Method and analyzer for determining a measured value of a measured quantity in process automation technology

Abstract

Method for determining a measured value of a measurement quantity in process automation technology in a liquid or gaseous medium by means of an optical sensor, which has at least one transmitter (17.1) for sending transmitted light with at least two wavelengths (1-7), and a receiver (17.2) associated with the transmitter (17.1) for receiving received light, comprising the steps: - Taking a sample (13) of the medium (15), - Mixing the sample (13) with one or more reagents (16), - Applying an excitation signal to the transmitter (17.1) to generate the transmitted light, wherein the transmitted light is converted into the received light by interaction, in particular by absorption, with the mixed sample (13) depending on the measured quantity, - Generating a receiver signal using the receiver (17.2) from the converted received light, - Determining the measured value based on the receiver signal and a calibration function, characterized by the fact that The aging of the reagents is taken into account when determining the measured value. In particular, the calibration function includes a term that accounts for aging. the reagents taken into account, where the calibration function is created using a kernel method, where the kernel method is the Support Vector Machine or the Kernel Fisher discriminant, where the measured value c(x) is given by c ( x ) = a 0 + ∑ i = 1 N ai ⋅ k ( x , xi ) is calculated using the kernel function k ( x , xi ) = exp ( − μ | | xi − x | | 2 ) , with x a data vector comprising the receiver signal of at least two wavelengths, x i Support vectors, a0, ai coefficients, µ is a kernel parameter, and N is the number of support vectors as a natural number, in particular between 50 and 300.

Description

[0001] The invention relates to a method for determining a measured value of a quantity used in process automation technology in a liquid or gaseous medium by means of an optical sensor. The invention further relates to an analyzer for determining a measured value of a quantity, in particular for analyzing at least one ion concentration, especially an ammonium concentration.

[0002] For the purposes of this invention, the term "analyzer" refers to a measuring apparatus used in process automation technology that uses a wet chemical method to measure specific substance concentrations, such as the ion concentration in a medium to be analyzed. A sample is taken from the medium to be analyzed. This sample is usually taken fully automatically by the analyzer itself, for example, by means of pumps, hoses, valves, etc. To determine the concentration of a specific species, reagents specially developed for that substance and stored in the analyzer housing are mixed with the sample to be measured. The resulting color reaction of this mixture is then measured using a suitable measuring instrument, such as a photometer. More precisely, the sample and reagents are mixed in a cuvette and optically measured at different wavelengths using a transmitted light method.The measured value is determined on the receiver side based on light absorption and a stored calibration model.

[0003] Methods for determining the measured value of a measurand using an analyzer are known in the prior art. DE 10 2011 007 011 A1 describes an analyzer for the automated determination of a measurand in a liquid sample. Calibration standards are used for calibration. Calibration is performed by photometric absorption measurement.

[0004] CA 2 818 940 A1 describes a method for measuring at least one property of a predominantly liquid sample. The application includes a linear pH calibration equation as a calibration function.

[0005] Application US 2013 / 0265568A1 also relates to the analysis of samples. A primary purpose is the identification of solid (including powdered) and liquid materials with a fast measurement cycle time of approximately 2 to 15 seconds and with a method that does not require sample preparation. No indication of consideration for the aging of the reagents is disclosed.

[0006] The following difficulties arise when designing a corresponding calibration model: The relationship between absorption and substance concentration is not linear, and external disturbances influence this relationship. Fig. Figure 1 shows a typical curve of absorbed light as a function of substance concentration, exemplified here by ammonium concentration, for various wavelengths 1-7. It can be seen that this relationship is not linear, especially for higher concentrations. Furthermore, a user is interested in allowing the analyzer to run independently for as long as possible, i.e., without (re-)calibration, (re-)adjustment, or the replacement of reagents.

[0007] The invention is based on the objective of providing a calibration model that accurately reflects the relationship between absorption and substance content while simultaneously being robust against disturbing influences.

[0008] The problem is solved by a method comprising the following steps: taking a sample of the medium; mixing the sample with one or more reagents; applying an excitation signal to the transmitter to generate the transmitted light, wherein the transmitted light is converted into the received light by interaction, in particular by absorption, with the mixed sample, depending on the measured quantity; generating a receiver signal from the converted received light using the receiver; and determining the measured value based on the receiver signal and a calibration function. The method is characterized in that the aging of the reagents is taken into account when determining the measured value; in particular, the calibration function includes a term that considers the aging of the reagents. The aging of the reagents is identified as one of the main causes for a measured value that deviates from the true measured value.It is often the case that newer reagents are mixed with older reagents, making it impossible to accurately determine the age of the reagents. The method according to the invention allows the measured value to be determined correctly even with older reagents. Readjustment of the analyzer is not necessary.

[0009] In the Fig. 2a-c uses a simple calibration model; more precisely, piecewise linear interpolation between certain points was used as the calibration model and stored in a table. Fig. Figure 2a shows the linear interpolation model for fresh reagents. Fig. 2b with 8-week-old reagents and Fig. Figure 3c shows 12-week-old reagents. The graph displays concentrations predicted by the linear calibration model as well as the true ammonium concentrations. Measurements were taken in 1 mg / L increments. The concentrations were predefined by a so-called sampler. This is a programmable device that automatically mixes samples of a predetermined, arbitrarily large ammonium concentration. These samples are then drawn into the cuvette of an analyzer and measured. The true measurement is marked with reference numeral 21, and the interpolated measurement in a table is marked with reference numeral 20. It can be seen that the accuracy of the model decreases significantly across the entire measurement range with increasing reagent age. Furthermore, it is observed that with older reagents, the model becomes completely insensitive above a certain concentration.

[0010] According to the invention, the calibration function is created using a kernel method.

[0011] According to the invention, the kernel method is either the Support Vector Machine or the Kernel Fisher Discriminant. These two methods make it possible to determine the calibration function in such a way that the aging of the reagents is taken into account.

[0012] According to the invention, the measured value c(x) is determined by c(x)=a0+∑i=1Nai⋅k(x,xi) calculated using the kernel function k(x,xi)=exp(−μ||xi−x||2), with x a data vector comprising the receiver signal of at least two wavelengths, x i Position vectors, a0, a i coefficients, µ a kernel parameter, and N the number of support vectors as a natural number, in particular between 50 and 300.

[0013] The data vector includes at least one of the following parameters: ambient temperature of the analyzer, temperature inside the analyzer, sampling time, procedure duration, reagents, reagent age, and / or reagent mixing ratios. These parameters influence the calibration function and can be taken into account by using a kernel method to determine it.

[0014] In an advantageous embodiment, the support vectors, coefficients and kernel parameters are determined in advance.

[0015] The reagents should preferably not be replaced for at least 10 weeks, ideally 12 weeks. This saves the user time and money.

[0016] In a preferred embodiment, the method is carried out in an analyzer for analyzing at least one ion concentration, in particular the ammonium concentration.

[0017] The task is further solved by an analyzer designed to perform a procedure as described above.

[0018] In an advantageous embodiment, the analyzer comprises a higher-level unit, in particular a transmitter, wherein the higher-level unit performs the method described above.

[0019] Preferably, the higher-level unit performs the calculation of the kernel function and the measured value.

[0020] The invention is explained in more detail with reference to the following figures. They show Fig. 3 the analyzer according to the invention, and Fig. 4a-c substance concentrations determined using the method according to the invention.

[0021] The method according to the invention is used in an analyzer 9, which will first be described.

[0022] An analyzer 9 measures specific substance concentrations, for example the ion concentration in a medium to be analyzed, using wet chemical methods. An analyzer 9 according to the invention measures the ammonium concentration. Other ions to be measured include, for example, phosphate, nitrate, etc.

[0023] For this purpose, a sample 13 is taken from the medium 15 to be analyzed. Usually, the sample 13 is taken fully automatically by the analyzer itself, for example by subsystems 14 such as pumps, hoses, valves, etc. To determine the concentration of a specific species, reagents 16, specially developed for the respective concentration and stored in the analyzer housing, are mixed with the sample 13 to be measured. This is done in Fig. Figure 3 is symbolic; in reality, different containers with various reagents are provided and dispensed via the pumps, hoses, valves, etc. mentioned, and possibly mixed. Separate pumps, hoses, and valves can also be used for each step (taking the sample, mixing reagents, etc.).

[0024] The resulting color reaction of this mixture is then measured using a suitable measuring device, for example, a photometer 17. For this purpose, sample 13 and reagents 16 are mixed in a cuvette and optically measured using a transmitted light method with at least two different wavelengths. Light of at least two wavelengths 1-7 is transmitted through sample 13 by means of a transmitter 17.1. A receiver 17.2 is associated with the transmitter 17.1 to receive the transmitted light. The measured value is generated at the receiver based on the light absorption and a stored calibration function. The transmitter 17.1 comprises, for example, one or more LEDs, i.e., one LED per wavelength, or a corresponding light source with broadband excitation. The receiver 17.2 can comprise, for example, one or more photodiodes.

[0025] The analyzer 9 further comprises a transmitter 10 with a microcontroller 11 and memory 12. The analyzer 9 can be connected to a fieldbus via the transmitter 10. The analyzer 9 is also controlled via the transmitter 10. For example, the microcontroller 11 initiates the extraction of a sample 13 from the medium 15 by sending corresponding control commands to the subsystems 14. The measurement by the photometer 17 is also controlled and regulated by the microcontroller.

[0026] According to the inventive method, the aging of the reagents is taken into account when determining the measured value using light absorption and the calibration function. In particular, the compensation for aging is not achieved by means of a single, isolated mathematical term; rather, the aging compensation is incorporated into all coefficients of the calibration model.

[0027] The absorptions across the individual wavelengths 1-7 are considered together as a vector quantity, the data vector x, which here contains the individual absorptions as entries. However, other configurations are also possible, in which the vector x also includes measured quantities such as one of the parameters ambient temperature of the analyzer, temperature inside the analyzer, sampling time, process duration, reagents, and / or mixing ratios of reagents. It is often the case that newer reagents are mixed with older reagents, making precise age determination impossible. In many applications, it is desirable that the reagents remain unchanged for at least 10 weeks, and especially 12 weeks, and that the analyzer is not adjusted.

[0028] To account for the aging of the reagents, the calibration function is created using a kernel method. Fig. Figures 4a-c show that kernel methods can provide more accurate measurements than simple linear interpolation, despite increasing reagent age, as shown in Figure 20. Fig. Figures 2a-c illustrate this. The Support Vector Machine (SVM) and the Kernel Fisher Discriminant (KFD) are examples of such kernel methods. For clarity, only the KFD method is shown in the figures.

[0029] The greater computing and storage effort required for the methods can be accepted, since, as with the analyzer 9 described here, often only 1 measurement value is generated every 10 minutes.

[0030] The following is a brief overview of the basic principles of kernel methods.

[0031] Kernel methods transform data (in our case, the data vector x with absorptions for different wavelengths) using a nonlinear, multidimensional mapping, where the data in the new space can be linearly approximated. This multidimensional mapping does not need to be explicitly computed. Functions that satisfy a certain condition do so implicitly by calculating the dot product of two transformed vectors. In the literature, this is called the "kernel trick"—in this context, such functions are referred to as kernel functions.

[0032] All algorithms that can be represented by scalar products of data vectors can be transformed into a nonlinear version using such kernel functions. Well-known examples of such algorithms are the Support Vector Machine (SVM) and the Kernel Fisher Discriminant (KFD).

[0033] The trained model then looks like this: a new data vector x is assigned a substance content c(x), e.g. an ammonium content, by c(x)=a0+∑i=1Nai⋅k(x,xi) with the kernel function k(x,xi)=exp(−μ||xi−x||2), with x i as position vectors, a0, a i as coefficients, µ a kernel parameter, and N the number of support vectors as a natural number, in particular between 50 and 300, although this can also be larger or smaller.

[0034] The support vectors applicable to analyzer 9 x i , coefficients a0 and a i , and the kernel parameter µ are determined in advance, e.g. in the laboratory under laboratory conditions, i.e. at a specific, controlled temperature, humidity, air pressure, etc.

[0035] The coefficients a0 and a iThe parameters are determined during the training of the kernel method. In SVM, for example, "training" mathematically means solving a quadratic optimization problem. In KFD, training is performed by solving a regularized least-squares problem. KFD is equivalent to the so-called "Kernel Ridge Regression." Further details are known to those skilled in the art from the literature.

[0036] The kernel parameter µ is determined by dividing a training dataset into M equal parts, where M is a natural number. M-1 parts are then used for training. The kernel parameter is varied within a specific range until the trained model produces the smallest error on the omitted "foreign" data portion. This is repeated M times, each time with a different omitted data portion. Finally, the kernel parameter that produced the smallest error on average is chosen. This procedure is called cross-validation.

[0037] The method according to the invention is carried out by the analyzer 9. As already mentioned, the analyzer 9 comprises a transmitter 10, wherein the claimed method is carried out by the transmitter 10.

[0038] The support vectors, coefficients and kernel parameters are hard-coded in the firmware of Transmitter 10. Reference symbol list 1-7 1st-7th wavelength 9 Analyzer 10 Transmitter 11 microcontrollers 12 storage locations 13 Sample 14 subsystems out of 9 15 Medium 16 reagents 17 photometers 17.1 Channel 17.2 Recipients 20 Measured value through lookup table 21 True Measured Value 23 KFD a coefficients c substance content k Kernel function n Number of support vectors x data vector x i Support vector µ Kernel parameter

Claims

[1] Method for determining a measured value of a quantity used in process automation technology in a liquid or gaseous medium by means of an optical sensor comprising at least one transmitter (17.1) for sending transmitted light with at least two wavelengths (1-7), and a receiver (17.2) associated with the transmitter (17.1) for receiving received light, comprising the steps: - Taking a sample (13) of the medium (15), - Mixing the sample (13) with one or more reagents (16), - Applying an excitation signal to the transmitter (17.1) to generate the transmitted light, wherein the transmitted light is converted into the received light by interaction, in particular by absorption, with the mixed sample (13) depending on the measured quantity, - Generating a receiver signal using the receiver (17.2) from the converted received light, - Determining the measured value based on the receiver signal and a calibration function, characterized by , that The aging of the reagents is taken into account when determining the measured value. In particular, the calibration function includes a term that accounts for aging. the reagents taken into account, where the calibration function is created using a kernel method, where the kernel method is the Support Vector Machine or the Kernel Fisher discriminant, where the measured value c(x) is given by c(x)=a0+∑i=1Nai⋅k(x,xi) is calculated using the kernel function k(x,xi)=exp(−μ||xi−x||2), with x a data vector comprising the receiver signal of at least two wavelengths, x i Support vectors, a0, ai coefficients, µ is a kernel parameter, and N is the number of support vectors as a natural number, in particular between 50 and 300. [2] Method according to claim 1, wherein the data vector x further comprises at least one of the parameters ambient temperature of an analyzer (9), temperature in the analyzer (9), sampling time, process duration, reagents, reagent age, and / or mixing ratios of reagents. [3] Method according to claim 1 or 2, wherein the support vectors x i The coefficients a0, ai and kernel parameters µ are determined in advance. [4] Method according to at least one of claims 1 to 3, wherein the reagents are not replaced for at least 10 weeks, in particular 12 weeks. [5] Method according to at least one of claims 1 to 4, wherein the method is carried out in an analyzer (9) for the analysis of at least one ion concentration, in particular the ammonium concentration. [6] Analyzer (9) for determining a measured value of a measured quantity, in particular for analyzing at least one ion concentration, in particular an ammonium concentration, wherein the analyzer (9) is configured for carrying out the method according to at least one of claims 1 to 5. [7] Analyzer (9) according to claim 6, wherein the analyzer (9) comprises a superior unit (10), in particular a transmitter, wherein the superior unit (10) performs the method according to at least one of claims 1 to 5. [8] Analyzer (9) according to claim 7, wherein the superior unit (10) performs the calculation of the kernel function and the measured value.

Images