234221A1 CA 002234221 A CA002234221 A CA 002234221A CA 2234221 A CA2234221 A CA 2234221A CA 2234221 A1 CA2234221 A1 CA 2234221A1
- Authority
- CA
- Canada
- Prior art keywords
- cooking
- neural network
- variables
- pulp
- dynamic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Abandoned
Links
Classifications
-
- D—TEXTILES; PAPER
- D21—PAPER-MAKING; PRODUCTION OF CELLULOSE
- D21C—PRODUCTION OF CELLULOSE BY REMOVING NON-CELLULOSE SUBSTANCES FROM CELLULOSE-CONTAINING MATERIALS; REGENERATION OF PULPING LIQUORS; APPARATUS THEREFOR
- D21C7/00—Digesters
- D21C7/12—Devices for regulating or controlling
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/0265—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion
- G05B13/027—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion using neural networks only
Landscapes
- Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Paper (AREA)
- Investigating Or Analyzing Materials Using Thermal Means (AREA)
- Polysaccharides And Polysaccharide Derivatives (AREA)
- Electric Ovens (AREA)
- Control Of High-Frequency Heating Circuits (AREA)
Abstract
The aim of the invention is to determine the point in time during pulp cooking at which the viscosity and/or kappa number of the pulp produced have predetermined values. Since these values can be determined only after the cooking process, appropriate models must be used. To that end, use is made of a neural network whose initial value is the desired viscosity and/or kappa number from which the requisite cooking time can be determined and whose initial values are measurement values of the cooking process. According to the invention, the neural network is adapted during the course of a continuous cooking reaction on the basis of a dynamic model of the changing measurement values. Suitable measurement values during pulp cooking are in particular the SO2 content of the cooking acid and the colour value and electrical conductivity or pH value. In the associated arrangement, the model (10) used is a combination of analytical models (101 to 103, 201 to 204) of the process variables with the neuronal network (100, 200).
Description
FILE, PIN IPI TH!~
T~ TRANSI J~T
Description Process for determining the end point of the cooking of pulp, and arrangement for controlling the cooking time when cooking pulp in a reactor The invention relates to a process for determining the end point in the cooking of pulp, at which time the chemical and/or physical variables for the cooking are characterisic, whereby at least one neural network is used in which measured variables occurring during cooking are input and which gives as initial variable at least the variable characteristic for the cooking. In addition, the invention also relates to the associated arrangement for controlling the cooking time when cooking pulp in a reactor, using control models adapted from the process and at least one neural network.
When cooking pulp, process variables which describe the quality of the cooking process are essentially the viscosity and the kappa number of the pulp produced, moreover, its yield being decisive in practice. Under fixed temperature and pressure conditions and known chemical initial concentrations, the pulp quality achieved depends on the reaction, that is to say on the cooking time. In practice, the cooking time is to be determined in advance as accurately as possible, the viscosity and the kappa number nevertheless not being able to be measured during the cooking process. For these reasons, the values of the above process variables could until now only be estimated.
A control process for producing pulp by pressure and temperature control is known from EP-A-0 590 433 in which a neural network and a fuzzy controller are used in parallel. Usually, the neural network is adapted once by means of the controller. Alternatively thereto, the adaptation should also be continuously in operation AMENDED PAGE
and adapted to tearing resistance or yield. Since these values are determined in the laboratory, this means an adaptation after a cooking of pulp has been carried out.
With the inventor's publication in "Proceedings of the 6th International Conference on Neural Networks and their Industrial and Cognitive Applicationsl', September 1993, Nimes (FR), pages 25 to 32, the use of neural networks for the above problem has already been suggested. In this case, the so-called permanganate number, which can be derived from measured variables, is determined as controlled variable. Using a prescribed core algorithm, a cooking '_ime can thus be forecast by the trained neural network, and is compared with the actual cooking times. In the case of deviations, the neural network is retrained. The retraining of the neural networks is carried out between the individual cooks ("batch") and is complicated.
The object of the invention is therefore to make the latter retraining of the neural networks superfluous. It is intended to specify a process and to provide an associated arrangement with which the end point of the cooking of pulp can be fixed in accordance with a model which can be adapted during the process.
According to the invention, the object is solved therein that the input variables are input into the neural network via units with a dynamic model for the respective measured variable and that, as output variable, in particular the viscosity and/or kappa number, are maintained, whereby the neural network is adapted during a running cooking and used to adapt the dynamic model of the changing measured variables.
AMENDED PAGE
- 2a -In the associated arrangement for controlling the cooking time in the cooking of pulp in a reactor, using control models adapted from the process which contains at least one neural network, the control model is a combination of the neural network with dynamic analytical models of the process variables. In this arrangement, a unit for the on-line learning of the neural network during the cooking is advantageously present.
In the case of the invention, the process model may be adapted automatically, which is therefore of importance since the cooking protcesses]...
AMENDED PAGE
rl~n~ Cl~kFl ;g C-'~LiC~ eut ~cL~ he cooks ("batch") and is complicated.
The object of the invention is therefore to/make the latter retr~; n; ng of the neural networks super~ uous.
It is intended to specify a proce6s and to p ~vide an associated arrangement with which the end po~nt of the cooking of pulp can be fixed in accordance ~ith a model which can be adapted during the process. /
According to the invention, ~ he object is achieved with the following process stfps:
- use is made of such a neural n~twork whose output variable is the viscosity and/~r kappa number, from which the cooking time neede~ to reach the predeter-mined value6 can be dete~mined, and whose input variables are measured v~riables of the cook, - the neural network is a~apted during a rllnn; ng cook, the changing measure~ variables being u~ed for the adaptation of a d ~ ic model.
In this case, the S02 ~ontent of the cooking liquor, its colour value and it~ electrical conductivity or pH are preferably used as ~easured variables.
In the as~ociated arrangement for controlling the cooking time in/the cooking of pulp in a reactor, using control models/adapted from the process and at least one neural netwo~k, the model is a combination of dynamic analytical ~ odels of the process variables with the neural ne ~ork. In this arrangement, a unit for the on-line le ~ning of the neural network during the cook is advant~geoucly present.
/ In the case of the invention, the process model ma~ be adapted automatically, which is therefore of ~ o~t~n~ rinc~ t~ co~kl~.y ~ oc~s~~, as so-called batch proce~ses, have a characteristic which varies with time.
Within the scope of the invention, the adaptation of the neural network during the cooking process results in a two-stage prediction of the viscosity and/or of the kappa number, static and dynamic models being combined.
The static model of the viscosity and/or of the kappa number comprises a neural network which is able to estimate the value of the viscosity and/or of the kappa number for a specific time from the measurements of the colour, the conductivity or the pH and the S02 content for the same time. For this purpose, the measured values such as S02 content, the colour and the conductivity or the pH are modelled as dynamic variables. The generation of such dynamic models is made possible by the measure-ment of the variables during the cooking process.
The dynamic models for the S02 content, the colour and the conductivity or the pH are employed in order to calculate in advance the values of these vari-ables for specific times in the end portion of the cook, from which the determination of the viscosity and/or of the kappa number is then carried out with the aid of the neural network. This results in a time topt for which the estimation of the viscosity and/or of the kappa number is sufficiently close to the desired values.
The significant fact in the invention is that on-line learning is now possible both during and also afterthe cooking process. The dynamic models for the S02 content, the colour and the conductivity or pH are continuously renewed as soon as new measurements are available. The cooking time may be recalculated from these values, using the unchanged neural network. By contrast, the neural network is updated only following the cooking process, when current laboratory measurements are available.
Further details and advantages of the invention emerge from the following description of the figures of exemplary embo~;m~nts, using the drawing, in which, as block diagrams, Figure 1 shows a neural network for monitoring the cooking of pulp,~5 Figure 2 show6 a model specifically for the viscosity prediction with the possibility of on-line adaptation, Figure 3 shows a variant of Figure 2.
In the following, the figures are partly described together.
Shown in Figure 1 iB a reactor l for the cooking of pulp, to which wood and cooking chemicals are fed as raw materials and which supplies pulp as the cooking product. With regard to the setting of predetermined pressure and/or temperature conditions, the reactor 1 is operated by means of a suitable control and/or regulation system 5. The control and/or regulation system 5 is based on a model 10, which may be adapted via a unit 11 to the actual conditions. The unit 11 may be realized by means of suitable software.
The cooking process for the cooking of pulp is in principle a dynamic process, for which reason it must be-described by a dynamic model. However, a dynamic model of a specific variable can only be derived if this variable can be measured during the process. A measurement in particular of the viscosity or a determination of the kappa number is however only possible at the end of the cooking process.
For the latter reason, the viscosity in particu-lar can be determined only via a static model which is generally [lacuna] only for the time interval in which the last cooking processes are valid, for which purpose the other process variables are used. The prescriptions for the viscosity model can in this case be obtained from such variables as are measured during the process. The latter are, for example, the concentration of S02, the colour and the conductivity or pH.
In order to be able to use the specified viscos-ity model in the example of the cooking process for theprediction of the viscosity value at a later time, the values for the S02 concentration, the colour and the conductivity or pH for this time have to be forecast. For this purpose, use i8 made of dynamic models of these variables, which supply estimates for the variables for a desired time t based on known initial values. This results in the prediction in particular of the viscosity for a specific future time as a two-stage process.
For the two-stage process, the values for a specific time are therefore initially forecast with the aid of the dynamic models of the input variables, from which forecast values the desired estimation of the viscosity i8 obtained with the aid of the static viscos-ity model. This is realized in Figure 2: a neural network100 for determin;ng the viscosity is shown, upstream of which are connected dynamic models 101 to 103 for the SO2 content, the colour and the conductivity or pH, which represent the input variables for the neural network.
Connected between them is a unit 110 for adapting these measured values to the desired time. The neural network 100 has, in addition to inputs 111 to 113 for the adapted values of SO2 content, colour and conductivity or pH, further inputs 106 ff. for temperature, pressure, time and other process conditions. The output value 120 characterizes the viscosity, from which the currently required cooking time for the r~nning cook may be deter-mined by means of comparison in the unit 121 with known viscosity values. Furthermore, there is a timing unit 122 which can be coupled back to the input value.
In Figure 3, the pH, the temperature T, the content of free SO2 and the pressure P are taken into account in particular as process variables. The neural network used here is designated by 200 and has a series of inputs 211 to 217 via which, on the one hand, the starting values for t = 0 and, on the other hand, the rnnn;ng process variables are fed. In addition to the overall SO2 content, the liquor-to-wood ratio FV, the wood moisture content and the CO content are also to be taken into account as starting values.
The process variables are in each case assigned units 201 to 204 for the purpose of dynamic modelling, said units being able to be connected via switches 221 to 224 into the input lines for pH, temperature T, content of free SO2 and pressure P. Connected between the modelling units 201 to 204 and the neural network 200 are now in each case integration units 210, 220, 230 and 240 which effect a time integration of the process variables up to the respective instantaneous time. As a result of such a time integration of the measured values, the values entered into the neural network 200 in each case take into account the previous history since the begin-ning of the cook and thus ensure an adaptation of the measured values, corre~po~ing to the unit llO from Figure 2. Via the output 220 of the neural network a visco~ity function ~(t) is output, u6ing which an indica-tion of the viscosity variation is carried out on an indicator unit 206. Now, it is important that the viscos-ity function ~(t) is a function not only of the variationwith time of the cook but also reflects to the same extent the quality and/or yield of the pulp produced as the functional parameters which are decisive in practice.
This function is thus ~uitable in a particular way for the control of the cook. The determination of the cooking time is carried out in a way correspo~; ng to the units 121 and 122 from Figure 2.
Using the proces~ de~cribed by reference to Figures 1 to 3 and the associated arrangement, the analytical preknowledge about the cooking process is combined in a particularly advantageous manner with a neural network. The result is thus a prediction model with a suitably trained neural network into which, via suitable dynamic models of the variables, their behaviour with time flow~. Thus, the neural network, which, viewed intrinsically, represents a static model, is given a dynamic a~ a result of the on-line adaptation and, in particular, the retr~;n;ng which wa~ previously necessary at certain intervals in cooking pauses is made ~uperflu-ous.
Trials in practice in the cooking of pulp haveshown that better results are achieved than in the previously used, mo~tly statistical models.
T~ TRANSI J~T
Description Process for determining the end point of the cooking of pulp, and arrangement for controlling the cooking time when cooking pulp in a reactor The invention relates to a process for determining the end point in the cooking of pulp, at which time the chemical and/or physical variables for the cooking are characterisic, whereby at least one neural network is used in which measured variables occurring during cooking are input and which gives as initial variable at least the variable characteristic for the cooking. In addition, the invention also relates to the associated arrangement for controlling the cooking time when cooking pulp in a reactor, using control models adapted from the process and at least one neural network.
When cooking pulp, process variables which describe the quality of the cooking process are essentially the viscosity and the kappa number of the pulp produced, moreover, its yield being decisive in practice. Under fixed temperature and pressure conditions and known chemical initial concentrations, the pulp quality achieved depends on the reaction, that is to say on the cooking time. In practice, the cooking time is to be determined in advance as accurately as possible, the viscosity and the kappa number nevertheless not being able to be measured during the cooking process. For these reasons, the values of the above process variables could until now only be estimated.
A control process for producing pulp by pressure and temperature control is known from EP-A-0 590 433 in which a neural network and a fuzzy controller are used in parallel. Usually, the neural network is adapted once by means of the controller. Alternatively thereto, the adaptation should also be continuously in operation AMENDED PAGE
and adapted to tearing resistance or yield. Since these values are determined in the laboratory, this means an adaptation after a cooking of pulp has been carried out.
With the inventor's publication in "Proceedings of the 6th International Conference on Neural Networks and their Industrial and Cognitive Applicationsl', September 1993, Nimes (FR), pages 25 to 32, the use of neural networks for the above problem has already been suggested. In this case, the so-called permanganate number, which can be derived from measured variables, is determined as controlled variable. Using a prescribed core algorithm, a cooking '_ime can thus be forecast by the trained neural network, and is compared with the actual cooking times. In the case of deviations, the neural network is retrained. The retraining of the neural networks is carried out between the individual cooks ("batch") and is complicated.
The object of the invention is therefore to make the latter retraining of the neural networks superfluous. It is intended to specify a process and to provide an associated arrangement with which the end point of the cooking of pulp can be fixed in accordance with a model which can be adapted during the process.
According to the invention, the object is solved therein that the input variables are input into the neural network via units with a dynamic model for the respective measured variable and that, as output variable, in particular the viscosity and/or kappa number, are maintained, whereby the neural network is adapted during a running cooking and used to adapt the dynamic model of the changing measured variables.
AMENDED PAGE
- 2a -In the associated arrangement for controlling the cooking time in the cooking of pulp in a reactor, using control models adapted from the process which contains at least one neural network, the control model is a combination of the neural network with dynamic analytical models of the process variables. In this arrangement, a unit for the on-line learning of the neural network during the cooking is advantageously present.
In the case of the invention, the process model may be adapted automatically, which is therefore of importance since the cooking protcesses]...
AMENDED PAGE
rl~n~ Cl~kFl ;g C-'~LiC~ eut ~cL~ he cooks ("batch") and is complicated.
The object of the invention is therefore to/make the latter retr~; n; ng of the neural networks super~ uous.
It is intended to specify a proce6s and to p ~vide an associated arrangement with which the end po~nt of the cooking of pulp can be fixed in accordance ~ith a model which can be adapted during the process. /
According to the invention, ~ he object is achieved with the following process stfps:
- use is made of such a neural n~twork whose output variable is the viscosity and/~r kappa number, from which the cooking time neede~ to reach the predeter-mined value6 can be dete~mined, and whose input variables are measured v~riables of the cook, - the neural network is a~apted during a rllnn; ng cook, the changing measure~ variables being u~ed for the adaptation of a d ~ ic model.
In this case, the S02 ~ontent of the cooking liquor, its colour value and it~ electrical conductivity or pH are preferably used as ~easured variables.
In the as~ociated arrangement for controlling the cooking time in/the cooking of pulp in a reactor, using control models/adapted from the process and at least one neural netwo~k, the model is a combination of dynamic analytical ~ odels of the process variables with the neural ne ~ork. In this arrangement, a unit for the on-line le ~ning of the neural network during the cook is advant~geoucly present.
/ In the case of the invention, the process model ma~ be adapted automatically, which is therefore of ~ o~t~n~ rinc~ t~ co~kl~.y ~ oc~s~~, as so-called batch proce~ses, have a characteristic which varies with time.
Within the scope of the invention, the adaptation of the neural network during the cooking process results in a two-stage prediction of the viscosity and/or of the kappa number, static and dynamic models being combined.
The static model of the viscosity and/or of the kappa number comprises a neural network which is able to estimate the value of the viscosity and/or of the kappa number for a specific time from the measurements of the colour, the conductivity or the pH and the S02 content for the same time. For this purpose, the measured values such as S02 content, the colour and the conductivity or the pH are modelled as dynamic variables. The generation of such dynamic models is made possible by the measure-ment of the variables during the cooking process.
The dynamic models for the S02 content, the colour and the conductivity or the pH are employed in order to calculate in advance the values of these vari-ables for specific times in the end portion of the cook, from which the determination of the viscosity and/or of the kappa number is then carried out with the aid of the neural network. This results in a time topt for which the estimation of the viscosity and/or of the kappa number is sufficiently close to the desired values.
The significant fact in the invention is that on-line learning is now possible both during and also afterthe cooking process. The dynamic models for the S02 content, the colour and the conductivity or pH are continuously renewed as soon as new measurements are available. The cooking time may be recalculated from these values, using the unchanged neural network. By contrast, the neural network is updated only following the cooking process, when current laboratory measurements are available.
Further details and advantages of the invention emerge from the following description of the figures of exemplary embo~;m~nts, using the drawing, in which, as block diagrams, Figure 1 shows a neural network for monitoring the cooking of pulp,~5 Figure 2 show6 a model specifically for the viscosity prediction with the possibility of on-line adaptation, Figure 3 shows a variant of Figure 2.
In the following, the figures are partly described together.
Shown in Figure 1 iB a reactor l for the cooking of pulp, to which wood and cooking chemicals are fed as raw materials and which supplies pulp as the cooking product. With regard to the setting of predetermined pressure and/or temperature conditions, the reactor 1 is operated by means of a suitable control and/or regulation system 5. The control and/or regulation system 5 is based on a model 10, which may be adapted via a unit 11 to the actual conditions. The unit 11 may be realized by means of suitable software.
The cooking process for the cooking of pulp is in principle a dynamic process, for which reason it must be-described by a dynamic model. However, a dynamic model of a specific variable can only be derived if this variable can be measured during the process. A measurement in particular of the viscosity or a determination of the kappa number is however only possible at the end of the cooking process.
For the latter reason, the viscosity in particu-lar can be determined only via a static model which is generally [lacuna] only for the time interval in which the last cooking processes are valid, for which purpose the other process variables are used. The prescriptions for the viscosity model can in this case be obtained from such variables as are measured during the process. The latter are, for example, the concentration of S02, the colour and the conductivity or pH.
In order to be able to use the specified viscos-ity model in the example of the cooking process for theprediction of the viscosity value at a later time, the values for the S02 concentration, the colour and the conductivity or pH for this time have to be forecast. For this purpose, use i8 made of dynamic models of these variables, which supply estimates for the variables for a desired time t based on known initial values. This results in the prediction in particular of the viscosity for a specific future time as a two-stage process.
For the two-stage process, the values for a specific time are therefore initially forecast with the aid of the dynamic models of the input variables, from which forecast values the desired estimation of the viscosity i8 obtained with the aid of the static viscos-ity model. This is realized in Figure 2: a neural network100 for determin;ng the viscosity is shown, upstream of which are connected dynamic models 101 to 103 for the SO2 content, the colour and the conductivity or pH, which represent the input variables for the neural network.
Connected between them is a unit 110 for adapting these measured values to the desired time. The neural network 100 has, in addition to inputs 111 to 113 for the adapted values of SO2 content, colour and conductivity or pH, further inputs 106 ff. for temperature, pressure, time and other process conditions. The output value 120 characterizes the viscosity, from which the currently required cooking time for the r~nning cook may be deter-mined by means of comparison in the unit 121 with known viscosity values. Furthermore, there is a timing unit 122 which can be coupled back to the input value.
In Figure 3, the pH, the temperature T, the content of free SO2 and the pressure P are taken into account in particular as process variables. The neural network used here is designated by 200 and has a series of inputs 211 to 217 via which, on the one hand, the starting values for t = 0 and, on the other hand, the rnnn;ng process variables are fed. In addition to the overall SO2 content, the liquor-to-wood ratio FV, the wood moisture content and the CO content are also to be taken into account as starting values.
The process variables are in each case assigned units 201 to 204 for the purpose of dynamic modelling, said units being able to be connected via switches 221 to 224 into the input lines for pH, temperature T, content of free SO2 and pressure P. Connected between the modelling units 201 to 204 and the neural network 200 are now in each case integration units 210, 220, 230 and 240 which effect a time integration of the process variables up to the respective instantaneous time. As a result of such a time integration of the measured values, the values entered into the neural network 200 in each case take into account the previous history since the begin-ning of the cook and thus ensure an adaptation of the measured values, corre~po~ing to the unit llO from Figure 2. Via the output 220 of the neural network a visco~ity function ~(t) is output, u6ing which an indica-tion of the viscosity variation is carried out on an indicator unit 206. Now, it is important that the viscos-ity function ~(t) is a function not only of the variationwith time of the cook but also reflects to the same extent the quality and/or yield of the pulp produced as the functional parameters which are decisive in practice.
This function is thus ~uitable in a particular way for the control of the cook. The determination of the cooking time is carried out in a way correspo~; ng to the units 121 and 122 from Figure 2.
Using the proces~ de~cribed by reference to Figures 1 to 3 and the associated arrangement, the analytical preknowledge about the cooking process is combined in a particularly advantageous manner with a neural network. The result is thus a prediction model with a suitably trained neural network into which, via suitable dynamic models of the variables, their behaviour with time flow~. Thus, the neural network, which, viewed intrinsically, represents a static model, is given a dynamic a~ a result of the on-line adaptation and, in particular, the retr~;n;ng which wa~ previously necessary at certain intervals in cooking pauses is made ~uperflu-ous.
Trials in practice in the cooking of pulp haveshown that better results are achieved than in the previously used, mo~tly statistical models.
Claims (10)
1. Process for determining the end point of the cooking of pulp, at which time the chemical and/or physical variables for the cooking are characteristic, whereby at least one neural network is used, for which measured variables occurring during the cooking are input as input variables and which outputs, as output variable, at least one physical or chemical variable characteristic for the cooking, characterized in that the input variables are input into the neural network via units with a dynamic model for the respective measured variable and that, as output variable, especially the viscosity and/or the kappa number are maintained, whereby the neural network is adapted during a running cooking and the changing measured variables being used for the adaptation of the dynamic model.
2. Process according to claim 1, characterized in that the measured variables are the SO2 content of the cooking liquor, its colour value and its electrical conductivity or pH.
3. Process according to claim 1, characterized in that, in addition to the dynamically changing variables, use is furthermore made of the time (t), the temperature (T) and other process conditions (x1, x2) as input variables for the neural network.
4. Process according to claim 1, characterized in that the dynamic models are adapted during a cook ("batch"),and the neural network is adapted after a cook ("batch").
5. Arrangement for controlling the cooking time in the cooking of pulp in a reactor, having a control model adapted to the process which contains at least one neural network, whereby a process according to claim 1 or one of the claims 2 to 4 is applied, characterized in that the control model is a combination of the neural network (100, 200) with dynamic models (101- 103,201 - 204) of the process variables.
6. Arrangement according to claim 5, characterized in that the analytical models (101, 102, 201, 203) describe the dynamic behaviour at least of the SO2 concentration, of the electrical conductivity or of the pH in the cooking liquor.
7. Arrangement according to claim 5, characterized in that further models (103, 202, 204) describe the dynamic behaviour of the colour value and of the temperature and of the pressure.
8. Arrangement according to claim 5, characterized in that a unit (110) for the on-line learning of the neural network (100) is connected between the neural network (100) and the dynamic models (101 to 103) for the process variables.
9. Arrangement according to claim 5, characterized in that integration units (210, 220, 230, 240) for the integration of the process variables up to the current cooking time are connected between the neural network (200) and the dynamic models (201 to 204) for the process variables.
10. Arrangement according to claim 5, characterized in that the output (219) of the neural network (200) drives an indicator unit (206) for the viscosity function (~(t)), from which the end point ofthe cook and the quality and/or yield of the pulp may be derived.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| DE19537539 | 1995-10-09 | ||
| DE19537539.4 | 1995-10-09 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CA2234221A1 true CA2234221A1 (en) | 1997-04-17 |
Family
ID=7774378
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA002234221A Abandoned CA2234221A1 (en) | 1995-10-09 | 1996-09-26 | Process for determining the final point of pulp cooking and an arrangement for controlling the pulp cooking time in a reactor |
Country Status (7)
| Country | Link |
|---|---|
| EP (1) | EP0854953B1 (en) |
| AT (1) | ATE189715T1 (en) |
| CA (1) | CA2234221A1 (en) |
| DE (1) | DE59604434D1 (en) |
| NO (1) | NO981650D0 (en) |
| WO (1) | WO1997013916A2 (en) |
| ZA (1) | ZA968469B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO1999028548A1 (en) * | 1997-11-26 | 1999-06-10 | Siemens Aktiengesellschaft | Control device for a continuos digester for the production of cellulose |
| CA2752471C (en) | 2009-02-13 | 2017-04-25 | Abb Research Ltd. | A method and a system for optimization of parameters for a recovery boiler |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB747108A (en) * | 1953-10-01 | 1956-03-28 | Courtaulds Ltd | Improvements in and relating to the production of wood pulp |
| DE2123497C3 (en) * | 1970-05-15 | 1978-08-03 | Mo Och Domsjoe Ab, Oernskoeldsvik (Schweden) | Process for the manufacture of sulphate pulps to obtain pulps with a predetermined degree of delignification |
| US3941649A (en) * | 1972-07-14 | 1976-03-02 | Mo Och Domsjo Aktiebolag | Process for obtaining a predetermined Kappa number in sulfate pulping |
| US4086129A (en) * | 1975-11-03 | 1978-04-25 | International Telephone And Telegraph Corporation | Process for controlling the intrinsic viscosity of sulfite pulp |
| KR880000744B1 (en) * | 1982-11-24 | 1988-05-04 | 더 뱁콕 앤드 윌콕스 캄패니 | Sulfite digester rate of delignification system |
| DE3641785A1 (en) * | 1986-03-31 | 1987-10-08 | Wolfen Filmfab Veb | Process for producing pulp |
| EP0445321B1 (en) * | 1990-03-05 | 1994-06-01 | Siemens Aktiengesellschaft | Process for making pulp in a continuous digester |
| DE9017325U1 (en) * | 1990-12-21 | 1992-01-02 | Siemens AG, 8000 München | Process control system for controlling pulp production |
| EP0590433B1 (en) * | 1992-10-02 | 1999-08-25 | Siemens Aktiengesellschaft | Control process for the production of pulp by control of pressure and temperature |
-
1996
- 1996-09-26 WO PCT/DE1996/001841 patent/WO1997013916A2/en active IP Right Grant
- 1996-09-26 DE DE59604434T patent/DE59604434D1/en not_active Expired - Fee Related
- 1996-09-26 AT AT96942247T patent/ATE189715T1/en not_active IP Right Cessation
- 1996-09-26 CA CA002234221A patent/CA2234221A1/en not_active Abandoned
- 1996-09-26 EP EP96942247A patent/EP0854953B1/en not_active Expired - Lifetime
- 1996-10-08 ZA ZA968469A patent/ZA968469B/en unknown
-
1998
- 1998-04-08 NO NO981650A patent/NO981650D0/en not_active Application Discontinuation
Also Published As
| Publication number | Publication date |
|---|---|
| WO1997013916A3 (en) | 1997-07-17 |
| DE59604434D1 (en) | 2000-03-16 |
| ATE189715T1 (en) | 2000-02-15 |
| EP0854953B1 (en) | 2000-02-09 |
| NO981650L (en) | 1998-04-08 |
| ZA968469B (en) | 1997-04-09 |
| WO1997013916A2 (en) | 1997-04-17 |
| NO981650D0 (en) | 1998-04-08 |
| EP0854953A2 (en) | 1998-07-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP5006806B2 (en) | Process for controlling the efficiency of the causticization process | |
| US5581459A (en) | Plant operation support system | |
| JPH05324006A (en) | Device and method of controlling process | |
| US5472273A (en) | System for determining the development status of a mass such as bread dough in a powered mixer | |
| CA2418226A1 (en) | On-line calibration process | |
| KR100188926B1 (en) | Defrosting method and its apparatus of ga fuzzy theory refrigerator | |
| CA2234221A1 (en) | Process for determining the final point of pulp cooking and an arrangement for controlling the pulp cooking time in a reactor | |
| EP1290507B1 (en) | Method and system for monitoring and analyzing a paper manufacturing process | |
| EP0905229A3 (en) | Process and device to determine and control the physiologic condition of microbial cultures | |
| JPH063243A (en) | Method for estimating physical property of rubber polymer | |
| CN115058555B (en) | Intelligent soft measurement method and system for measuring carbon content of converter endpoint | |
| Nielsen et al. | Identification and control of nutrient removing processes in wastewater treatment plants | |
| JP3100803B2 (en) | Fuzzy control device | |
| JPH06199904A (en) | Method for changing continuous operation condition of continuous polymerization process | |
| CN116380205A (en) | Flue air volume error soft correction method and system based on multi-metadata fusion | |
| CN116772284A (en) | Heating furnace temperature control method | |
| JP2023519112A (en) | Process control method and process control system | |
| Fogarty et al. | System identification for load balancing with the genetic algorithm | |
| SU1430431A1 (en) | Method of automatic control of instrument sulfate pulp digestion process | |
| Gough et al. | Kraft continuous digester effective alkali control | |
| Leiviskä | INTRODUCTION TO APPLICATIONS IN PULP AND PAPER INDUSTRY | |
| Eerikainen | Artificial intelligence in extrusion cooking process modelling and control. | |
| Wang et al. | A Hybrid Intelligent Controller Based on" Consultation of Doctors" | |
| Leiviskä | INTELLIGENT APPLICATIONS IN PULP AND PAPER INDUSTRY | |
| JPS6394308A (en) | Batch control diagnosis device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| FZDE | Discontinued |