DE102013113170A1 - Method for determining a measured value and analyzer for carrying out the method - Google Patents

Abstract

The invention relates to a method for determining a measured value of a measured variable of the process automation technology in a liquid or gaseous medium by means of an optical sensor (17) comprising at least one transmitter (17.1) for transmitting transmitted light with at least two wavelengths (2, 3, 4, 5 ), and a receiver (17.2) associated with the transmitter (17.1) for receiving received light, comprising the steps of: applying to the transmitter (17.1) an excitation signal for generating the transmitted light, wherein the transmitted light (2, 3, 4, 5) is converted into the received light by interaction with the medium (15) as a function of the measured variable; Generating a receiver signal by means of the receiver (17.2) from the converted received light; Determining the measured value based on the receiver signal. The method is characterized in that an outlier (7) of the measured value is detected via a multivariate method for detecting outliers. The invention further relates to an analyzer (9) for carrying out the method.

Description

The invention relates to a method for determining a measured value of a measured variable of the process automation technology. The invention further relates to an analyzer for carrying out the method.

For the purposes of this invention, the term "analyzer" is intended to mean a measuring apparatus of process automation technology which measures certain substance contents, for example the ion concentration in a medium to be analyzed, using a wet-chemical method. A sample is taken from the medium to be analyzed. Usually, the sample is taken fully automatically by the analyzer itself, for example by pumps, hoses, valves, etc. To determine the substance content of a particular species to be determined, reagents specially developed for the respective substance content and stored in the analyzer housing are mixed with the sample to be measured. A resulting color reaction of this mixture is then measured by means of a suitable measuring device, for example by means of a photometer. More specifically, sample and reagents are mixed in a cuvette and optically measured at different wavelengths by transmitted light. On the basis of the light absorption and a stored calibration model, the measured value is determined on the receiver side.

In addition to the reaction cycle described above, several cycles of mixing and rinsing in the cuvette are necessary. As a result, the entire measuring cycle is slow, so only about every 10 minutes a measured value is obtained.

Due to the long-lasting measurements, incorrect measurements should be avoided. Incorrect measurements are caused, for example, by air bubbles in the cuvette. In general, a measurement error is a measurement that does not seem plausible for various reasons, so for example, a measurement that is not possible for physical reasons - such as a negative concentration. This is called an outlier.

Such outlier detection using smoothing (e.g., by a median filter) of the measurements themselves does not make sense. Because of the large time interval between two measured values, it is possible for strong changes in measured values to occur that falsely vanish after such smoothing. In addition, such a smoothing filter always needs a certain settling time, which additionally falsified the data.

The invention has for its object to reliably detect outliers of a series of measurements with a large time interval between the individual measuring points.

The object is achieved by a method by means of an optical sensor having at least one transmitter for transmitting transmitted light having at least two wavelengths, and a receiver associated with the transmitter for receiving received light, comprising the steps of: applying an excitation signal to the transmitter to generate the Transmitted light, wherein the transmitted light is converted by interaction with the medium depending on the measured variable in the receiving light; Generating a receiver signal by means of the receiver from the converted received light; and determining the measured value based on the receiver signal. The method is characterized in that an outlier of the measured value is detected by a multivariate method for detecting outliers.

Thus, it is even possible to reliably detect the outlier. In an advantageous embodiment, the multivariate method comprises determining the Mahalanobis distance, wherein an outlier has a Mahanalobis distance that is greater than a permissible maximum value.

berechnet, mit x einem Datenvektor umfassend die zumindest zwei Wellenlängen, µ dem Erwartungswertvektor des Datenvektors, und Σ der Kovarianzmatrix vom Datenvektor x.According to a preferred embodiment, the Mahanalobis distance d through

calculated, with x a data vector comprising the at least two wavelengths, μ the expected value vector of the data vector, and Σ the covariance matrix from the data vector x.

In an alternative development, the formula can also d (x) = (x - μ) Σ ^-1 (x - μ) ^T be used.

Both formulas are relatively easy to implement. In addition, low storage and computing capacity is needed. The Mahanalobis distance is a robust and reliable way to detect outliers. For both formulas mentioned results in an equivalent result as far as the outlier detection is concerned.

As an alternative to the Mahanalobis distance, a "nearest neighbor" classification, such as the k-nearest neighbor algorithm, can be used.

According to an advantageous development, the method is used in an analyzer for the analysis of at least one ion concentration, in particular the ammonium concentration, carried out, wherein reagents are used for the analysis. Alternatively, for example, phosphate or nitrate ions are determined.

Preferably, the process is carried out every 10 minutes.

In one embodiment, the data vector further comprises at least one of analyzer ambient temperature parameters, analyzer temperature, sampling time, processing time, reagents, reagent age, and / or reagent mixing ratios. This allows even more accurate detection of outliers.

In one embodiment, the determination of the measured value is aborted if an outlier is detected. This avoids unnecessary measurements, thus saving time and money.

The object is further achieved by an analyzer for analyzing at least one ion concentration, in particular the ammonium concentration, wherein the analyzer is designed for carrying out the method according to at least one of the aforementioned embodiments.

According to an advantageous development of the analyzer comprises a data processing unit, in particular a transmitter, wherein the data processing unit executes the method according to at least one of the above embodiments.

In an advantageous embodiment, covariance matrix and expected value vector are stored in the higher-order unit. Covariance matrix and expected value can be created in advance in a design under laboratory conditions, so that the user can rely on it.

The invention will be explained in more detail with reference to the following figures. Show it

1 an analyzer according to the invention,

2 a typical reaction course in a measurement with multiple wavelengths, and

3 a multivariate distribution function.

The method according to the invention finds application in an analyzer 9 , An analyzer 9 uses wet-chemical methods to measure certain substance contents, for example the ion concentration in a medium to be analyzed. An inventive analyzer 9 measures the ammonium concentration. Other ions to be measured are, for example, phosphate, nitrate, etc..

From the medium to be analyzed to a sample 13 from a medium 15 taken. Mostly the sample 13 fully automatically by the analyzer itself, for example by various equipment 14 such as pumps, hoses, valves, etc. taken. To determine the substance content of a particular species to be determined, reagents specially developed for the respective substance content and stored in the analyzer housing are used 16 with the sample to be measured 13 mixed. This is in 1 symbolically represented, in reality, different containers are provided with different reagents and removed via the addressed pumps, hoses, valves, etc. and optionally mixed.

A resulting color reaction of this mixture is then by means of a suitable measuring device, for example by means of a photometer 17 , measured. More precisely, sample 13 and reagents 16 mixed in a cuvette and optically measured with at least two different wavelengths in the transmitted light method. This will be light of at least two wavelengths 2 . 3 . 4 . 5 through a transmitter 17.1 to the test 13 Posted. The transmitter 17.1 a recipient is assigned 17.2 to receive the transmitted light. On the basis of the light absorption and a stored calibration model, the measured value is thus generated on the receiver side. The transmitter 17.1 is for example one or more LEDs, ie one LED per wavelength or a corresponding light source with broadband excitation. The recipient 17.2 can be about a photodiode.

The analyzer 9 further includes a transmitter 10 with a microcontroller 11 including memory 12 , About the transmitter 10 can the analyzer 9 be connected to a fieldbus. Next is the analyzer 9 over the transmitter 10 controlled. For example, taking a sample 13 from the medium 15 through the microcontroller 11 by appropriate control commands to the equipment 14 causes. Also, the measurement is made by the photometer 17 controlled and regulated by means of the microcontroller.

In 2 is a typical course of the received light signal during a multi-wavelength response cycle 2 . 3 . 4 . 5 to see. For example, during the reaction time, the minimum for each wavelength becomes 2 . 3 . 4 . 5 certainly. Other typical values are also possible. All minima are then further processed to a measured value with the aid of a calibration model.

If one now wants to decide on the basis of these minima or also on the basis of other sampling points during the course of the reaction, how reliable the current measurement is, the multivariate method according to the invention is used.

This is in 3 displayed. In the distribution function, two parameters are exemplarily plotted, for example, two different wavelengths. Other possible parameters include analyzer ambient temperature, analyzer temperature, sampling time, process time, reagents, regeneration age, and / or reagent mixing ratios.

In 3 is next to the reliable data 8th an outlier 7 to recognize, which is visible only by this type of application. The method according to the invention can recognize this outlier and is thus a method which evaluates how probable the occurrence of a particular data vector (ie the outlier 7 ) is against the background of the already observed data (ie the reliable data 8th ).

Such a method is based for example on the "nearest neighbor classification" (English: nearest neighbor method), for example by the k-nearest-neighbor algorithm.

Another method involves determining the Mahalanobis distance. Let x be a data vector whose entries are the minima mentioned above, but also the other parameters already mentioned above. Furthermore, let μ be the expected value of as many and representative data as possible that does not contain any outliers, such as the reliable data 8th , Finally, let Σ be the covariance matrix of the same data. Then the Mahalanobis distance d is defined as

As an alternative formula can d (x) = (x - μ) Σ ^-1 (x - μ) ^T be used. Although this does not give the same values as in the top formula, the results for outlier detection are equivalent.

If the Mahalanobis distance of the current data vector x exceeds a certain maximum value, the current measurement is considered as an outlier.

The data vector x can assume different forms. For example, one can only take sample points from a very early point in the course of the reaction. So you can see the outlier very early and stop the measurement and repeat if necessary.

The covariance matrix and the expected value are calculated using representative data. Thus, for example, the covariance matrix and the expected value can be created in advance under laboratory conditions, so that it is ensured for the user that these two represent reliable data. In the laboratory, n best possible and representative data vectors x _{i are} collected, where n is sufficiently large, on the order of at least 100 times more than the dimension of the data vectors).

Covariance matrix and mean are fixed in memory 12 the transmitter 10 deposited. The Mahanlobis distance is relatively easy to implement by the described formula. In addition, low storage and computing capacity is needed. Determining the Mahanalobis distance has proven to be a robust way to detect outliers.

LIST OF REFERENCE NUMBERS

1

 reaction course

2

 red

3

 yellow

4

 green

5

 blue

6

 Multivariate distribution function

7

 Runaway

8th

 Reliable data

9

 analyzer

10

 transmitter

11

 microcontroller

12

 Storage

13

 sample

14

Equipment for 9

15

 medium

16

 reagents

17

 photometer

17.1

 transmitter

17.2

 receiver

d

 Mahanalobis distance

t

 Time

TR

 reaction time

x

 data vector

Σ

 covariance matrix

μ

 expected value

Claims (10)

Method for determining a measured value of a measured variable of the process automation technology in a liquid or gaseous medium by means of an optical sensor ( 17 ), the at least one transmitter ( 17.1 ) for transmitting transmitted light with at least two wavelengths ( 2 . 3 . 4 . 5 ), and one the transmitter ( 17.1 ) associated receiver ( 17.2 ) for receiving received light, comprising the steps of: - loading the transmitter ( 17.1 ) with an excitation signal for generating the transmitted light, wherein the transmitted light ( 2 . 3 . 4 . 5 ) by interaction with the medium ( 15 ) is converted into the received light as a function of the measured variable, - generating a receiver signal by means of the receiver ( 17.2 ) from the converted received light, - determining the measured value on the basis of the receiver signal, characterized in that an outlier ( 7 ) of the measured value is detected via a multivariate method for detecting outliers. The method of claim 1, characterized in that the multivariate method comprises determining the Mahalanobis distance, an outlier having a Mahanalobis distance greater than an allowable maximum value. The method of claim 2, wherein the mahanalobis distance d is through

is calculated, with x a data vector comprising the at least two wavelengths, μ the expected value vector of the data vector, and Σ the covariance matrix of the data vector x. Method according to at least one of claims 1 to 3, wherein the method is used in an analyzer ( 9 ) is carried out for the analysis of at least one ion concentration, in particular the ammonium concentration, using reagents for the analysis. The method of claim 4, wherein the method is performed every 10 minutes. The method of at least one of claims 4 or 5, wherein the data vector further comprises at least one of analyzer ambient temperature parameters, analyzer temperature, sampling time, method duration, reagents, reagent age, and / or reagent mixing ratios. Method according to at least one of claims 1 to 6, wherein the determination of the measured value is aborted when an outlier is detected. Analyzer ( 9 ) for the analysis of at least one ion concentration, in particular the ammonium concentration, wherein the analyzer for carrying out the method according to at least one of claims 1 to 7 is configured. Analyzer ( 9 ) according to claim 8, wherein the analyzer ( 9 ) a data processing unit ( 10 ), in particular a transmitter, wherein the data processing unit ( 10 ) performs the method according to at least one of claims 1 to 7. Analyzer ( 9 ) according to claim 9, wherein covariance matrix and expected value vector are stored in the higher-order unit.

Images