DD277701A1 - METHOD FOR THE AUTOMATIC CONTROL OF BIOTECHNICAL FABRICATION PROCESSES BASED ON A KNOWLEDGE-BASED ON-LINE CONTROL SYSTEM - Google Patents

Abstract

The invention relates to a method for the automatic control of biotechnical material conversion processes on the basis of a knowledge-based on-line control system, in which microorganisms are involved. In order to determine the switchover point for process state changes and periodic process control, a knowledge-based on-line control system is used, which is structured in a parallel real-time and dialog / consulting module with a specific task spectrum and comprises three knowledge bases, whereby the knowledge base I forms the communication base of the two modules, the knowledge base II contains the rules and facts for fault detection and cause localization and the knowledge base III serves to determine the optimal control vector. This control system allows a flexible response to biological variations or changing environmental conditions in complex fermentation systems.

Description

Yd = Yk (D-Yk (2).

12. The method according to claim 10, characterized in that as an intelligible classifier a Trennebenenklassifikator (classifier on the basis of Hyperf laugh) is used, being used as the decision of the classifier evaluating the current difference of the corresponding class membership values <Yo> in the decision making ,

The value <Yd> is determined according to the following relationship:

Yo = X ^t * W- B- A.

13. The method of claim 1 to 11 or 1 to 10 and 12, characterized in that preferably the reference variables nutrient supply (supply of C and energy source, O ₂ -, nutrient salt and trace elements) and / or temperature, pH in the fermentation medium , Energy input, residence time, amount of fermentation medium in the reactor, amount and composition of the gas entry into the Fermontor use.

Field of application of the invention

The invention relates to a method for the automatic control of biotechnical material conversion processes in which microorganisms, which may also be genetically modified, are involved in pure odor mixing. These processes may include the extraction of biomass, the production of P.uucts, the destruction of waste or the extraction of enrichment ore.

Characteristic of the known technical solutions

There are ancestors of the biotechnical material conversion are known in which by an alternating gradual change in the carbon substrate concentration in Fermontationssystem, the duration of the periods of the length dor phases of the cell cycle is adjusted, a Vorbosserunn the effectiveness of the process, especially a reduction in the use of raw materials is achieved. (DDPS 140150, DD-PS 219035).

This scheduling control of a partially synchronized microorganism routine operates at fixed periods, that is, this control principle does not take into account the input disturbances of the fermentation system, such as changes in temperature, inlet concentrations, inflows and raw material quality, on the metabolic activity of the microorganisms in the fermentor. In addition, the variability of biological properties, such. As the duration of the phases of the cell cycle, are not taken into account. However, these disturbances and variability require different periods for a favorable process control. This requirement can not meet the solutions described above or in an off-line control only with greatly increased effort.

Furthermore, methods for the automatic control of processes of biotechnological material conversion are known or proposed, which meet the above requirement for adaptation of the process control due to disturbances and variability of the fermentation system to some extent (DD-WP 243296 and DD-WPC 12Q / 306778.3). Since the process adaptation in the first control system (DD-WP 243296) is realized only by a prognosis filter, considerable problems with regard to reliability and adaptation quality occur in this method. H. accurately determining the process state changes that lead to productivity and effectiveness losses. Diener's case also applies, albeit to a lesser extent, to the proposed hierarchical control system (DD-WPC 12Q / 306778.3), mainly because the results of situational awareness are taken for granted without checking their biological significance. The limited flexibility of the situation recognition algorithms, in particular the forecast filter, and lack of robustness in the occurrence of disturbances in the measured value detection limit the applicability of these controls and require a high effort for the retraction of this automatic process control. Further productivity and effectiveness losses are due to the fact that these systems do not allow operator intervention in specific process situations without interrupting the control and do not allow secure experience for optimal process control in the form of rules (no algorithms) and off-line to be won Anaiysenwerte to integrate into the determination of optimal control.

In known solutions measures for fault detection and -Ursachenbestimmung are not respected or in too small a degree, although it is known that in unfavorable cases, even brief changes in environmental conditions in bioreactors can lead to irreversible damage to the microorganisms or undesirable processes.

Object of the invention

The aim of the invention is to increase the material and energy efficiency and / or the productivity of the process, the improvement of stability, the reduction of susceptibility and the saving of manpower for process control.

Furthermore, in the event of the occurrence of non-controllable disturbances in the bioreactor (defective transducers or control loops), damage to the process and reactor should be averted or kept as low as possible.

The invention also aims to significantly reduce the cost of retraction of the automatic control or for the consideration of technological design and / or procedural changes.

Essence of the invention

It is an object of the present invention to provide such an on-line learning control system which, by flexibly responding to biological variations or changing environmental conditions in the complex fermentation system (hierarchical structure), allows for optimal and reliable determination of the changeover timing in process state changes and periodic processing , The control must be influenced by the operator deh at any time. This concerns both the determination of the control vector as well as the change of model parameters as well as parameters of the situation recognition during the process control, in order to minimize the introduction effort of the automatic control. Furthermore, it must be guaranteed to incorporate knowledge of experience for optimal process control in the form of rules and facts as well as off-line analysis values in the determination of the optimal control. Non-controllable disturbances in the fermentation system (disturbances of the measuring sensor, control circuits) are to be detected quickly, their cause (s) to be determined and, if necessary, redundancies to be activated. What is essential is that the complex control system has great flexibility, i. H. a high degree of adaptivity with regard to various biotechnological processes as well as technological objectives. In the event of a breakdown in the field of actuators, the knowledge-based control in open-circuit operation as a guidance system must be functional.

According to the invention, the object is achieved by using a knowledge-based (ie based on elements of artificial intelligence) on-line control system, which is structured in real-time and dialogue or consulting modules and three knowledge bases (or relatively independent components of a Knowledge base), where the knowledge base I is designed as a superordinate dynamic knowledge base and which receives declarative knowledge for the characterization of all the proxy size and the current process state, the knowledge (rule) base II contains the rules and facts for fault location and cause localization and the knowledge (rule) basis III using algorithmic software (situation detection, model simulation) to determine the optimal control vector. The Wosen dor Systomstruktur bostoht in Q <"- _r ^, Jlolon work of real-time and dialogue / consultation modules and their communication übor diegomoinsamo dyne nischo knowledge base woboi also performed on oinem third channel model simulations to determine dor Optimalon control werdon whose results dor Ontsprechondon Rule Base (III) is constantly in place for submission stohon. For clio implementation of the controller, therefore, a multitasking computer operating system with at least three channel or odor oino equivalent hardware-side solution is to be provided

 Dom real-time module can be assigned to the following tasks:

- schnollo data acquisition and, if necessary, primary daton processing,

- Aiuoigo and storage of the Daton (Moßworto, controlled variables),

- Determination and determination of disturbances on the basis of the corresponding knowledge base (II),

Determination of the optimal control on the basis of the special knowledge base (III), whereby the inventive design of this knowledge base takes into account both the epigenetic and metabolic regulatory level of the microorganisms as well as short-term disturbances in the fermentation system, which among other things can influence the residence time of the microorganisms in the fermenter be and

- Output of the control vector.

The dialing module ensures the following range of tasks:

Input and storage of the current process data (e.g., setpoints),

- Input of off-line analysis values (with time of sampling),

- control interventions by the operator (at any time), such as switching of the operating mode, technologically conditioned parameter changes (eg pumping rates of pumps, setpoints) and change of the model parameters or the characteristics of the situation detection,

- Output of fault messages with indication of the possible cause of the fault (s),

- graphic evaluations of the course of the process,

- Consultation when changing the process state and

- Conducting an active dialogue (computer request / operator response) in the case of a not definable by the control system process or reactor state.

The control structure according to the invention proves to be superior to all other known or proposed controls, because the described complex task or functional spectrum of real-time and dialogue module is realized in parallel under real-time conditions while ensuring a high flexibility in terms of use both for controlling various biotechnological processes as well changing technological objectives, which is reflected, among other things, in the greatly reduced effort required to implement this control system.

The qualitative and quantitative knowledge of the individual process variables is structured in a frame-like manner by means of the definition of composite objects, whereby the frame structure takes into account the temporal validity of the information. In the common dynamic knowledge base (I), the categories (predicates) "Constant", "Relative" and "Variable" are distinguished.The category "Constant" contains information with a permanent character, such as the listing of all process variables with name, unit, type ( Control, on-line or off-line analysis value), identification as a model parameter, if applicable, and technological or physical / technical information used for fault detection and localization, such as. B. technologically related limits, maximum slew rate and tolerance range for controlled variables. Changes are only necessary if new control loops or measuring points are integrated into the reactor system, new analytical methods are introduced or major technological modifications are made to the bioreactor. These changes can be made via any word processing system and require no special knowledge of the control system. The category "Relative" includes data on the process variables listed in "Constant", which must also be changeable during the process control, such as setpoints for the controls, correction factors for measured variables and the identification of a process variable as a feature for the situation detection, d. H. The Prognosis Filter and / or an adaptive classifier (/ 1 / Möckel, B. Bley, Th .: Böhme, B. Heinritz, B. Gerhardt, M .: Method for the automatic control of processes of the

Biotechnical transformation based on a hierarchical tax strategy. DD-VVPC12 Q / 306778.3. 121 Möckel, B .; Bley, Th .; Böhme, B .: Cyclic control of continuous biotechnological processes on the basis of a hierarchical control system. In: J.Systems Analysis Modeling Simulation (3) 1989).

Furthermore, this category contains all information on the current process status, such as operating mode (batch, fedbatch, continuous operation), control regimes, fault messages and decision aids (consultation). In the "Variable" category, in addition to the current values of the process variables listed in "Constant", a process-specific number of past values are included for trend determination.

The rule base II contains the procedural knowledge for processing the information stored in the common dynamic knowledge base in the sense of a comprehensive fault detection and -Ursachenbestimmung (specific maturity tests for all process variables, tolerance test for controlled variables and complex evaluation of individual results). If disturbances occur in the detection of process variables which are specified as characteristics of the situation recognition or as model parameters, these are automatically deleted as features or the last meaningful value is retained as model parameter in order to avoid incorrect decisions of the control system.

An essential basis for the knowledge base for the determination of the optimal control vector (III) is formed by information about the situation recognition (prognosis filter, learning classifier) and model simulations. At the same time, the systems are assigned the characteristics specified in the common dynamic knowledge base at the sampling times. Transfer model parameter vector. By modeling the epigenetic level of regulation is essentially detected, whereas the Si'untionserkennung both epigenetic and metabolic regulations as well as short-term disturbances of environmental conditions in the fermentor (eg fluctuations in substrate or Laugenzufluß), among other things influence on the residence time of microorganisms in the Fermenter or the duration of the phases in the cell cycle have reacted. The prognostic filter employed in the method according to the invention is the Gaussian filter with a spociellom triggering mechanism which is particularly suitable for use under real-time conditions. This triggering mechanism, responsive to changes in the oinzelnon percent size trend, is based on the constant determination of prediction <SP (i)> and current stochastic <o (i)> allor features, as well as the comparison of the quotient <Q> of these two magnitudes of the gosamton morkmalvoctor with oinom according to the fishorvertoilung given Gronzwert <Q>.

Dor lornfähigo classifier, which assigns ooror current Morkmalvektor oinor typical process situation, can be designed in Orfindungsgornäßon ancestor both as a Schworpunkt- as well as Tronnobenonklassifikator.

All Elemonto the situational design are described in detail in / 1 / and / 2 /.

For the determination of oplimolone cyclic control by model, this ancestor employs a structured state modulus or a similar modulus sufficiently describing the dynamics of microorganism populations, which is mathematically manageable under true-zoon conditions {121 and / 3 / Bley, Th .; State-structure models of MLCrobial growth. Acta Biotechnol. 7 [1987], p. 173).

The use of the knowledge-based control system according to the invention with the above-described structure (multitasking operation of the structural elements) makes it possible to use more complex, adequately described models of the process, whose execution schedules for simulation calculations for determining the optimal control can be significantly greater than the sampling time Statements of the model calculation by the knowledge-based system as a function of the age of the result will be included in the decision. In the method according to the invention, the main task of the knowledge base for determining the optimal control is the complex evaluation and interpretation of the statements obtained by the situation recognition and model simulation, including assured heuristic knowledge (empirical knowledge) stored in the form of rules and the evaluation of important ones the balance of biotechnological processes, such as carbon, nitrogen and oxygen balance By taking account of process-specific knowledge, it is possible, in particular, to examine the biological conspicuousness of tax decisions.

This knowledge base uses a specific evaluation scheme (point table), which quantifies the validity or quality of the statements of the subsystems.

In order to achieve a continuous evaluation spectrum of the subsystems of the situation recognition, in addition to the yes / no decision with respect to state changes of the process at the forecast filter the current value of the above-described quotient (Q) and the learning classifiers the current difference of the corresponding class membership values Yo ( see IM, 121) are included in the decision-making process. The value γο is determined according to the following relation, for the fuzzy center classifier:

Y ₀ = Yk (D -Yk (2) with

1 ti ~ H O) * -

1 + Z UXi C j) -Ck, «(JMZf) * i * l, ... XMAX

{Ck, i (i = 1 ... XMAX; k = 1 .... 2, XMAX: number of features / centroid of the element X _{t of} the feature vector with respect to class K), F (symmetric extension of the Xi (i = 1 ... XMAX)), B (the factor by which the membership value at the class limit of the mode value has decreased) and D (Pai ameter for the degree of fuzziness)) and the separation level classifier:

Yo = X ^T * WBA

{X: current measured value vector (feature vector), W: normal vector of the dividing plane, B: threshold component and A:

Ruck Transfer Zone}.

If contradictory information of the subsystems involved in the determination of the optimal control vector makes it impossible to make a definite statement about the process state and consequently control, the control module will output a control proposal via the dialog module in conjunction with the alert / request of the operator to evaluate this proposal (rejection / acceptance ).

The operator is always able to interrupt the automatic output of the control vector indefinitely, so that the system operates in off-line mode without modifying the control vector.

The invention is illustrated by the following examples:

example 1

The yeast strain Candida maltosa was grown on sucrose in a 10 L fermenter aerobically and continuously. The sucrose was fed to the mold gate in aqueous 5% solution. The aqueous nutrient brine contained per liter

8,000 mg N as NH ₄ Cl 800 mg K as K ₂ CO ₃ 800 mg P as H ₃ PO ₄ (85%) 160 mg Mg as MgSO ₄ -7 H ₂ O

and Spurensalzverbindungen in sufficient quantity. The temperature was 32 ⁰ C, the pH 4.2. The aeration rate was 501 / kg h. The residence time was 4 h. After a partial synchronization by a time regime (cf.

WP DD 219036) was controlled by the method according to the invention.

Process variables were continuously monitored and monitored by the system as follows: temperature, agitator, air mongo, masso; pH word, Geiöstsauorstoffkonzentration and biomass concentration in the fermentor, CO, - and O ₁ -GoMaIt the exhaust air, Durchflußmonge dor substrate and drain pump. Furthermore, as an off-line anolysonoroid gravimetrically determined HTS value and the substrate concentration in the fermentor as well as the carbon and oxygen balance integrated in the knowledge base III were included in the determination of the optimal control.

Dio Wisson's Theory of Optimal Control Constraint was also able to refer back to model simulation results, using a three-dimensional differential-locked system (see BLEY et al., Stud, hiophys., 98 (1983)).

119), as well as taking into account the information about situational recognition.

The parameters of the Gaussian filter used were - R = 12 H = 4 Z = 3 V = 6-8

and the parameters of the interface classifier were j

S = 2 N = 0.3

Despite occurring process disturbances, the degree of synchronization of the population of 0.26 (Boim time regime) could be increased to 0.35. ;

The process was characterized by rapid achievement of stable periods after these process disturbances. In dan stabilized phases, the yield coefficient was Y = 0.55.

Example 2

The extremely thermophilic bacterial population TP5 / 85 ZIMET 11093 was grown in a fermenter of 41 net volumes

continuously fermented in the chemostat for biomass formation.

The medium had the following composition:

(NH ₄ I ₂ SO ₄ 6.0 g / i KH ₂ PO ₄ 2.1 g / i MgSO ₄ χ 7H ₂ O 1.1 g / i yeast extract 0.5 g / i glucose 30.0 g / i CuSO ₄ X 5H ₂ O 0.14 g / i MnSO ₄ χ 4H ₂ O 0.12 g / i ZnCl ₂ 0.07 g / i CoSO ₄ χ 7H ₂ O 0.06 g / i FeCI ₃ 0.001 g / l

The temperature was 70 ° C, the pH was 6.9. The Saue ^r furnish was at a stirrer speed of 1250min 'and ensures a l.uftdrucksatz of 100l / kgh standard air.

Fermenting was carried out at an average flow rate of 0.83 h "'(n _max = 1.63 h ~') and a substrate residual concentration of 0.9-1.1 g / kg (K, = 2.2 g / l).

Since it can lead to fluctuations in the carbon substrate consumption in process disturbances, was on the

Metering pumps controlled the residence time. Furthermore, nitrogen imitations are possible by controlling the

Nitrogen substrate dosing must be overcome quickly.

The following process variables were constantly detected by the system and processed Lvw. monitors: temperature,

Stirrer speed, air quantity, mass; pH value, dissolved oxygen concentration and biomass concentration in the fermentor,

CO ₂ - and O ₂ content of the exhaust air, flow rate of the carbon substrate, nitrogen substrate and drain pump. Furthermore, as off-line analysis values, the gravimetrically determined BTS value, the glucose, phosphorus and nitrogen concentration in the fermentor and the carbon, nitrogen and oxygen balance integrated in the knowledge base III were used to determine the optimum control.

The knowledge base for determining the optimal control was constantly receiving the results of the model simulation using a standard three-dimensional system of differential equations (see BLEY, Th. Et al., Stud, biophys., 78 [1980111]).

was and the messages of the situation detection.

The parameters of the Gaussian filter were:

R = 10 H = 3 Z = 3 V = 7

The parameters for the fuzzy centroid classifier are: B = 0.5 D - 2 F = 0.6

The following process codes were achieved:

Productivity: r "= 10,95g / kgh

· Yield coefficient. Y _{x / S} = 0.45 g / g

Biomass concentration: X = 13.14g / l

Substrate concentration: S, = 0.8g / l

Claims (11)

A method for automatically controlling biotechnical material conversion processes by continuous or partially continuous growth of naturally occurring or genetically engineered cells on all usable C and energy sources, characterized in that a knowledge-based (ie based on elements of artificial intelligence) on-line Control system, which is structured in parallel real-time and dialog / consulting modules, each with a specific task spectrum and comprises three knowledge bases (or relatively independent components of a knowledge base), the knowledge base I as a superordinate dynamic knowledge base forms the communication basis of the two modules and the declarative knowledge for the characterization of all process variables and the current process state takes, the knowledge (rule) base-ll contains the rules and facts for fault detection and -Ursachenlokalisierung and the knowledge (rule) base ill un ter use of algorithmic software (situation detection, model simulation) to determine the optimal control vector is used, reliably detects process state changes and reacts flexibly in terms of optimal process management on this or determines the optimal period in processes with cyclically controlled command values. 2. The method according to claim 1, characterized by a real-time module of the following functions: - fast data acquisition and, if necessary, primary data processing, - Display and storage of data (measured values, controlled variables), - Disruption detection and determination on the basis of the appropriate knowledge base (II), Determination of the optimal control on the basis of the special knowledge base (III), whereby the inventive design of this knowledge base both the epigenetic and metabolic regulatory level of the microorganisms as well as short-term disturbances in the fermentation system, including the residence time of the microorganisms in the fermentor or the duration the phases in the cycle can be taken into account and - Output of the control vector. 3. The method according to claim!, Characterized by a dialogue / advisory module of the following functions: Functions enables or includes: - input and storage of the current process data (eg setpoints), - Input of off-line analysis values (with time of sampling), - control intervention by the operator (at any time), such as switching of the operating mode, technologically conditioned parameter changes (eg pump delivery rates, setpoint values) and modification of the model parameters or characteristics of the situation recognition, - Output of fault messages with indication of the possible cause of the fault (s). - graphic evaluations of the course of the process, - Consultation when changing the process state and - Conducting an active dialogue (computer request / operator response) in the case of a not definable by the control system process or reactor state. 4. The method according to claim 1 to 3 ,? characterized in that a multitasking or real-time operating system with at least three channels or an equivalent hardware-side solution is used to implement the control. 5. The method according to claim 1, characterized in that in the common dynamic knowledge base (I) the qualitative and quantitative knowledge of the individual process variables by the definition of composite objects is stored structured frame-like, the frame structure of the temporal validity of the information takes into account. 6. The method according to claim 1 and 5, characterized in that on the basis of a second specific knowledge or rule base (II), the procedural knowledge for processing contained in the common dynamic knowledge base current process information (current Meßwertvektor, process state) and contains technological as well as physical / technical information on the individual process variables, a comprehensive malfunction detection and -Ursachenlokalisierung is performed. 7. The method according to claim 1,5 and 6, characterized in that on the basis of a third specific knowledge base (III) determines the optimal control vector automatically or technologically or interference-related borderline cases (not clearly identifiable process state) the operator a reasoned proposal for The determination of the optimal control vector is based on the complex evaluation and interpretation of the statements obtained by the situation recognition and model simulation, including secured heuristic knowledge (empirical knowledge), which is stored in the form of rules and the evaluation of important to the specific process coordinated elemental biotechnical processes, such as carbon, nitrogen and oxygen biofuel. 8. The method according to claim 1 to 7, characterized in that in simulations for determining the optimal control as models structured models, preferably state models, are used, which describe the dynamics of a microorganism population especially in the epigenetic Reguldtionsebene sufficient and the statements of the model calculation of the according to the age of the result weighted in the decision making. 9. The method according to Anspnch 1 to 8, characterized in that as learning elements (subsystems) of the situation detection a prognosis filter and a classifier are used, wherein to achieve a continuous evaluation spectrum of statement quality (-safety) of these subsystems in addition to the actual yes / No decision regarding state changes of the process, a further variable evaluating this decision is included in the decision making process. 10. The method according to claim 10, characterized in that a Gaussian filter (cf., / 1 /, 121) is used as a prognosis filter, wherein the respective current value of the quotient <Q> acts as the decision of the Gaussian filter evaluating size. 11. The method according to claim 10, characterized in that a focal classifier (based on the fuzzy theory by Zadeh) is used as an adaptive classifier, wherein as the decision of the classifier evaluating size the current difference of the corresponding class membership values <Yd> in the decision making is included. The value <Yd> is determined according to the following relationship: