WO2008003575A2 - Measuring arrangement comprising a state monitoring system, and method for monitoring the state of a measuring arrangement - Google Patents

Abstract

The invention relates to a measuring arrangement comprising at least one sensor for detecting a measuring quantity and emitting a sensor signal dependent on the measuring quantity, and a data processing unit for evaluating the sensor signal. The data processing unit comprises a data memory which stores different types of operating data of the sensor, and determines values for at least two different state quantities of the measuring arrangement, each of said at least two state quantities containing values of at least two different types of operating data. The data processing unit then classifies a state of the measuring arrangement on the basis of the values of the at least two state quantities, and releases the result of the classification. The inventive method can be implemented in the measuring arrangement, but in principle is carried out outside the measuring arrangement in a data processing system.

Description

 Measuring arrangement with condition monitoring and method for condition monitoring of a measuring arrangement

The subject relates to a measuring arrangement with condition monitoring and a method for condition monitoring of a measuring arrangement.

At present, a variety of diagnostic functions are provided for measuring arrangements which comprise at least one sensor and a device for measuring value processing, for example a measuring transducer. For example, the diagnosis of measurement or state variables with the monitoring by means of warning and alarm limit values of a minimum and maximum value are state-of-the-art today. This ensures the monitoring of individual sensor properties of interest.

For the diagnosis of individual, for the function of the system very important attributes of the method of prediction based on simple mathematical models is known, such as in the patent DE 102 09 318 B4 entitled "Method for determining the wear-dependent residual life of an electrochemical measuring sensor "or the patent DE 101 41 408 B4 entitled" Method for determining the calibration interval time of electrochemical Messsensoreri "is disclosed. This allows the prediction of a time when a limit will be reached or exceeded.

The published patent application DE 10 2004 012 420 A1 entitled

"Monitoring device for the load of measuring probes by influences from the measuring environment" also takes into account the current characteristics of the measuring environment as well as the history of the process conditions.Thus, depending on a load model, an evaluation of the already loaded load of the measuring system is possible Derive statements about the remaining service life of the system.

Although the above approaches already to some extent

To enable diagnostic capability, an improved knowledge of the qualitative state of a measuring arrangement in the measuring mode is desirable.

In Feid, M .; Pandit, M .: "Systematic design of a signal-based

Fault Diagnosis System ", at - Automatisierungstechnik 54 (2006) 1, Oldenbourg or in Hengen, H .; Feid, M., Pandit, M .:" Monitors Learning Classification Methods at a Glance ", at - Automatisierungstechnik 52 (2004) 3, Oldenbourg Extensive statistical approaches for the diagnosis of sensor systems are discussed. The A special feature of these methods is that a comprehensive methodology for the feature search and then for feature selection is necessary.

In Schafer, R .; Jungmann, M .; Werthschützky, R .: "Self - diagnosis in the

Flow measurement by combination of differential pressure and vortex measurement ", tm - Technisches Messen 67 (2000) 9, Oldenbourg, it is shown, how the redundancy of the measured value acquisition and the availability of a measured value history can be used to perform a plausibility check of the measured value.

Based on the high demands on the quality, availability and safety of a measuring device and its measured value (e) a comprehensive diagnosis of the measuring device is necessary. In addition to the measured value, information is to be made available to a higher-level system or to a supervisor of the measuring device, which makes it possible to diagnose the sensor and the transmitter in detail and / or the entire measuring arrangement with regard to the stated requirements.

To meet this requirement, the measuring arrangement according to claim 1 and the method according to claim 8 is given according to the invention.

The diagnosis should include an assessment of the current state of the measurement arrangement and preferably a prognosis of the future state.

The evaluation of the current state relates essentially to

Availability, safety and quality, from which statements on the plausibility of the measured value and the credibility can be derived.

In the prognosis of a future state, the times at which a maintenance measure (calibration, cleaning, replacement of wearing parts, renewal of consumables or replacement of partial / total system) will be necessary with a certain probability of interest.

The measuring arrangement according to the invention comprises at least one sensor for

Detecting a measured variable, and outputting a measured variable-dependent sensor signal,

A data processing unit which evaluates the sensor signal, wherein

The data processing unit comprises a data memory in which different types of operating data of the sensor are stored,

The data processing unit values for at least two different ones

State variables of the measuring arrangement determines where in each of these at least two state variables in each case enter values of at least two different types of operating data, wherein

The data processing unit further based on the values of at least two State variables classified a state of the measuring device, and the result of the

Classification issues. The inventive method may be implemented in the measuring arrangement, but in principle also outside of the measuring arrangement in a separate

Data processing system to be performed. The operating data preferably go with a predetermined weight in the

State variables. In a presently preferred embodiment of the invention, the

State variables may be main components that have been determined by means of a principal component analysis. The operating data go in one by means of

Principal Component Analysis determined weighting into the major components. To carry out the principal component analysis can for an ensemble of

Measurement arrangements and their immediate process environment, operating data are included, which belong to a specific and interesting for diagnostic purposes state of a particular class. Such classes may, for example, be in need of maintenance (generally), needing calibration, polluted, drifting measured value, measured value dripped off, defective, etc. Operating data or characteristics of the measurement arrangements under consideration are, for example

• directly measurable quantities such as internal resistance, closed circuit electric power, and temperature

• Physically derivable quantities such as pH value, noise

• Mathematically derivable quantities from history such as mean values, variation, temporal trends, derivatives and

• other statistical and model-based quantities.

[0021-] Further operating data are calibration data such as zero point and span,

Load data of a sensor, for example pressure, vibrations, dirt load in the measuring medium, etc.

All data can be determined by the measuring arrangement or by means of auxiliary sensors and the measuring arrangement can be provided in a suitable manner.

Operating data for optical sensors are, for example:

Pollution degree of a process window

Age of an LED / photodiode or luminosity (of a reference signal)

[0029] Calibration characteristics Dynamic characteristics (time constant).

Operating data for amperometric sensors are, for example:

Membrane resistance

Current integral (electrolyte aging)

[0034] dynamic characteristics

Potential difference anode ref. -> layer structure at oxygen sensor

Load data (temperature, pH, pressure, CIP cycles ...) can also play a role in these sensors.

Result of the data acquisition is an extensive data table in the line by line, the measurement arrangements (sensors and Messunformer) and column by column the operating data or characteristics for the different measuring arrangements can be found. This representation can be viewed geometrically as a multidimensional space in which each measurement arrangement is characterized as a point. Both the geometric representation and the tabular representation contain the complete information.

To evaluate this data, a principal component analysis of the data is carried out, resulting in the example of two to four main components. A general presentation of the principal component analysis is, for example, in Backhaus, K .; Erichson, B .; Plinke, W .; Weiber, R .: "Multivariate Methods of Analysis An application-oriented introduction", 2006, Springer, given as a mathematical method for evaluation, for example, the so-called NIPALS method is suitable.

By means of the principal component analysis it is identified which operating data enter into the definition of a state of a measuring arrangement to what extent. Insofar as the number of main components is less than the number of types of operational data or characteristics, the number of dimensions for evaluating a measuring arrangement is reduced to a practical level by the principal component analysis.

As a result, the state of a measuring arrangement can be represented as a point on a plane or in, for example, a two to four-dimensional state space. The individual dimensions (main components) contain information of several characteristics or operating data and are thus no longer directly physically interpretable as in the complete representation.

This reduced representation depicts the considerable information content of the sum of the operating data for the evaluation of a measuring arrangement. operating data with little or no variation are ignored. If the sensors are coded according to their state in this reduced "diagnostic state space", it can be recognized whether these state maps reproduce themselves If this is the case, a simple diagnostic function can be implemented as a classifier in a sensor or a measurement arrangement with a transformation rule derived from the principal component analysis ,

The state classes can be diagnosed via the transformation and simple classifiers (polynomial of the main components).

In one embodiment of the invention, the prediction of times, from which the monitored sensor leave a certain state or from which he will assume a particular state. In most cases, the sensor states depend on several operating data, which are difficult to describe due to inherent physical and chemical relationships. By transforming the considered sensor into the space of the main components, all inherent relationships are mapped and the prediction becomes, with partly simple mathematical models such as e.g. linear regression possible.

In addition, a so-called diagnostic teaching is possible in this way. That is, when an in-use sensor is manually judged by the operator to be out of order due to experience or expertise, its characteristics can be measured and thus the data pool expanded. Thus, with powerful measuring devices, the acknowledgment of vague diagnostic messages and thus learning is possible. Sensor states in which the individual characteristics do not exceed the respective limit values; Thus, due to their internal relationships or history can be correctly evaluated.

The invention will now be explained with reference to an exemplary embodiment shown in the drawings.

It shows:

Fig. 1 is a schematic representation of some elements of a measuring arrangement according to the invention;

FIG. 2 shows a general structure of a feature data matrix for performing a multivariate analysis, in particular a principal component analysis; FIG.

FIG. 3 shows an example of a feature matrix for an ensemble of pH sensors for performing a principal component analysis; FIG. and

Fig. 4 shows the result of the principal component analysis for the ensemble pH sensors from Fig. 3.

As shown in FIG. 1, a measuring arrangement according to the invention comprises a

Sensor 1 and a data processing unit 2 with a data memory 3, in which the measured variable, and other operating data are stored. On the one hand, the operating data may be sensor data such as the measured variable or sensor temperature or sensor impedance, for example, or additional information such as calibration data (slope or zero point) or externally determined process data, for example the number of CIP cycles. From the operating data and data histories stored in the data memory, derived operating data, which are of interest for determining the sensor state, can be generated by means of a software module, which is designated as operator 4. This derived data may be, for example, differentiated or integrated sensor data or data describing a spectrum or the noise of the sensor. The operating data are then fed to a classifier 5, which determines the values of the main components based on the operating data with the respectively required, predetermined weightings (loads). Thus, the position of the sensor or the mass arrangement is determined in the state space, and the classifier can carry out the classification of the state based on the position in the state space.

Which operating data for the evaluation of the sensor state with which weighting are considerable is determined for a specific sensor type with a multivariate analysis, in particular principal component analysis on the basis of a sufficiently large ensemble of measuring arrangements of this sensor type.

For this purpose, several as "well" found sensors and several as "bad" found sensors measured in their characteristics, as shown in Fig. 2 in general and in Fig. 3 for the example of a pH sensor.

Without setting limit values for the individual features, a

Main component analysis is performed, which compresses the relevant information and disregards low-content features.

The result of the principal component analysis is shown in FIG. In the left part of Fig. 4 it can be seen that the two groups of sensors, namely, "good" and "poor" in the two-dimensional diagnostic state space spanned here, form again separate positions in the diagnostic state space via two main components. A classification according to "good" and "bad" as indicated by the ellipses, can thus in the measuring operation after Determination of the values for the two main components easily done. The weighting with which the individual sensor characteristics or operating data enter into the main components is shown in the right-hand part of FIG. 4.

In a further development of the invention, the predictive diagnosis is the

Changes in the position of a sensor in the diagnostic state space are evaluated by means of a trend analysis, and a prognosis is made as to when the position will leave the area of the current class or how long it will take to reach the area of another class.

The embodiment has been described with reference to FIGS

Main component analysis with two main components described. Of course, the diagnostic state space may also include more than two major components, for example, three or four. In principle, other methods of multivariate analysis can be used instead of the principal component analysis.

Claims

claims 1. Measuring arrangement, comprising at least one sensor for detecting a Measured variable, and outputting a measured variable-dependent sensor signal, a data processing unit which evaluates the sensor signal, wherein the data processing unit comprises a data memory in which values of different types of operating data of the sensor are stored, the data processing unit values for at least two different state variables of the measuring arrangement determined in each of these At least two state variables in each case enter values of at least two different types of operating data, wherein the data processing unit further classifies a state of the measuring arrangement based on the values of the at least two state variables and outputs the result of the classification. 2. Measuring arrangement according to claim 1, wherein the operating data received with a predetermined weighting in the values of the state variables. 3. Measuring arrangement according to claim 2, wherein the state variables Main components are those that have been determined by means of a principal component analysis, and enter into the values of the operating data in a weighting determined by the principal component analysis in the main components. 4. Measuring arrangement according to one of claims 1 to 3, wherein the operating data, directly measured variables and / or derived variables include. 5. Measuring arrangement according to one of the preceding claims, wherein the Operating data include sensor data, internal resistance, electrical energy flow, temperature, calibration data such as zero point and span pH, noise, averages, variation, timing trends, derivatives, and / or other statistical and model-based quantities. 6. Measuring arrangement according to one of the preceding claims, wherein the Operating data Load data of a sensor, such as pH value, pressure, temperature of vibration, dirt load in the measuring medium, and / or the number of cleaning cycles. 7. Measuring arrangement according to one of the preceding claims, wherein the Sensor is a potentiometric sensor, in particular a pH sensor, an amperometric sensor, in particular a gas sensor, or an optical sensor. 8. A method of classifying the condition of a sensor, comprising Steps: Acquire values for different operating data types of the sensor; Determining the values for at least two different state variables of the measuring arrangement, wherein in each of these at least two state variables values of at least two different types of operating data are received; and classifying the state based on the values of the at least two state variables. 9. The method of claim 8, wherein the operating data with a predetermined Weighting into the values of the state variables. 10. The method of claim 8 or 9, wherein the state variables Main components are those that have been determined by means of a principal component analysis, and enter into the values of the operating data in a weighting determined by the principal component analysis in the main components.