DE10251192A1 - Method for monitoring of continuous technical process esp. for chemical reaction, involves on-line measurement of process variables such as concentration of educts and products at input and output of process - Google Patents

Abstract

A method for monitoring a technical process for carrying out a chemical reaction involves (a) on-line measurement of several process variables, especially the concentration of educts and products at the input and output of the continuous technical process, and arithmetic averaging over largely constant time intervals Delta T, (b) calculation of one or more process parameters, especially the throughput (turnover) by data validating for each time interval Delta T and using the measured (see a) process variables, (c) the process parameter calculated under (b) is used for regulating the technical process in which the process parameter serves as the regulating variable, (d) an actual accuracy of the process variables (measured under a) is measured in time intervals of 10 to 30 hours, by means of data validation. The calculated actual accuracy is used (e) as an instantaneous previous time interval Delta T2.

Description

The invention relates to a method for surveillance a continuous process engineering process to carry out a chemical reaction, in which process parameters, e.g. Sales and selectivity which serve as control variables, be calculated.

To monitor a process engineering process, continuous measurements of various process variables such as flow, concentration, temperature and pressure are carried out during operation. However, not all desired sizes are accessible for direct measurement or can only be measured with great technical effort. In addition, the measurements are fundamentally subject to both systematic and statistical measurement errors. If the measured variables are used to control or monitor the process, high demands are placed on the accuracy of the measurements, since the measurements otherwise only insufficiently describe the state of the process, which can lead to misinterpretations and thus incorrect control of the process. Regular maintenance and calibration of the measuring devices make it possible to achieve high accuracy. Despite the high accuracy, the measured values do not meet the mass and / or energy balance of the process. In order to be able to monitor the process engineering process on the basis of parameters or quality parameters such as sales and selectivity that cannot be measured directly, the measurement values must be adjusted when they are calculated so that they meet the mass and / or energy balance and, if necessary, other boundary conditions. This method of adapting measured values under boundary conditions is known as data reconciliation. An overview of the method of data validation can be found, for example, in CM Crowe, Data reconciliation - progress and challenges, J. Proc. Cont., Vol. 6, No. 2/3, 89-98, 1996. There are numerous commercial software products for the application of this method in process engineering processes, among others ADVISOR ^® from Aspentech, DATACON ^® from Simsci (Invensys group), RTO + ^® from MDC Technology, SIGMAFINE ^® from OSI SOFT and VALI ^® from Belsim.

When applying data validation on a process engineering process to carry out a chemical reaction using a process model which describes the chemical reaction, and a mathematical optimization algorithm estimates for measured Process variables determined, which meet the balance equations of the process model, with the measurement errors of the measured process variables minimized become. The solution of the optimization problem is to use a quality function, e.g. based on a To minimize the series of measurements under the boundary conditions of the process. The quality function is a maximum likelihood estimator in general. For the Case of a Gaussian The normal distribution of the errors is the quality function from the sum of the weighted Error squares formed.

It is therefore critical the failure behavior know the measurements to make an assumption about their distribution function and to be able to make their parameters. For this purpose, the a priori estimated Accuracy of the respective measuring instrument, which for example derived from empirical values or from manufacturer's information.

Furthermore, incorrect data is important for data validation Measuring device or a possible one Drift of the measuring devices due to aging, poor maintenance, irregular calibration, or the like. to consider.

In addition to adopting a suitable one Error distribution function is also the choice and parameterization of the process model used. As a process model, simple Balance models as well as complex procedural models are used become. The models can both stationary be dynamic as well. Depending on the type of process model used, i.e. depending on whether there are linear or nonlinear boundary conditions to the numerical solution linear or non-linear optimization algorithms of the optimization problem, such as. an SQP algorithm (SPQ: sequential quadratic program), used. In special cases is also an analytical solution of the optimization problem possible.

From the US 5,504,692 A method for data validation, in particular of flow measurements and pressure and temperature measurements, is known in particular, in which the weighting factors of the measurement errors decrease when the measurement errors are large. The measurement error of a signal is large if it either exceeds a certain limit or if it does not correspond to the statistical behavior of the errors from previous periods. For this purpose, the current measurement error is compared with the moving average of the historical measurement errors and the standard deviation used as weight is adjusted so that all current measurement errors are within a permissible range. This procedure eliminates outliers, although the definition of the permissible band, as in the US 5,504,692 described, is only reliable for measurement errors that are normally free of mean values.

A disadvantage of the application of data validation in process engineering processes according to the prior art is the insufficient known accuracy of the measured and estimated values of the measured process variables, e.g. flow, concentration, temperature and pressure or other variables dependent on the composition of the educt and product streams, such as pH -Value, density or the like, as well as the determined process parameters, for example conversion, selectivity, separation performance of an apparatus, in particular a distillation column, energy consumption, energy efficiency. One reason for this is the above-mentioned change in errors over time due to aging, calibration or poor maintenance.

Object of the present invention is a method of surveillance a process engineering process for carrying out a chemical reaction to provide, which does not have the disadvantages mentioned. In particular, a method is to be found by means of which process parameters, such as Sales and selectivity which as controlled variables in the process engineering process, determined with greater accuracy can be.

According to the invention, the object is achieved by the features of claim 1.

• a) Online-Messung mehrerer Prozessgrößen, insbesondere der Konzentration von Edukten und Produkten am Eingang und Ausgang des kontinuierlichen verfahrenstechnischen Prozesses, und arithmetische Mittelung über im Wesentlichen konstante Zeitintervalle ΔT₁ von 10 Sekunden bis 60 Minuten, wobei die Anzahl der gemessenen Prozessgrößen größer ist als die zur Beschreibung des Prozesses mit Hilfe eines Prozessmodells minimal notwendige Anzahl,
• b) Berechnung einer oder mehrerer Prozesskenngrößen, insbesondere des Umsatzes, aus den unter a) gemessenen Prozessgrößen mittels Datenvalidierung für jedes Zeitintervall ΔT₁ unter Annahme einer Fehlerverteilung der Messwerte und unter Verwendung des Prozessmodells sowie unter Verwendung einer momentanen Genauigkeit zur Gewichtung der gemessenen Prozessgrößen,
• c) Verwendung der unter b) berechneten Prozesskenngröße zur Regelung des verfahrenstechnischen Prozesses, wobei die Prozesskenngröße als Regelgröße dient,
• d) Berechnung einer aktuellen Genauigkeit in Zeitintervallen ΔT₄ von 10 bis 30 Stunden der innerhalb eines Zeitintervalls ΔT₂ von 10 bis 30 Stunden nach a) gemessenen Prozessgrößen mittels Datenvalidierung gleichzeitig für jedes Zeitintervall ΔT₃ von 1 bis 60 Minuten innerhalb des Zeitintervalls ΔT₂, wobei das Zeitintervall ΔT₄ größer oder gleich dem Zeitintervall ΔT₂ und das Zeitintervall ΔT₃ größer oder gleich dem Zeitintervall ΔT₁ ist und die gemessenen Prozessgrößen in dem Zeitintervall ΔT₃ arithmetisch gemittelt werden,
• e) Verwendung der unter d) berechneten aktuellen Genauigkeit als momentane Genauigkeit für die Gewichtung der gemessenen Prozessgrößen unter b).

The invention relates to a method for monitoring a continuous process engineering process, which comprises at least the following steps:

• a) Online measurement of several process variables, in particular the concentration of starting materials and products at the entrance and exit of the continuous process engineering process, and arithmetic averaging over essentially constant time intervals ΔT ₁ from 10 seconds to 60 minutes, the number of measured process variables being greater than the minimum number required to describe the process using a process model,
• b) calculation of one or more process parameters, in particular the turnover, from the process variables measured under a) by means of data validation for each time interval ΔT ₁ , assuming an error distribution of the measured values and using the process model and using an instantaneous accuracy for weighting the measured process variables,
• c) use of the process parameter calculated under b) to regulate the process engineering process, the process parameter serving as a controlled variable,
• d) calculation of a current accuracy in time intervals ΔT ₄ of 10 to 30 hours of the process variables measured within a time interval ΔT ₂ of 10 to 30 hours according to a) by means of data validation simultaneously for each time interval ΔT ₃ of 1 to 60 minutes within the time interval ΔT ₂ , wherein the time interval ΔT _{4 is} greater than or equal to the time interval ΔT ₂ and the time interval ΔT _{3 is} greater than or equal to the time interval ΔT ₁ and the measured process variables are arithmetically averaged in the time interval ΔT ₃ ,
• e) Use the current accuracy calculated under d) as the current accuracy for weighting the measured process variables under b).

The continuous process engineering process in the sense of the present invention serves to carry out a chemical reaction. It can be a single or multi-stage Act with or without feedback from Partial streams. In a preferred embodiment The invention relates to reactions that take place in a particular Response time incomplete expire. In a further preferred embodiment, runs alongside the desired Main reaction at least one undesirable side reaction, so that high selectivity based on the main reaction for the economics of the process is crucial. In a particularly preferred embodiment The invention of the process engineering process is the catalytic Oxidation of ethylene to ethylene oxide, in which the oxidation of Ethylene to carbon dioxide and water occur as an undesirable side reaction.

Which process variables are selected for online measurement depends on respective process, i.e. of the process of performing a chemical reaction, and that used to describe the process Process model. These are directly measurable quantities, with to which the state of the process can be described. decisive requirement for The data validation is that the number of measured quantities is redundant is, i.e. that the process is overdetermined based on the measurements is. Only the redundancy allows the measured values to be adjusted under Minimizing their mistakes. Preferably the concentration of educts and products at the entrance and exit of the process and / or the pH and / or the density and / or the pressure and / or the temperature and / or the Flow of the educt flows and product flows measured online.

With the help of the data validation, process parameters are calculated in accordance with the measured values arithmetically averaged in substantially constant time intervals ΔT ₁ from 10 seconds to 60 minutes, ie the calculation is carried out as soon as possible for the measurement of the process variables. These parameters are not directly measurable variables, but can only be determined on the basis of the measured process variables and a process model. Turnover and / or selectivity are preferably determined as process parameters. Alternative process parameters are, for example, the current efficiency of a separation apparatus, for example a distillation column, the degree of contamination of an apparatus or the energy efficiency of a process or an apparatus.

Before the measured values in the Data validation, they are checked for their validity, for outliers to eliminate. For this purpose, the measured values are based on a plausibility check an upper and lower limit and based on their change dependent on subject to time. Upper and lower limit values for permissible measured values result from operational experience, for example. Another possibility for outlier detection is a posteriori outlier detection, i.e. an outlier detection after implementation data validation. This is a comparison of the calculated measurement errors with the accuracies previously used to weight the measurement errors of the measuring devices performed. Is the calculated measurement error for a measured value significantly larger than that expected accuracy, this is a high probability an outlier and the data validation can be carried out again without this measured value.

wobei Φ die Gütefunktion, e_i,k der Messfehler der Messgröße i am Messgerät k, x_i der validierte Wert der Messgröße und x m / i,k der Messwert der Messgröße, σ _i,k die momentane Genauigkeit, h die Gleichungen des Prozessmodells und F die Prozesskenngrößen darstellt.The optimization problem of data validation can be described mathematically as follows (Equation 1):

where Φ the quality function, e _{i, k} the measurement error of the measured variable i on the measuring device k, x _i the validated value of the measured variable and xm / i, k the measured value of the measured variable, σ _{i, k} the current accuracy, h the equations of the process model and F the process parameters.

The current accuracy σ _{i, k} in the sense of the present invention depends on the error distribution of the measured process variables. In the case of a Gaussian normal distribution, the current accuracy is the standard deviation; in the case of a bimodal distribution, the current accuracy is described by the standard deviation, the outlier probability and the ratio of the standard deviation of measurement error and outlier (see also IB Tjoa and LT Biegler, Simultaneous strategies for data reconciliation and gross error detection of nonlinear systems, Comp. Chem. Eng., Vol. 15, No. 10, 679-690, 1991).

Hierin ist Φ die Gütefunktion, e_i,k der Messfehler der Messgöße i am Messgerät k, x_i der validierte Wert der Messgöße und x m / i,k der Messwert der Messgöße, σ _i,k die momentane Genauigkeit, h die Gleichungen des Prozessmodells und F die Prozesskennrgößen.Various statistical distributions, such as the frequently occurring Gaussian normal distribution or a bimodal distribution, can be used as an error distribution function. In the case of a Gaussian normal distribution, the data validation is adjusted using the least squares method, ie the sum of the weighted error squares is minimized as follows (Equation 2):

Herein Φ is the quality function, e _{i, k} the measurement error of the measured quantity i on the measuring device k, x _i the validated value of the measured quantity and xm / i, k the measured value of the measured quantity, σ _{i, k} the current accuracy, h the equations of the process model and F the process parameters.

The process model used is preferably a stationary one or dynamic mass balance or a stationary or dynamic mass and energy balance.

If a stationary process model is used, the frequency for calculating the process parameters is finished given the time constant of the process. As a measure of the time constant, for example serve the time it takes until the measured value of a process variable at the output of the process, e.g. of the reactor, after change an influencing factor, e.g. of the intake quantity, to half of the new stationary Value has risen.

The determined using data validation Process parameters used to control the process engineering process. there they serve as control variables by the determined value is treated as the actual value of the parameter.

In time intervals of ΔT ₄ of 10 to 30 hours, preferably 24 hours, a current accuracy of the process variables measured within a time interval ΔT ₂ of 10 to 30 hours is also simultaneously for each time interval ΔT ₃ of 1 to 60 minutes within the time interval ΔT by means of data validation ₂ , wherein the time interval ΔT _{4 is} greater than or equal to the time interval ΔT ₂ and the time interval ΔT _{3 is} greater than or equal to the time interval ΔT ₁ . For this purpose, the valid measured values are first arithmetically averaged within the time interval ΔT ₂ in intervals ΔT ₃ of 1 to 60 minutes. Subsequently, data validation takes place simultaneously for each time interval ΔT ₃ within the interval ΔT ₂ . The same process model and the same error distribution are used as in the data validation at a single point in time in step b) of the method according to the invention. Furthermore, the assumption is made that the current accuracy in the interval .DELTA.T _{2 is} constant, ie the period of time .DELTA.T ₂ is chosen so that the current accuracy of the measuring devices does not change significantly, for example not more than 0.5%. The current accuracy σ _{i, k} in the sense of the present invention is the standard deviation of the measured values in the case of the Gaussian normal distribution or, in the case of a bimodal distribution, the instantaneous accuracy is determined by the standard deviation, the outlier probability and the ratio of the Standard deviation of measurement errors and outliers described.

wobei wiederum Φ die Gütefunktion, e_ij,k der Messfehler der Messgröße i am Messgerät k zum Zeitpunkt j,x_ij der validierte Wert der Messgröße i am Messgerät k zum Zeitpunkt j, x m / i,j,k m Messwert der Messgröße i am Messgerät k zum Zeitpunkt j, σ_i,k die aktuelle Genauigkeit der Messrgöße i am Messgerät k und h die Gleichungen des Prozessmodells bezeichnet.The current accuracy is determined using the method according to the invention by means of data validation simultaneously in each interval ΔT ₃ within an interval ΔT ₂ . The optimization problem is as follows (Equation 3):

where in turn Φ the quality function, e _{ij, k} the measurement error of the measured variable i on the measuring device k at the time j, x _ij the validated value of the measured variable i on the measuring device k at the time j, xm / i, j, km measured value of the measured variable i on the measuring device k at time j, σ _{i, k} denotes the current accuracy of the measured variable i on the measuring device k and h denotes the equations of the process model.

In a first preferred embodiment The current accuracy thus determined becomes direct in the method as the current accuracy for weighting the errors in the data validation used after step b).

In a second preferred embodiment, the current accuracy σ _{i, k, new} for data validation according to b) by a moving average of the current accuracy σ _{i, k} and the previous current accuracy σ _{i, k, alt formed} as follows (Equation 4): σ i, k, new = aσ i, k + (1 - a) σ i, k, old

For the constant a applies: 0 <a <1.

The constant a can be chosen arbitrarily small. Your choice depends on the time interval ΔT ₂ . A is preferably chosen in the range from 0.01 to 0.5.

The advantage of the method according to the invention is that for monitoring process parameters not serving the process engineering process determined on the assumption of an accuracy estimated a priori Need to become. The Process parameters, for example Turnover and selectivity, can using the method according to the invention be calculated at least 1% more accurately than using the estimated a priori Accuracy, thereby regulating the process of a lesser Uncertainty is subject. In the case of the calculation of selectivity as a process parameter this means, for example, that the selectivity after the method according to the invention with 70.5% +/- 1% can be specified in comparison to the selectivity under Use of the a priori estimated Accuracy determined with 70% +/- 2% becomes. Since the process parameters as Control variables serve improves the method according to the invention the plant economy.

Because the current accuracy of each measuring device after the inventive method is determined from the measurements themselves, it reflects the current one Condition of the measuring device, i.e. its accuracy and the distribution of its measurements. On the other hand, the method according to the invention allows event-controlled Maintenance of the measuring devices used, since the current accuracy of a measuring device in relation to the intended accuracy of the device as an indicator of the maintenance condition can be used. Under the intended accuracy of a meter can be both the best instantaneous observed in the past Accuracy as well as the accuracy estimated a priori from the manufacturer's information be understood.

Another advantage of the invention lies in the fact that the maintenance effort for the calculation program, with which the data validation is carried out is reduced because of changes in the accuracy of measuring devices, e.g. due to exchange, calibration or failure, not done manually Need to become, but automatically recognized and optimally taken into account because they are direct flow into the data validation.

The inventive method is based on the in 1 illustrated schemes explained in more detail.

The process engineering process involves the catalytic oxidation of ethylene to ethylene oxide according to reaction 1: C ₂ H ₄ + ½O ₂ → C ₂ H ₄ O where as a side reaction the total oxidation of ethylene according to reaction 2: C ₂ H ₄ + 3O ₂ → 2CO ₂ + 2H ₂ O takes place.

With the help of data validation, the process parameters sales U _{1 of} the main reaction and sales U _{2 of} the side reaction are determined, which results in the selectivity S as follows (equation 5): S = U 1 / (U 1 + U 2 )

Knowledge of sales and selectivity allows the process to be controlled using selectivity as the control variable and the value calculated using data validation as the actual value considered the controlled variable becomes. This controls, among other things. the temperature control of the Reaction.

The process quantities of the four components ethylene, oxygen, ethylene oxide and carbon dioxide are measured at the inlet and at the outlet of the reactor with the help of a mass spectrometer at intervals of maximum 1 min and a gas chromatograph at intervals of maximum 8 min. The measured values are arithmetically averaged at intervals ΔT ₁ of 2 min. With a total of 4 measurements for each component is the process over-determined for determining sales. The reading 1 the measurement data is carried out by the data processing program 6 , which is installed on a process computer. Then the review takes place 2 of the measurement data instead of their validity, ie outliers are eliminated. To eliminate the outliers, the measured values are checked for upper and lower bounds. These barriers were specified manually from the operating experience of the plant operator, ie they reflect the measured values to be expected in all permissible operating states of the plant. All measured values are archived 3 ,

For data validation 4 According to b) and d), the stationary balance equations for the four components are used as the process model, taking into account the main and the secondary reaction (see reactions 1 and 2).

Both in data validation 4 after b) as well as after d) a Gaussian normal distribution is assumed, so that the quality function is formed by the sum of the weighted squares of errors (see equation 2).

The turnover U ₁ and U ₂ and the selectivity S are calculated by means of data validation according to b) at intervals ΔT ₁ of 2 min in a calculation program implemented directly on the process control system 6 , The calculated values are archived or displayed in the process control system 5 ,

For calculating 8th In terms of current accuracy, the optimization problem after step d) is solved once a day (ΔT ₄ = 24 hours) from the 14 last valid hourly mean values (ΔT ₂ = 14 hours; ΔT ₃ = 60 min) at the same time. The current accuracy is incorporated into the data validation after step b) using a moving average according to equation 4 with a = 0.5. This calculation 8th takes place on a process computer which is connected to the process control system by means of a communication program 7 connected is.

The process parameters conversion U ₁ and U ₂ and selectivity S could be calculated with this method with a 0.5 to 1% higher accuracy than using an a priori known accuracy of the measuring devices from the manufacturer's information.

Claims (9)

Method for monitoring a continuous process engineering process for carrying out a chemical reaction, comprising at least the following steps: a) Online measurement of several process variables, in particular the concentration of starting materials and products at the entrance and exit of the continuous process engineering process, and arithmetic averaging over essentially constant Time intervals ΔT ₁ from 10 seconds to 60 minutes, the number of measured process variables being greater than the minimum number necessary to describe the process using a process model, b) calculation of one or more process parameters, in particular the turnover, from the measurements measured under a) Process variables using data validation for each time interval ΔT ₁ assuming an error distribution of the measured values and using the process model and using an instantaneous accuracy to weight the measured process variables, c) Extension of the process parameter calculated under b) for controlling the process engineering process, the process parameter serving as a control variable, d) calculation of a current accuracy in time intervals ΔT ₄ of 10 to 30 hours of the measured within a time interval ΔT ₂ of 10 to 30 hours according to a) Process variables using data validation simultaneously for each time interval ΔT ₃ from 1 to 60 minutes within the time interval ΔT ₂ , the time interval ΔT _{4 being} greater than or equal to the time interval ΔT ₂ and the time interval ΔT _{3 being} greater than or equal to the time interval ΔT ₁ and the measured process variables in arithmetically averaged the time interval ΔT ₃ , e) using the current accuracy calculated under d) as the current accuracy for weighting the measured process variables under b). Method according to claim 1, characterized in that instead of the current accuracy used according to e) as the instantaneous accuracy for the weighting of the measured process variables under b), a moving average of the current accuracy from d) and the instantaneous accuracy from the respectively preceding time interval ΔT ₂ is used. Method according to one of claims 1 or 2, characterized in that that the continuous process engineering process is a chemical one Reaction is in which the starting materials are not fully implemented. Method according to one of claims 1-3, characterized in that that the continuous process engineering process is a chemical one Reaction with at least one side reaction. A method according to claim 4, characterized in that the continuous process engineering process is the catalytic Oxidation of ethylene to ethylene oxide with the oxidation of ethylene to carbon dioxide and water as a side reaction. Method according to one of claims 1-5, characterized in that the process variables measured under a) the concentration of starting materials and products at the entrance and exit of the process and / or the pH value and / or the density and / or the pressure and / or the temperature and / or the flow of the educt streams and product streams. Method according to one of claims 1-6, characterized in that that the process parameters calculated according to b) are sales and / or Selectivity and / or the current efficiency of a separator, especially one Distillation column, and / or the degree of contamination of an apparatus and / or the energy efficiency of a process or one Apparatus. Method according to one of claims 1-7, characterized in that the process model used under b) is a stationary mass balance or a dynamic mass balance or a stationary mass and energy balance or a dynamic mass and energy balance is. Method according to one of claims 1-8, characterized in that the error distribution assumed under b) is the Gaussian normal distribution or is a bimodal distribution.