EP0947625B1 - Process and apparatus for controlling and optimizing the process of chemical recovery during cellulose production - Google Patents

Description

The invention relates to a process for process control and process optimization of chemical recovery in the production of pulp using at least one state model and / or process model. In addition, the invention relates to an apparatus for carrying out the method.

The production of pulp by the so-called pulp digestion is carried out by boiling wood chips using appropriate cooking chemicals either in a continuous or in a batch process. Sulphure or sulphite pulp cooking usually involves a recovery process for the cooking chemicals.

The process control in the recovery of cooking chemicals is difficult, as important quality parameters of cooking chemicals - the sulphate pulping example, the active alkali concentration, sulfidity, sulfate content, sulfur content, Na ₂ S, Na ₂ SO ₄ , Na ₂ CO ₃ , Na ₂ O, CaO, NaOH, NaCl, K ₂ CO ₃ , CaCO ₃ , Ca (OH) ₂ and in sulfite pulping, for example, the chemicals SO ₂ , HSO ₃ ^- , SO ₃ ^- , S ₂ O ₃ ^- , SO ₄ ^- , MgO , Active MgO, dead burned MgO - measured in the laboratory only after a time delay. For this reason, a meantime erroneous production can not be excluded. To avoid the risk of mis-production, therefore, the production process is usually operated with a greater safety margin in terms of Kochchemikalienqualität than it would actually be necessary.

• Der Prozeß wird nach Erfahrungswerten gesteuert. Regeleingriffe sind nur in geringem Umfang möglich. Die Qualität der Kochchemikalien kann nur im nachhinein bestimmt werden. Eine große Streuung der Qualitätsparameter ist daher häufig nicht vermeidbar.
• Es erfolgt ein analytisches Messen einzelner Komponenten der Kochflüssigkeit mit beispielsweise Titrationsautomaten.

The prior art often uses the following strategies:

• The process is controlled based on experience. Control interventions are only possible to a limited extent. The quality of the cooking chemicals can only be determined with hindsight. A large spread of quality parameters is therefore often unavoidable.
• There is an analytical measurement of individual components of the cooking liquid with, for example, automatic titration.

In addition, it has already been proposed to provide infrared spectroscopy for process evaluation, for which a number of publications exist. From the DE 195 10 009 A1 a method for process control of a paper machine is known. Selected spectral characteristics are used to measure product quality parameters. Especially for process control at the headbox of a paper machine neural networks are used, which were trained on the measured in the laboratory product parameters. Furthermore, from the DE 195 10 008 A1 a process for process control in the pulp and paper production known in which spectral characteristics of different wavelengths for the determination of the starting materials and for the evaluation of the raw material quality are derived. Correction signals are also to be determined for the plant control or regulating devices, in particular with neural networks. In this method, the Meßsignalaufnahme takes place directly on the raw material "wood", such as wood chips or wood fibers, or directly on the raw material "waste paper".

According to the US 5,378,320 In the sulphate pulp production, the absorption is measured at selected wavenumbers by IR spectroscopy and correlations are used to determine different alkali concentrations. It is proposed to determine the lignin content of pulp with spectroscopic measurements.

Finally, out of the DE 36 16 051 A1 a method for controlling the digestion of lignocelluloses using FTIR spectroscopy. By measurement in the MIR (middle infrared) range, the residual lignin content and the concentration of the cooking chemicals is determined, for example the content of free SO ₂ in the sulfite boiling.

Besides that is from the DE 42 21 404 A1 a combustion plant for the spent liquor of a digester with a control device for the combustion air, in which the amount and temperature of an atomized waste liquor in the plant depending on a measure of the ash quality is such that the combustion products of the waste liquor as large amounts of hydratable starting materials for contain the recovery of digestion chemicals.

Finally, in the US 5 616 214 A , of the US 5,364,502 A and the US 5,282,931 proposed to use spectroscopy especially in connection with the recovery of digestion chemicals in pulp production. In these documents samples are taken from the process and examined at defined wavelengths. Apart from that, with the JP 5-321183 A proposed to provide fuzzy control to control the combustion of black liquor in a recovery boiler such that fuzzy rules are established from operating practice and correlated with a suitable membership function.

From the WO 98/10137 A1 For example, a method and apparatus for process control and process optimization in causticizing a green liquor in the sulfate pulp manufacturing process is known. In this case, continuous infrared absorption spectra are measured on the green liquor and on the white liquor and the spectra are evaluated to form parameters. Control signals are set up from the parameters and laboratory measurements in order to control the lime intake in the lime slaker used for causticisation.

When not prepublished prior art according to the DE 196 53 530 A1 , of the DE 196 53 532 A1 , of the DE 196 53 479 A1 and the DE 196 53 477 A1 is focused on the application and evaluation of pulping spectroscopy when bleaching

Paper stocks and in the operation of refiners and paper machines. For these applications, it is proposed in each case to measure at least one point continuous spectra of electromagnetic radiation and / or continuous spectra of mechanical properties, continue to detect at least one location discrete physical and / or chemical properties, by mathematical evaluation of the continuous spectra characteristics form and from the characteristics, the physical and / or chemical properties and laboratory measurements of the product properties a state model and possibly a process model set up.

In contrast, it is an object of the invention to improve the process control specifically for the recovery of chemicals in the production of pulp and to provide an associated device.

The object is achieved in a method of the type mentioned according to the invention with the features of claim 1, wherein advantageous developments are specified in the dependent claims. The associated device is specified in the single claim.

The method of detecting the continuous spectra is alternatively applicable to the sulfite or sulfate process. In the present invention, spectroscopic measurements are carried out specifically on the chemical streams of the recovery plant, ie on sulphate digestion, for example on the soda melt after the recovery boiler, in the green, white and black liquors, and on sulphite digestion, for example in the ash stream or on the ash or in the cooking liquors. carried out. Advantageously, for measuring continuous spectra of electromagnetic radiation absorption, emission and / or luminescence spectroscopy in the spectral range from 0.1 .mu.m to 400 .mu.m, preferably between 0.4 .mu.m and 100 .mu.m. The continuous spectra are evaluated with suitable calculation methods and allow a statement about the expected chemical composition of the cooking liquid. On the basis of these statements can be intervened in the process regulating, eg by changing the air distribution in the recovery boiler, the CaO supply in the so-called Kaustifizierer, by addition of so-called "make-up" chemicals and by regulating the mixing ratio of Black liquor, white liquor and dilution water.

The improved chemical recovery is particularly advantageous for pulp production with homogenization of the quality of the cooking liquor at a high level. This achieves an increase in the yield and the production quantity, a reduction in the use of chemicals and / or energy as well as the saving of auxiliary substances in the bleaching stage subsequent to the pulp production.

The invention is in relation to the above-mentioned prior art in essential points, such as in particular the measuring location, the measurement method, the processing of the spectra and the modeling further developed.

If electromagnetic waves are used in the wavelength range between 100 nm and 400 μm, preferably in the range from 0.4 μm to 100 μm, not only absorption, emission or luminescence spectra but also so-called Raman spectra can be measured. For example, the absorption spectroscopy can be done in transmission, diffuse reflection or attenuated total reflection (ATR = a ttennuated t otal r eflection). The excitation to luminescence can eg by the irradiation of electromagnetic radiation, such as UV radiation or by a specific chemical reaction such Chemiluminescence, done; In contrast, the emission is excited, for example, by irradiation with electrons. When measured in the infrared range (IR: 800 nm to 20 μm), Fourier transform infrared spectroscopy (FTIR) can preferably be used. In the case of inhomogeneous samples, the spectroscopic measurement is repeated several times for a plausibility comparison.

• durch Fourier-Transformation.
• Bei der Messung der Absorption durch diffuse Reflexion durch Umrechnung in sog. Kubelka-Munk-Einheiten und Korrektur von Mehrfachstreueffekten
• durch Normierung und Glättung der Spektren
• durch Ermittlung von für die Modellbildung ungeeigneten Spektren. Die Ausschaltung ungeeigneter Messungen kann z.B. durch Vergleich mit Referenzspektren erfolgen
• durch Bildung von Mittelwerten bei mehreren Spektren zu einer Probennahme.

Preferably, the spectra are preprocessed as follows:

• through Fourier transformation.
• In the measurement of absorption by diffuse reflection by conversion into so-called Kubelka-Munk units and correction of multiple scattering effects
• by normalization and smoothing of the spectra
• by determining spectra unsuitable for modeling. The elimination of unsuitable measurements can be done eg by comparison with reference spectra
• by forming averages at several spectra for sampling.

• PCA-Verfahren (principle component analysis) oder auch sog. Hauptkomponentenanalyse
• PLS-Verfahren(partial least square), einer der Fachwelt bekannten Rechenmethode unter Verwendung kleinster Quadrate
• Neuronale Netze
• Analytische Beschreibung der Spektren, z.B. im Bereich des Infraroten (IR) durch Lage, Intensität und Breite der wichtigsten Absorptions- oder Emissionspeaks, Ermittlung dieser Größen z.B. mit einfachen Minimum-/Maximum-Verfahren oder der
• zweite Ableitung der Spektren

After these first processing steps, the following computer-assisted methods for determining characteristic quantities can be used in whole or in sections on the spectra, the description of the spectra being essentially based on its main components:

• PCA-method (p rinciple c omponent a nalysis) or so-called. Principal Component Analysis
• PLS method (partial least square), a method known in the art using least squares
• Neural Networks
• Analytical description of the spectra, eg in the range of the infrared (IR) by position, intensity and width of the most important absorption or emission peaks, determination of these quantities eg with simple minimum / maximum methods or
• second derivative of the spectra

The parameters are used to model the desired quality parameters.

The state models for calculating the quality parameters can be structured with a sufficiently large number of data based on neural networks, fuzzy systems, multi-linear regression models or combinations thereof. As an alternative to purely data-driven models, combined models are also possible in which additional analytical knowledge is introduced. The process models can be constructed in the same way and with the same means.

Laboratory measurements on intermediate and end products are used to construct the models. The training of the models or their validation is carried out on the basis of the laboratory values and can be repeated at certain intervals, with only a partial readjustment is possible.

A pre-processing integrated into model verification ("Novelty Detection") according to the older, not pre-published DE 196 322 45 A1 can indicate in time the need for a new training phase in the current process.

Figur 1

ein Schema der Meßstellen bei der Chemikalienrückgewinnung,

Figur 2

als Ausschnitt von Figur 1 ein entsprechendes Schema für die Kaustifizierung,

Figur 3

ein kontinuierliches Spektrum von optischen Messungen an Kochflüssigkeit,

Figuren 4 und 5

jeweils ein Zustandsmodell für die Qualitätsparameter der Kochflüssigkeit für den Sulfataufschluß und den Sulfitaufschluß,

Figur 6

ein Prozeßmodell für die Chemikalienaufbereitung beim Sulfatzellstoff,

Figur 7

ein Prozeßmodell für die Chemikalienaufbereitung beim Sulfitzellstoff,

Figur 8

ein dynamisches Prozeßmodell,

Figur 9

den schematischen Aufbau einer Prozeßoptimierung zum Steuern der Chemikalienaufbereitung,

Figur 10

eine Variante zu Figur 8 unter Einbeziehung des dynamischen Modells gemäß Figur 7,

Figur 11

schematisch die Vorverarbeitung und Verdichtung der Spektren und

Figur 12

eine Vorrichtung zur optimierten Prozeßführung bei der Chemikalienaufbereitung.

Further details and advantages of the invention will become apparent from the following description of exemplary embodiments with reference to the drawings. Show it

FIG. 1

a schematic of the measuring points in chemical recovery,

FIG. 2

as a section of FIG. 1 a corresponding scheme for caustification,

FIG. 3

a continuous spectrum of optical measurements of cooking liquid,

FIGS. 4 and 5

in each case a state model for the quality parameters of the cooking liquor for sulphate digestion and sulphite digestion,

FIG. 6

a process model for the treatment of sulphate pulp,

FIG. 7

a process model for the chemical treatment of sulfite pulp,

FIG. 8

a dynamic process model,

FIG. 9

the schematic structure of a process optimization for controlling the chemical treatment,

FIG. 10

a variant too FIG. 8 including the dynamic model according to FIG. 7 .

FIG. 11

schematically the preprocessing and densification of the spectra and

FIG. 12

a device for optimized process control in the treatment of chemicals.

The figures will be described below partially together. The same or equivalent parts have corresponding reference numerals.

In FIG. 1 the structure of a continuously working chemical treatment is shown, wherein in the present context, the Meßstellenschema is clarified. In this case, it is assumed that a known cooker for producing pulp by cooking raw materials, in particular in the form of wood chips, in a suitable cooking liquid. Denoted at 11 is the inlet for the wood into the digester 10, at 12 the supply line for the cooking liquid and at 13 the outlet for the finished pulp. The digester 10 is followed by a blow tank 14 and a plurality of scrubbers 15, 15 ', 15 "with associated filtrate tanks 16, 16', 16". From the washer the finished pulp is spent.

The filtrate is passed through at least one unit 17, advantageously units 17, 17 ', 17 ", ... for multistage evaporation, into a so-called black liquor tank 18 and from there into a recovery boiler 20 for cooking chemicals. In recovery boiler 20, the non-reusable waste liquors referred to as black liquor are to be incinerated and the recyclable cooking chemicals recovered. An electric filter 21 removes the fly ash. The soda melt accumulating at the bottom of the recovery boiler 20 flows into a melt-dissolving tank 22. This is followed by a green liquor clarification tank 23 and a caustification plant in which chemical conversion reactions for liquor extraction take place. This plant consists essentially of a quenching tank 24, several individual Kaustifizierern 25, 25 ', 25 ", a white liquor clarification tank 26 and a cooking liquor tank 27. The Kaustifizierung plant is associated with a scrubber 28 with filter 29 and a lime kiln 30. The there burned lime is returned to the extinguishing tank 24. From the partial illustration to FIG. 2 the operation of the staggered caustics 25, 25 ', 25''and25''' becomes clear. There is a computer 105 for receiving and processing characteristic data, in particular of physical or chemical properties, such as flow m̊ and density ρ, available.

Due to the desired continuous operation of the system, the illustrated chemical treatment is relatively complex: it can be used alternatively after the sulphate process or after the sulphite process. In this case, the process state is to be detected at various significant points of the process chain and the process to be optimized, for which measurement quantities must be recorded. In FIG. 1 for this purpose spectrometers for recording individual continuous spectra A at corresponding points in the process chain, B, C, D, E, F, G, H and I of electromagnetic radiation and in the partial view according to FIG. 2 With appropriate modifications, spectrometers are mounted at suitable measuring points for the spectra F 'and G'. With this arrangement, preferably in the infrared range (IR: 1-25 μm), spectroscopic measurements are carried out online in the cooking liquid. Such a measurement can be carried out, for example, with the aid of a known ATR probe or a known FTIR spectrometer.

The measured spectra are smoothed and normalized by signal preprocessing. Subsequently, a decomposition into the main components and / or the identification of important peaks can take place. The main components can then be modeled, e.g. according to multilinear regression methods, the concentrations of the following quantities are calculated: effective alkali, sulphidity, carbonate, sulphate and thiosulphate.

The latter values can be known, among other values, in process models for the calculation of the quality parameters of in FIG. 1 produced cooking liquid 12 are used.

A regulatory intervention is conceivable at various points in the recovery process, e.g. by replenishment of chemicals, by temperature variations or by changes in the circulation.

In FIG. 3 31 exemplifies the course of an IR spectrum of cooking liquid in the wavenumber range 1500 to 900 cm ^-1 . To obtain the spectra A to I in the FIGS. 1 and 2 can also be measured in other wavenumber ranges.

The basis of the FIG. 3 exemplified spectra are analyzed by a variety of mathematical methods, As already mentioned, this essentially involves a preprocessing of the spectra, the introduction of analytical knowledge and a possible outlier detection, in order to ensure a correct determination of parameters as intended. It is essential that alone from the continuous spectra by a suitable mathematical evaluation such characteristics can be formed specifically for the chemical streams in the recovery process and that from the characteristics and laboratory measurements of the chemical concentrations a state model and / or additionally with process properties a process model is formed. Furthermore, discrete physical and / or chemical properties for further processing can be detected at the chemical streams.

The latter applies to the cooking chemicals both in the sulfate process and in the sulfite process: In particular, in the sulfate process as the chemical recovery process descriptive chemical concentrations in particular the active alkali concentration, the sulfidity, the sulfate content and the proportion of Na ₂ S, Na ₂ CO ₃ and NaOH used. When sulfite are used as the chemical recovery process described chemical concentrations of the acids, the total SO _2, HSO ₃ ^- -, -SO ₃ ^- -, S ₂ O ₃ ^- -, SO ₄ ^- concentrations and for the bases, the MgO Concentration, divided into so-called deadburned MgO and active MgO, used.

In the FIGS. 4 and 5 40A and 40B each indicate a related state model of the cooking chemicals from which their properties, such as concentration, are calculated. From this, the essential quality parameters for the cooking liquid to be processed can be derived. For this purpose, from the spectra either with the so-called principal component analysis (PCA) or With the so-called PLS method (partial least square), the parameters are determined and entered into the model.

Continue to be in Fig. 4 and Figure 5 selected discrete mechanical or chemical properties and entered into the state model 40. However, the state model 40A or 40B can optionally be formed solely with parameters PC1 to PCn determined from the continuous spectra.

The model 40A or 40B may e.g. advantageously be formed by a neural network. In modeling, use is generally made of modern information-technological calculation methods, in particular also evolutionary or genetic algorithms.

In FIG. 6 For example, a process model for chemical recovery by the sulphate process is designated 50. Entered here are, for example, discrete physical and chemical properties, such as temperature, pH and pressure and the state variables of the cooking liquid as an output of the state models, for example, according to the FIGS. 4 or 5 , The process models can also be formed solely with the parameters determined from the continuous spectra.

In FIG. 7 a corresponding process model 60 is given specifically for the sulfite process. If discrete physical and / or chemical properties are to be used, in addition to the optical spectrum, for example, the flow rates and / or the temperatures can be specified, which in FIG. 6 is shown. The pH value for sulfite digestion can be entered.

It is expedient to create your own models for each quality parameter to be calculated. To increase the prediction accuracy For each parameter partial models can be formed.

According to FIG. 8 can the process model according to FIG. 6 also be designed as a dynamic process model 70. The input variables are here according to the input variables in FIG. 6 in each case at discrete points in time k,..., (kn). The same applies to the chemical concentration. From this, the dynamic process model 70 determines the chemical concentration at future times (k + 1.

When setting up the models, except for neural networks, fuzzy methods can also be used. Combined neuro-fuzzy systems can be used.

After setting up the state and / or the process models, it is particularly important to validate the validity of the models, which can be done by an online training of the individual models or sub-models. It may be important for the practice to make a review of the results obtained by computer-aided selection of all information-bearing data. This method was proposed as so-called "Novelty Detection" and allows in the current production process, to introduce new data sets in the evaluation. In the case of inconsistent results, it is necessary to retrain the models.

The sizes obtained from the described modeling methods are used for process control and process optimization in the recovery plant. This is in FIG. 9 the process generally indicated at 80, from which the current process state results from the spectra with 81. The process model is designated here by 82, from which the data is put into a unit for the cost function 83, which simultaneously with data for costs and prices from the unit 84 is charged. An optimizer 85 determines therefrom the manipulated variables 86, which are fed back into the process model 82 and also the optimal manipulated variables 87 for process control. These can also be switched through a switch 88 by the plant operator.

The same applies FIG. 10 for a litigation, in which a dynamic model according to the FIG. 8 is used. Here, a unit 89 with the dynamic model is additionally present, in which the current process state is entered on the one hand, and the optimal control variables, on the other hand, are specified.

Based FIG. 11 illustrates that a unit 91 for preprocessing and compression for the spectra of the total spectrum 90 is used, from which in the evaluation unit 92, for example, the parameters PC 1 to PC 10 are calculated. In the case of the principal component analysis used for this purpose, so-called scores are formed from a suitable number of spectra, for example between three and ten spectra, for the purpose of data reduction. From this, the input variables PC1,..., PCn, in particular the model according to FIG FIG. 4 calculated.

According to FIG. 11 the parameters PC1 to PCn flow into the state model 93 and into the process model 94, in which case additional discrete physical and / or chemical properties and the process description supplement the state model to the process model. From the state model 93, the product properties of the cooking chemicals and from the process model 94 the product properties of the cooking liquid are derived. Now all chemical additives can be calculated and optimized in terms of the targeted cost savings.

FIG. 12 shows, as with appropriate evaluation and optimization software on the basis of a computer, the computer 105 from FIG. 1 be intervened in an existing process control system. Optimized manipulated variables can be generated, which act as a process control system on a known automation device 100, which interacts with the actual system for carrying out the process. In principle, therefore, the existing facilities for chemical recovery by several spectrometers 101 to 103 and an associated, implemented as a software optimization program that runs on the computer 105, supplemented.

Claims (27)

1. Method for process control and process optimisation during chemical recovery in the manufacturing of cellulose using at least one state model or state model and process model, with the following features:

   a) At at least one point continuous spectra of electromagnetic radiation are measured at the chemical flows,

   b) By evaluation of the continuous spectra, characteristic values (PCl...PCn) for the chemical flows are formed,

   c) From the characteristic values (PCl...PCn) and laboratory measurements of the chemical concentrations the state model is set up and the setup of the process model may possibly include process characteristics,

   characterised in that,
   in method step b) the spectra are pre-processed and compressed, that as part of a mathematical evaluation, a main component analysis is performed on a predetermined number of spectra and for data reduction a corresponding number of scores is selected and that the characteristic values for the model formation are determined therefrom for the modelling.
2. Method according to claim 1, characterised in that discrete physical and/or chemical characteristics are detected at at least one point at the chemical flows.
3. Method according to claim 2, characterised in that the discrete physical and/or chemical characteristics are used for setting up the state model and if necessary the process model.
4. Method according to claim 1, characterised in that measurements are made at wavelengths of the electromagnetic radiation between 100 nm and 400 µm.
5. Method according to claim 4, characterised in that the electromagnetic radiation is detected as an absorption, emission, luminescence or a Raman spectrum.
6. Method according to claim 4, characterised in that the electromagnetic radiation is detected in transmission, direct or diffuse reflexion or attenuated total reflectance (ATR).
7. Method according to claim 2, characterised in that the electrical conductivity, the pH value, the temperature, the throughflows, the chemical concentration of the chemical flows are recorded as discrete physical and/or chemical characteristics.
8. Method according to claim 1, characterised in that the specific characteristic values of the spectra, especially the main components, are selected for describing the product state and entered directly into the state model.
9. Method according to claim 1, characterised in that the specific characteristic values of the spectra, especially the main components, are incorporated into the state model and that, at the output of the state model, the product characteristics are formed and entered directly into the process model.
10. Method according to claim 1, characterised in that unsuitable spectra for modelling are eliminated by plausibility checking.
11. The method according to one of the preceding claims, characterised in that neural networks and/or fuzzy logic are employed for modelling.
12. Method according to claim 1, characterised in that the measurements of the spectra are performed in the sulphate process both on the soda melts and also on the aqueous chemical flows such as green liquor, white liquor and black liquor.
13. Method according to claim 1, characterised in that the measurements of the spectra are performed in the sulphite process both on the ash stream and on the ash and on the cooking acids.
14. Method according to claim 12 or claim 13, characterised in that the characteristic variables obtained by evaluation of the spectra are included for closed-loop and/or open-loop control of the recovery process.
15. Method according to claim 14, characterised in that the quality parameters of the finished cooking fluid, especially the chemical concentrations, are modelled for the open-loop and/or closed-loop control of the chemical recovery.
16. Method according to claim 15, characterised in that the model approaches are employed for calculation of the chemical usage as well as for predicting the product quality.
17. Method according to claim 15, characterised in that the active alkali concentration, the sulphidity, the sulphate proportion as well as the proportion of Na₂S, Na₂CO₃ and NaOH are used as the chemical concentrations describing the chemical recovery process.
18. Method according to claim 15, characterised in that, for the acids, the overall SO₂, the HSO₃ ^-, SO₃ ^--, S₂O₃ ^--, SO₄ ^-- concentrations and for the bases the MgO concentration, divided into burned MgO and active MgO, are used as the chemical concentrations describing the chemical recovery process in the sulphite process.
19. Method according to claim 15, characterised in that the model approach with the quality parameters is used in the process optimisation.
20. Method according to one of the preceding claims, characterised in that a cost function is formed which is optimised with an optimiser by suitable variation of the adjustment variables.
21. Method according to claim 19, characterised in that the optimisation is undertaken by genetic algorithms.
22. Method according to claim 20, characterised in that a cost function for the production costs and/or a profit function is used as the cost function.
23. Method according to one of the preceding claims, characterised in that a dynamic model is used for checking the adjustment variables optimised by a static model, wherein a neural network in particular is used as the dynamic model.
24. Method according to one of the preceding claims, characterised in that the model and/or the submodels are trained online.
25. Method according to one of the preceding claims, characterised in that checking of results obtained (novelty detection) is carried out by computer-aided selection of all information-bearing data.
26. Method according to claim 25, characterised in that, if non-consistent results are present, re-training is carried out.
27. Apparatus for carrying out the method according to one of claims 1 to 26, consisting of at least one spectrometer (101, 102, 103), a unit for pre-processing and compressing the spectra, a digital computer (105) for mathematical evaluation of the continuous spectra for the purposes of calculating the characteristic variables (PC1...PCn) and for setting up the state model or the state model and process model from the characteristic variables (PCl...PCn) and if necessary the discrete physical and/or chemical characteristics as process characteristics, and also consisting of a process control system (100).

Images