METHOD FOR PROVIDING MEASURED VALUES OF A TECHNICAL PLANT, TECHNICAL SYSTEM AND METHOD FOR OPERATING THE TECHNICAL SYSTEM

Description

[0001] The present invention relates to a method for providing measured values ​​of a technical system, a technical system comprising the technical system, at least one measuring sensor and a program-controlled device, and a method for operating a technical system.

[0002] Technical systems, such as gas or wind turbines in power plants, are becoming increasingly complex. To ensure low-maintenance and efficient operation of such systems, monitoring and control systems are used. These systems employ sensors to measure data from the technical system, assess the operating status of the system (comprising the technical system and sensors) based on these measurements, and adjust operating parameters accordingly. For example, the optimal time to deactivate the technical system for maintenance, considering cost-effectiveness, can be determined based on the measured values, and the system can then be deactivated at that optimal time.

[0003] The assessment of the operating status of the technical system can be carried out manually by an operator of the technical system or automatically in a procedure carried out by a program-controlled device.

[0004] For example, US 2015 / 233730 A1 discloses a method for combining measurements of an aircraft's flight parameter from measurements of that parameter provided by a multitude of sensors.

[0005] Technical systems, such as gas or wind turbines, can be characterized by extreme conditions like high temperatures, pressures, or flow rates, which lead to increased error rates in the sensors used. Unreliable, faulty, or otherwise anomalous measurements can reduce the reliability of assessing the operational status of the system based on these measurements, potentially having a detrimental impact on its operation by altering its operating parameters.

[0006] EP 2 290 371 A1 discloses a calibration method for the prospective calibration of a measuring instrument. Each calibration point comprises a measurement signal from the instrument and a reference value corresponding to a reference measurement. Several possible slopes between the measurement points are determined. To avoid calibration errors, slopes outside a range defined by thresholds can be discarded.

[0007] EP 2 351 996 A1 discloses a method for determining at least one characteristic parameter for correcting measured values ​​from a Coriolis mass flow meter. The method comprises acquiring values ​​of a measured quantity, calculating at least one location parameter from the acquired values, and calculating at least one dispersion parameter from the acquired values ​​and the location parameter. These steps are repeated until the dispersion parameter reaches a threshold value. The measured values ​​are then corrected using the location parameter corresponding to the dispersion parameter.

[0008] Against this background, the present invention aims to propose a method for providing measured values ​​of a technical system that improves the reliability of the measured values ​​provided by the method.

[0009] The problem is solved by a procedure according to independent claims.

[0010] In particular, a method for providing measurement data from a technical system is proposed using a program-controlled device for carrying out the method. In this method, measurement data from at least one measurement series are acquired, with each measurement being provided by a sensor for a specific physical quantity in the technical system at a specific measurement time. The measurement data are categorized as normal or anomalous using a threshold comparison and at least one further processing stage. This further processing stage includes calculating several statistical location parameters for selected measurement data from one of the at least one measurement series and / or at least one statistical location parameter for selected measurement data from several measurement series.

[0011] The term "location parameter" here refers to a statistical parameter that more precisely describes the location, such as the center point, of a distribution comprising several measured values. One can speak of a "location" with respect to a data cloud.

[0012] An anomalous measurement, which may also be referred to as an unreliable measurement or outlier, is in particular a measurement which is assumed to be not related to the actual value of the associated physical quantity at the time of measurement in a way useful for assessing the operating condition of the technical system, due to a temporary or permanent anomaly, malfunction or failure of a measuring sensor.

[0013] The physical quantity being measured could be, for example, temperature, movement, vibration, pressure, or the like.

[0014] Each measurement series, for example, is a sequence of measured values ​​from a given sensor for a given physical quantity, preferably ordered according to measurement times. A sequence can also be ordered geometrically based on measurement locations. More than one measurement series can be recorded for a given physical quantity if redundant sensors are used. The measurement times can be chosen, for example, every hour, every minute, or every second. Irregular time intervals between measurement times are also conceivable.

[0015] Using the proposed method, obviously erroneous measurements can be categorized as anomalous by comparing threshold values. Furthermore, in a subsequent step of the procedure, measurements that are not obviously erroneous can be categorized as anomalous based on statistical criteria using the location parameter.

[0016] In further training, calculating one or more statistical location parameters for selected measurements from the same and / or different measurement series includes calculating several statistical location parameters for selected measurements from one of the measurement series and / or at least one statistical location parameter for selected measurements from several measurement series.

[0017] By calculating multiple position parameters for selected measurements from one of the measurement series, measurements that deviate from the multiple position parameters due to temporary sensor anomalies can advantageously be categorized as anomalous, while measurements that deviate from only one of the position parameters due to real physical transients can be categorized as normal. By calculating one position parameter for selected measurements from multiple measurement series, if redundant sensors are provided, individual measurements that deviate from the position parameter due to a fault in one of the multiple redundant sensors can be categorized as anomalous.

[0018] Categorizing measured values ​​as normal or anomalous increases their reliability. For example, when assessing the operating condition of a technical system, it is advantageous to consider only reliable measurements categorized as normal. This improves the system's operational safety because the adjustment of operating parameters is less influenced by anomalous measurements.

[0019] The task is further solved by a procedure for providing measured values ​​of a technical system, which includes carrying out a first process stage and a second and a third process stage.

[0020] The first process stage, the second process stage and the third process stage can be executed sequentially in any order or at least partially in parallel with each other.

[0021] The first stage of the procedure corresponds to the threshold comparison described above and below and includes recording at least one measured value from at least one series of measurements, comparing the at least one measured value with a predetermined threshold to generate a comparison result, and identifying the at least one measured value as a normal measured value or as an anomalous measured value of the first kind depending on the comparison result.

[0022] The first stage of the process advantageously allows measured values ​​to be identified as anomalous first-order measured values, which, based on given knowledge about the configuration of the technical system and / or about physical facts, are recognized as obviously erroneous measured values, such as negative temperatures in liquid water or negative pressures or other gross outliers that deviate significantly from the expected values.

[0023] The second process stage corresponds to the further process stage according to the embodiments described above or below and comprises: Acquiring multiple selected measurements from multiple measurement series, wherein the multiple selected measurements are provided by different measuring sensors for the same measurand and the same measurement time; calculating a statistical location parameter of the selected measurements; and for at least one of the acquired selected measurements: comparing the at least one measurement with the statistical location parameter and, if the at least one measurement deviates from the statistical location parameter by more than a predetermined relative deviation or a predetermined absolute deviation, identifying the at least one measurement as an anomalous measurement of the second kind.

[0024] In the second stage of the process, the statistical position parameter can be interpreted as a majority decision of redundant sensors regarding the assumed physically correct measured value. If a single measured value among several measured values ​​provided for the same quantity and the same measurement time deviates too much from the statistical position parameter, the measured value in question is identified as an anomalous. If a relative deviation is used as the criterion, the identification of such second-order anomalous measured values ​​can advantageously be performed without knowledge of the physical quantity being measured.

[0025] If an absolute deviation is used, existing knowledge about the physical quantity measured and the expected properties of the associated measuring sensor can be incorporated into the identification of anomalous measurements of the second kind by appropriately choosing the absolute deviation.

[0026] The third stage of the procedure can be part of the further stage of the procedure, as explained above and / or below, and includes: Acquiring a series of measurements, wherein the measurements are provided by a measuring sensor for the same measurand and different measurement times, and the measurements in the series are ordered chronologically; and for at least one of the acquired measurements in the series, each of the following steps: determining a first statistical location parameter and a first statistical dispersion parameter for a first predetermined number of measurements from the same series that precede the at least one measurement in time; determining a second statistical location parameter and a second statistical dispersion parameter for a second predetermined number of measurements from the same series that follow the at least one measurement in time; calculating a first quotient of the absolute value of the difference between the at least one measurement and the first statistical location parameter and the first statistical dispersion parameter;Calculate a second quotient from the absolute value of the difference between the at least one measured value and the second statistical measure of central tendency µf and the second statistical measure of dispersion; identify the at least one measured value as an anomalous measurement of the third kind if the first quotient is greater than or equal to a given first reference value and the second quotient is greater than or equal to a given second reference value, or as a normal measured value if the first quotient is less than the given first reference value or the second quotient is less than the given second reference value.

[0027] In the third stage of the process, the first quotient for a selected measurement describes the ratio of the selected measurement's deviation from a location parameter of a moving window of size (from measurements at least partially preceding the selected measurement in time, hereinafter referred to as the "preceding moving window") to a dispersion parameter of the moving window. This ratio thus represents a measure of the measurement's deviation from the location parameter of the distribution, normalized to the dispersion of the distribution of the at least partially preceding measurements.

[0028] The second quotient describes the ratio of the deviation of the selected measurement from a location parameter of a moving window of size (from measurements at least partially following the selected measurement in time, hereinafter referred to as the "following moving window") to a dispersion parameter of the moving window. This ratio thus represents a measure of the deviation of the measurement from the location parameter of the distribution, normalized to the dispersion of the distribution of the at least partially following measurements.

[0029] By comparing the quotients with predefined dimensionless reference values, measured values ​​that deviate too much from both the position parameter of the preceding sliding window and the position parameter of the subsequent sliding window, for example due to a temporary sensor anomaly, can be identified as anomalous measured values ​​of the third kind without knowledge of the physical measured quantity, while measured values ​​that deviate from only one of the two position parameters due to a physical transient of the measured quantity and are erroneously identified as anomalous measured values ​​by conventional smoothing with a centered sliding window can be identified as normal measured values.

[0030] The number of measurements in the preceding sliding window is preferably 5 to 200, particularly preferably 10 to 100, and most preferably 30 to 60. The number of measurements in the subsequent sliding window is preferably 5 to 200, particularly preferably 20 to 50, and most preferably 20 to 30.

[0031] The first comparison value is preferably greater than two and less than four, and is particularly preferably three. The second comparison value is preferably greater than or equal to two and less than four, and is particularly preferably three.

[0032] The identification of a measured value as a normal measured value or as an anomalous measured value of the second or third type according to the second or third stage of the procedure is included in the categorization step for a respective measured value and has the effect of further increasing the reliability of the provided measured values.

[0033] According to another embodiment, the statistical location parameter is a median, a mean, or a biweight mean; the first and second statistical location parameters are each a mean, a median, or a biweight mean; and the first and second statistical dispersion parameters are each a standard deviation, a mean absolute deviation, or a biweight standard deviation.

[0034] Preferably, the statistical location parameter is a median value, the first and second statistical location parameters are preferably means, the first and second statistical dispersion parameters are preferably standard deviations, and the second comparison value is most preferably two.

[0035] In embodiments, the third stage of the process further comprises identifying a respective measured value in the measurement series as noise if the first quotient is greater than or equal to a predetermined third reference value and the second quotient is greater than or equal to a predetermined fourth reference value. The predetermined third reference value is less than or equal to the predetermined first reference value, and the predetermined fourth reference value is less than or equal to the predetermined second reference value.

[0036] The third and fourth specified comparison values ​​are preferably each two.

[0037] Noise can be understood as an unreliable measurement value whose cause is not an anomaly, malfunction or failure of the associated measuring sensor, but which deviates excessively from the actual value of the associated physical quantity due to physical disturbances.

[0038] Excessively noisy measurement values ​​that are not due to a malfunction of the measuring sensor, but are not reliable enough to determine the actual value of the associated physical quantity, can be identified as noise, thereby further increasing the reliability of the provided measurement values.

[0039] In further embodiments of the method for providing measured values ​​of a technical system, identifying a measured value also includes changing the measured value if the measured value is not identified as a normal measured value.

[0040] By appropriately modifying a non-normal measured value, the reliability of the provided measured values ​​and thus the reliability of the assessment of the operating state of the technical system can be further improved.

[0041] In embodiments, modifying a measurement identified as anomalous or as noise includes marking the measurement with a label and / or removing the measurement from the series of measurements.

[0042] If measured values ​​identified as anomalous or as noise are flagged or removed from the measurement series, they can be disregarded when assessing the operational state of the technical system, thereby improving the reliability of the assessment. Removing such measured values ​​also reduces the amount of available measurement data, thus decreasing the computational effort required for data evaluation and assessment of the operational state of the technical system.

[0043] In embodiments, in the second process stage, if a measured value is identified as an anomalous measurement of the second type, the measured value identified as an anomalous measurement of the second type is removed from the multiple selected measured values, and the steps of the second process stage are repeated for the remaining selected measured values, provided that more than two selected measured values ​​remain.

[0044] If, during the majority decision according to the second stage of the procedure, a measurement is identified as an anomalous measurement of the second type, the statistical location parameter describing the majority decision is recalculated using only the remaining selected measurements, omitting the measurement identified as anomalous and therefore unreliable. This improves the reliability of the recalculated location parameter. The remaining selected measurements are then compared again with the more reliable, recalculated location parameter, potentially identifying another anomalous location parameter. These steps are repeated iteratively until either none of the remaining measurements are identified as anomalous, or only two measurements remain and a majority decision is no longer possible, in which case the last remaining measurements are identified as normal measurements.

[0045] Identifying a measurement that is identified as an anomalous measurement further includes storing classifying information about the severity of an anomaly, about a temporal correlation of the anomaly and / or about the multiple occurrence of the anomaly at different measurement sensors for the same measurement time.

[0046] An anomaly can be understood, in particular, as a situation that causes an anomalous measurement. Classifying information can thus be determined that allows conclusions to be drawn about the cause of an anomaly. These conclusions can be drawn by an operator or an automated system, such as an artificial neural network, and taken into account when assessing the operating status of the technical system, the technical equipment, and / or the measuring sensors.

[0047] In particular, classifying information about the severity of an anomaly can be determined and stored by the fact that the second stage of the procedure further includes: Classifying a measurement identified as an anomalous measurement of the second type as a severe anomaly if the measurement deviates from the statistical location parameter by more than a second relative or absolute deviation, or as a minor anomaly if the measurement identified as an anomalous measurement of the second type is not more than the predetermined second range of variation from the statistical location parameter. Here, the second deviation is either a predetermined second deviation that is larger than the predetermined first deviation, or a sum of the predetermined first deviation and a value determined from the selected measurements. Preferably, the second range of variation is the sum of the predetermined first range of variation and a quotient of the standard deviation of the selected measurements and the square root of the number of selected measurements.

[0048] Furthermore, the severity of an anomaly can be determined and stored, in particular, by the third stage of the procedure further comprising: classifying a measurement identified as an anomalous measurement of the third type as a severe anomaly if the first quotient is greater than or equal to a predetermined fifth reference value and the second quotient is greater than or equal to a predetermined sixth reference value, or as a less severe anomaly if the first quotient is less than the predetermined fifth reference value or the second quotient is less than the predetermined sixth reference value, wherein the predetermined fifth reference value and the predetermined sixth reference value are each greater than the predetermined first reference value and the predetermined second reference value, respectively, and preferably, for example, equal to four.

[0049] In particular, classifying information about a temporal correlation can also be determined and stored, so that the procedure further includes: Acquiring one or more series of measurements, wherein the measured values ​​are provided by one or more measuring sensors for the same measured quantity and different measurement times, and the measured values ​​in the one or more series of measurements are each ordered in time;and for each measured value from one of the measurement series that is identified as an anomalous measured value of the first, second, or third type, the classification of the respective measured value as a time-correlated anomaly if at least one further measured value for the measurement time preceding or following the measurement time of the respective measured value in the measurement series of the further measured value is also identified as an anomalous measured value of the same type, or as a time-isolated anomaly if none of the further measured values ​​for the measurement time preceding or following the measurement time of the respective measured value in the measurement series of the respective further measured value is identified as an anomalous measured value of the same type.

[0050] In particular, classifying information about multiple occurrences of an anomaly at different measuring sensors for the same measurement time can be determined and stored by the fact that the procedure further includes: Acquiring multiple selected measurements provided by different sensors for the same measurement quantity and at the same measurement time; determining the number of normal and anomalous measurements among the selected measurements; classifying the respective anomalous measurements among the selected measurements as a majority sensor anomaly if the number of anomalous measurements among the selected measurements is greater than a predefined maximum value or the number of normal measurements is less than a predefined minimum value, or as a specific sensor anomaly if the number of anomalous measurements is not greater than a predefined maximum value or the number of normal measurements is not less than a predefined minimum value.

[0051] In the aforementioned steps, classification may include storing classifying information, which for each anomaly includes information about the type of anomaly, a marker for temporally correlated or temporally isolated anomaly, a marker for severe or minor anomaly, and / or a marker for anomaly of a majority of sensors or anomaly of a specific sensor.

[0052] The classifying information according to the present embodiment is particularly advantageous in determining the causes of anomalies in measuring sensors.

[0053] The process of identifying and storing classifying information includes, for example, categorizing the measured values.

[0054] In further embodiments, the following steps are also carried out: Determine the number of measurements provided by a selected sensor and the number of anomalous measurements among those provided by the selected sensor; and deactivate the selected sensor if the proportion of anomalous measurements among those provided by the selected sensor exceeds a predefined acceptance value.

[0055] On the one hand, deactivating measuring sensors reduces the amount of data to be considered, and on the other hand, it reduces the energy consumption of all measuring sensors.

[0056] In further embodiments, only the subsequent process steps are carried out without a prior threshold comparison. In these cases, only a statistical analysis is performed, particularly using one or more location and / or dispersion parameters. Therefore, in alternative embodiments of the method, only the first, second, or third process step for categorizing the measured values ​​is performed.

[0057] In further embodiments, a technical system comprising a technical installation, at least one measuring sensor, and a program-controlled device is proposed, wherein the program-controlled device is configured to perform the previously or subsequently described method for providing measured values. The embodiments and features described for the proposed method apply accordingly to the proposed technical system.

[0058] Furthermore, a method for operating a technical system is proposed, comprising a technical installation, at least one measuring sensor, and a program-controlled device. This method includes carrying out the procedure according to the embodiment of the procedure described above or below, using measured values ​​from the at least one measuring sensor for at least one physical measurand in the technical installation. An operating parameter of the technical installation is thereby changed or controlled depending on the measured values ​​provided by the method for supplying reliable measured values.

[0059] Changing the operating parameters of the technical system can, in particular, include deactivating the system for maintenance. Since deactivation is performed with improved reliability based on the measured values ​​provided by the process, just-in-time maintenance can be advantageously implemented, thereby minimizing maintenance costs. The operating state of the technical system is thus defined based on the provided measured values. A closed-loop control system can also be established using the reliably provided measured values ​​to operate the system.

[0060] The technical equipment of the proposed system may in particular be a turbine, a compressor or a generator, especially a gas turbine or a wind turbine.

[0061] The maintenance procedure performed after deactivation can include an offline cleaning process of the guide vanes of a gas turbine. Alternatively, depending on the measured values ​​provided by the procedure, an online cleaning process can be initiated without deactivating the gas turbine.

[0062] The measuring sensor of the proposed technical system may, in particular, be a temperature sensor, a pressure sensor, a motion sensor or a vibration sensor.

[0063] Furthermore, a computer program product is proposed which, on a program-controlled device, causes the execution of the procedure according to one of the first to eighth embodiments.

[0064] A computer program product, such as a computer program tool, can be provided or delivered from a server on a network, for example, as a storage medium such as a memory card, USB stick, CD-ROM, DVD, or as a downloadable file. This can be done, for example, in a wireless communication network by transmitting the corresponding file containing the computer program product or tool.

[0065] Other possible implementations of the invention also include combinations of features or embodiments described previously or subsequently with regard to the exemplary embodiments, even if not explicitly mentioned. In such cases, the person skilled in the art will also add individual aspects as improvements or additions to the respective basic form of the invention.

[0066] Further advantageous embodiments and aspects of the invention are the subject of the dependent claims and the exemplary embodiments of the invention described below. The invention will be explained in more detail below with reference to preferred embodiments and the accompanying figures. Fig. 1 shows a schematic representation of a technical system with a technical installation that is set up to carry out a procedure for providing measured values ​​or measurement data. Fig. 2 shows a schematic representation of possible measurement data. Fig. 3 shows a schematic flowchart for a method for providing measured values ​​according to a first embodiment. Fig. 4 shows a schematic flowchart for the procedure according to a second embodiment. Fig. 5 shows a schematic flowchart for possible first, second and third process stages. Fig. 6 shows a flowchart of the steps of the second process stage. Fig. 7 shows the steps that are performed for a measured value according to the third procedure stage. Fig. 8 shows a representation of raw measurement data in a first case. Fig. 9 shows a representation of reliably provided measurement data. Fig. 10 shows a representation of raw measurement data in a second case. Fig. 11 shows a representation of reliably provided measurement data in the second case.

[0067] In the figures, identical or functionally equivalent elements have been given the same reference symbols, unless otherwise indicated.

[0068] Fig. 1 Figure 1 shows a schematic representation of a technical system 1 according to an embodiment, comprising a technical system 2, such as a gas turbine, three measuring sensors MS1-MS3, and a program-controlled device, such as an industrial computer 4. The measuring sensors MS1-MS3 are arranged inside the gas turbine 2 and connected to the industrial computer 4 via a wireless or wired connection 3 extending from the gas turbine 2. The industrial computer 4 is coupled to a display device 5 and a memory 6.

[0069] The measuring sensors MS1 - MS3 arranged inside the gas turbine 2 provide raw measurement data RMD to the industrial computer 4 via the connecting path 3.

[0070] The industrial computer 4 executes a computer program that initiates a procedure for providing measured values ​​of a technical system using the raw RMD measurement data provided by the temperature sensors.

[0071] In one variant, the industrial computer 4 can also be coupled via a (not shown) feedback line to a (not shown) control and / or regulation device for the gas turbine 1. In this variant, the industrial computer 4 further executes a computer program that initiates a procedure for operating the technical system 1 and, in particular, for changing an operating parameter of the gas turbine 2 depending on provided reliable measured values.

[0072] Fig. 2 This shows a schematic representation to illustrate how measurement data can be acquired and made available. Fig. 2 The horizontal axis represents the time as the number of each measurement point, and the vertical axis represents the sensor number. Vertically arranged, m measurement series for m redundant sensors each contain n measured values ​​x for n measurement points of the same physical quantity. In this description, the measurement data are denoted by the equation symbol xs,t, where s and t are integers, s = 1 ... m, and the values ​​in Fig. 2 The number of the respective sensor is plotted on the vertical axis, and t = 1 ... n, and the in Fig. 2 Number plotted on the horizontal axis of the respective

[0073] The measurement time is denoted by , where larger t represents later measurement times and smaller t represents earlier measurement times. In other words, the n measured values ​​xs , t of a measuring sensor MSs with the number s for t=1 ... n are arranged in temporal sequence of the measurement times of the respective measured values ​​of the measuring sensor MSs.

[0074] The measurement data from further measurement series can be recorded and schematically represented in the same way using measured values ​​y, z, ... for one or more additional physical quantities. For the sake of simplicity, the following description considers only measured values ​​x of the same physical quantity.

[0075] The measurement data is stored, for example, as a packet data stream, as a vector, as an array, as a linked list, or the like.

[0076] The term measured value refers, for example, to a data set that includes at least one numeric measurement data field, such as a floating-point value.

[0077] In variants, the data record of a measurement can include further identifying information, which may include numeric data fields, for example an integer value and / or binary values, such as an integer identifier indicating the sensor number of the sensor that provided the measurement, or a binary identifier indicating whether the measurement is a normal or anomalous measurement, a binary identifier indicating whether an anomalous measurement is a time-correlated or a time-isolated anomalous measurement, a binary identifier indicating whether an anomalous measurement is a severe or a minor anomaly, and / or a binary identifier indicating whether an anomalous measurement is due to an anomaly of a majority of sensors or an anomaly of a specific sensor.In some embodiments, the data set of a measured value may include a pointer that points to another data set in which characteristic information as described above is stored.

[0078] The following are examples of a method for providing measured values ​​from a technical system.

[0079] Fig. 3 Figure 1 shows a schematic flowchart of a procedure according to a first embodiment. The procedure comprises six steps S0 to S6, which are executed sequentially.

[0080] In step S0, the measuring sensors MS1...MSm in gas turbine 2, as shown in Fig. 1 As indicated, raw measurement data (RMD) is provided. The raw measurement data (RMD) includes measured values ​​for physical quantities from the respective measuring sensors MS1...MSm.

[0081] In step S1, RMD measures 7 are recorded from the provided raw measurement data, which include measurement values ​​xs,t for a physical measurement quantity x in m measurement series.

[0082] In step S2, individual measured values ​​xs,t are compared with a predefined lower threshold and a predefined upper threshold. Measured values ​​that are less than the predefined lower threshold or greater than the predefined upper threshold are categorized as anomalous. The remaining measured values ​​are categorized as normal.

[0083] In step S3, a further procedural stage is carried out, which includes the calculation of one or more statistical location parameters.

[0084] In one variant, step S3, for example, determines a statistical location parameter, such as a median value, for a selection of measured values ​​from several measurement series of multiple sensors at the same time t. The selected measured values ​​are compared sequentially with the median value and identified as anomalous if they deviate from the median value by more than a predetermined absolute or relative deviation. Otherwise, the respective measured values ​​are identified as normal.

[0085] In another variant, for example, in step S3, a preceding window is formed for a selected measured value with a first selection of measured values ​​from the same measurement series, corresponding to measurement times that precede the selected measured value in time, and a subsequent window is formed with a second selection of measured values ​​from the same measurement series, corresponding to measurement times that follow the selected measured value in time.

[0086] For the preceding and subsequent windows, a statistical measure of central tendency, such as the mean, and a statistical measure of dispersion, such as the standard deviation, are determined. For each window, the deviation of the measured value from the respective mean is normalized by the respective standard deviation and compared to a predefined reference value. If the normalized deviation is greater than the reference value for both the preceding and subsequent windows, the selected measured value is identified as anomalous; otherwise, it is identified as normal.

[0087] In step S4, the acquired measurement data are categorized. A measurement is categorized based at least on whether it was identified as normal or anomalous in step S3. In some embodiments, the categorization of anomalous measurements is further based on whether the anomalous measurement is a time-correlated or time-isolated anomalous measurement, whether it is a severe or minor anomaly, and / or whether it is due to an anomaly in a majority of sensors or in a specific sensor. Categorization can be performed by removing anomalous measurements from the acquired data. Alternatively, the anomalous measurements can remain in the acquired data and be marked to indicate that they are anomalous.Categorization can also include storing identifying information associated with the measurement, indicating, for example, whether an anomalous measurement is a severe or minor anomaly and / or whether an anomalous measurement is due to an anomaly of a majority of sensors or an anomaly of a specific sensor.

[0088] In step S5, the measurement data categorized in step S4 are made available for further use. Optionally, the provided measured values, optionally together with the classifying information, are visualized on the display unit 5. Optionally, the provided measured values ​​are stored in memory 6 and / or used to change operating parameters of the gas turbine 6. The now reliable measurement data are used, for example, to set an operating state of the technical system 2. Compared to the raw measurement data RMD, the data provided according to the procedure allows for simpler and more reliable operation of the system.

[0089] Fig. 4 shows a schematic flowchart illustrating a second embodiment of the method for providing measured values.

[0090] Measurement values ​​7, acquired from the raw measurement data RMD as temporally ordered measurement series, and predefined parameters 8 are entered into an anomaly detection unit (ADU) 9. The parameters 8 are, for example, threshold values ​​or information about statistical quantities to be calculated from the measurement data. The ADU 9 categorizes the measurement values ​​using the predefined parameters 8 into normal and anomalous measurement values, provides normal measurement values ​​10 as reliable and anomalous measurement values ​​11 as unreliable. The anomalous measurement values ​​11 are linked to characterizing information about the anomaly. The anomalous measurement values ​​11 are then transferred to an anomaly classification unit (ACU) 12.The ACU 12 performs a further categorization of the anomalous measurements based on the characterizing information and provides the anomalous measurements linked with the classifying information described above as an anomaly classification 13.

[0091] The function of the ADU 9 will now be explained using the following: Fig. 5 and 6 explained in more detail.

[0092] Fig. 5 Figure 1 shows a schematic flowchart for a method according to a second embodiment. Raw measured values ​​7, acquired as temporally ordered measurement series, are provided to the ADU 9. In the ADU 9, the measured values ​​7 successively pass through a threshold comparison filter 14, a majority decision filter 16, a statistical filter 18, and a noise filter 20. The majority decision filter 16 is only passed through if the prior determination 15 shows that redundant sensors are present, i.e., the number m of sensors for the same measured quantity is greater than 1. Predefined parameters 8 (8a, 8b, 8c, 8d) are provided to each of the filters 14, 16, 18, 20.

[0093] After each of filters 14, 16, 18, a determination S6, S7, S8 is made for each measurement xs,t of the measurement data 7 to determine whether the measurement was identified by the respective filter 14, 16, 18 as an anomalous measurement or as a normal measurement. If a measurement was identified as an anomalous measurement by the respective filter 14, 16, 18, the anomalous measurement is removed from the measurement data 7 and provided as anomalous measurement 11 (11a, 11b, 11c) linked to information characterizing the anomaly. Each measurement can thus be assigned characterizing information. This characterizing information includes details such as the value, position, magnitude, and number of anomalous measurements for the same measurement time, as well as information on whether the anomalous measurement was identified as an anomalous measurement of the first, second, or third type.

[0094] Measurement values ​​identified as noise by the noise filter are removed from measurement values ​​7 and discarded. After the noise filter, only measurement values ​​identified as normal remain in measurement values ​​7. These are provided as normal measurement values ​​10.

[0095] The following section uses the threshold comparison filter 14. Fig. 5 The threshold comparison filter 14 is provided with predetermined parameters 8a. These parameters include a lower threshold th1 and an upper threshold thu. The threshold comparison filter 14 performs a first processing stage. In this stage, the threshold comparison filter 14 compares each measured value xs,t of the 7 measured values ​​with the lower threshold th1 and the upper threshold thu. If one of the conditions from equation 1 below is met, the threshold value is identified as an anomalous measured value. x s , t < th l oder x s , t > th u

[0096] The first stage of the process filters out measured values ​​that, even when considered in isolation, clearly cannot correspond to reality based on the knowledge about the technical system or physical conditions expressed in the threshold values. These anomalous measured values ​​can then be disregarded, thus increasing the reliability of the assessment of the operational status of the technical system.

[0097] The following uses the majority decision filter 16. Fig. 5 The majority decision filter 16 is provided with predetermined parameters 8b and the measured values ​​7. The predetermined parameters 8b comprise a predetermined relative deviation RD and / or a predetermined absolute deviation AD. The majority decision filter 16 therefore performs a second procedural stage. Figur 6 shows a flowchart of the steps of the second process stage.

[0098] In step S10, a loop is initialized over the n measurement points included in the 7 measured values ​​by setting a measurement point counter t to 1.

[0099] In step S11, the measured values ​​xi,t with i=1...m for the m measuring sensors are selected from the measured values ​​7 for the measurement time j, and a variable s, which indicates the number of selected measured values ​​21, is set to m.

[0100] In step S12, it is determined whether the value s of selection 21 is less than or equal to two. If s is less than or equal to two, the measured values ​​of selection 21 are identified as the remaining normal measured values ​​22, and the procedure continues at step S16. If s is not less than or equal to two, the procedure continues at step S13.

[0101] In step S13, a median value µv(t) of the selected measurements is calculated for time t. The median is an example of a statistical measure of central tendency. Choosing the median as a measure of central tendency is advantageous because, based on the applicant's research, it has proven to be particularly robust against outliers or anomalous measurements. However, instead of the median, any other statistical measure of central tendency, such as a mean or a biweight mean, can also be calculated for µv.

[0102] In step S14, a loop is executed over all measured values ​​of the selection. Using a sensor counter i, it is determined for each i=1 to s whether the measured value xi,t deviates from the median value µ v by more than the predetermined deviation AD or RD. If an absolute deviation AD is specified, the determination is carried out by evaluating the condition from the following equation (2): x i , t − μ v t > AD

[0103] If a relative deviation RD is specified, the determination is carried out by evaluating the condition from the following equation (3). x i , t − μ v t μ v t > RD

[0104] If the condition for no i is met, i.e., if none of the selected measured values ​​xi,t deviates from the median value µ v by more than the predetermined deviation, the selected measured values ​​of selection 21 are identified as reliable, i.e., normal measured values ​​22, and the procedure continues with step S16.

[0105] If the condition for an i is met, the procedure continues with step S15.

[0106] In step S15, the measured value xi,t is identified as an anomalous measured value of type 11b (second type). The measured value xi,t is removed from the selection, and the variable s, which represents the size of the selection, is reduced by 1. The procedure then continues with step S12.

[0107] In step S16, it is determined whether further measurement times are available in the measurement data 7, i.e., whether the condition t <n erfüllt ist, wobei n die Anzahl der Messzeitpunkte in den Messdaten 7 ist. Falls ja, wird das Verfahren mit Schritt S17 fortgesetzt. In Schritt S17 wird der Zähler t um eins erhöht und das Verfahren wird mit Schritt S11 fortgesetzt.

[0108] If no further measurement points are available in step S16, i.e., t>=n, then all measured values ​​22 identified as normal during step S12 or S14 are provided as normal measured values ​​10. The normal measured values ​​10 can be provided as measured values ​​7 of a further processing stage or as normal measured values ​​10 as output from the ACU. The second processing stage ends after step S16.

[0109] As described above, in the second stage of the majority decision filter, 16 measured values ​​from different sensors are selected for the same measurement time and categorized as normal or anomalous based on their deviation from a position parameter determined for selection. This stage has the advantageous effect of filtering out the measured values ​​of individual faulty sensors from a majority of normal sensors, thereby increasing the reliability of the assessment of the operating condition of the technical system. Furthermore, the computational and evaluation effort is reduced due to the smaller amount of data.

[0110] The statistical filter 18 is then applied. Fig. 5 The statistical filter 18 is provided with predetermined parameters 8c and the measured values ​​7. The predetermined parameters 8c include the parameters wb, kb, wf, and kf, which are explained in more detail below. The statistical filter 18 performs a third processing stage.

[0111] In the third stage of the process, the statistical filter 18 acquires a series of measurements from the same sensor for the same measurand at different measurement times. For the sake of simplicity, the following describes the steps performed in the third stage with the measurements from a series of measurements for a sensor s, and the notation xs,t, which denotes a measurement for sensor s at measurement time t, is shortened to xt, i.e., measurement for measurement time t. It is understood that the third stage can be repeated for further series of measurements with the measurements of different sensors and different measurands, so that the entirety of all measurements 7 can be acquired using the third stage.

[0112] The recorded measurement series consists of n measured values ​​for n different measurement times. The following describes the steps performed according to the third procedure stage for a measured value xt at a selected measurement time t from the n measurement times. It is understood that the steps described below, according to the third procedure stage, can be performed in a loop t=1 to n for some or all of the n measured values ​​xt.

[0113] Fig. 7 shows the steps that are performed for a measured value according to the third procedure stage.

[0114] In step S18 of the third process stage, the measured value xt is related to statistical location parameters for two so-called windows. According to the present embodiment, a window is a selection of temporally successive measured values. However, the measured values ​​in the window do not necessarily have to follow each other seamlessly; for example, only every second or every third measured value can be selected. For illustration, a first window, also called the preceding window, is a selection of measured values ​​that at least partially precede the measured value xt in the measurement series, and a second window, also called the subsequent window, is a selection of measured values ​​that at least partially follow the measured value xt in the measurement series. The windows can contain the measured value xt, but preferably they do not contain the measured value xt to be categorized.The size, that is, the number of measured values ​​in the two windows, is determined by the predetermined parameters wb and wf and can, for example, be adapted to a predefined operating situation of the technical system from which the measured values ​​are retrieved. A preceding window is selected with a number wb of measured values ​​preceding the measured value xt, and a subsequent window is selected with a number wf of measured values ​​following the measured value xt.

[0115] In step S19 of the third stage of the procedure, a statistical location parameter, which denotes the location of a center point of the window's measurements, and a statistical dispersion parameter, which denotes the spread of the window's measurements, are determined for each of the two windows. Any suitable location parameter and any suitable dispersion parameter can be used for this purpose. According to the second embodiment, for example, a mean is determined as the location parameter and a standard deviation as the dispersion parameter. In one variant, however, a median value can also be used as the location parameter and a mean absolute deviation as the dispersion parameter. In another variant, a biweighted mean can be used as the location parameter and a biweighted standard deviation as the dispersion parameter. A different location parameter and / or a different dispersion parameter can be determined for the preceding window than for the subsequent window.

[0116] In step S20 of the third procedure stage, the deviation of the measured value xt from the mean is determined for both the preceding and the following time window, normalized by the standard deviation, and the result is compared with a predetermined reference value k. In other words, the condition from the following equation (4) is evaluated: x t − x ¯ b s ¯ b ≥ k b und x t − x ¯ f s ¯ f ≥ k f with: xt: Measured value for the measurement time txb: Mean of the wb measured values ​​of the preceding window (location parameter µ b ) sb: Standard deviation of the wb measured values ​​of the preceding window (dispersion parameter σ b ) xf: Mean of the wf measured values ​​of the following window (location parameter µ f ) sf: Standard deviation of the wf measured values ​​of the following window (dispersion parameter σ f ) kb: Predetermined first comparison value kf: Predetermined second comparison value

[0117] The parameters wb, kb, wf, and kf are included in the predetermined parameters 8c provided to the statistical filter 18. According to a variant of the second embodiment, wb = 50, kb = 3, wf = 25, and kf = 2.

[0118] If the condition from equation (4) is met, the measured value xt is identified as an anomalous measured value of the third type in step S21 according to the third procedure stage. Otherwise, the measured value xt is identified as a normal measured value in step S22.

[0119] After the steps of the third procedure stage described above have been carried out for all measurement times of the measurement series and for all measurement data series 7, the measured values ​​identified as anomalous are provided as anomalous measured values ​​11c, and the values ​​identified as normal are provided as normal measured values ​​10. The normal measured values ​​10 can be provided as measured values ​​7 of a further procedure stage or provided as normal measured values ​​10 as output from the ACU.

[0120] As described above, in the third stage of the statistical filter, 18 measured values ​​from the same sensor at different measurement times are examined for anomalies using statistical parameters from a preceding and a subsequent window containing measured values ​​from the same sensor. These values ​​are then categorized as normal or anomalous. This stage ensures that individual anomalous measured values, so-called outliers, are filtered out from the majority of normal measured values ​​from the same sensor and subsequently disregarded. This increases the reliability of the measured values ​​and thus the assessment of the operating condition of the technical system. Besides outliers, steep transients of the measured quantity can also cause a measured value to differ significantly from the preceding or subsequent measured values, resulting in an outlier.Before a simple smoothing procedure, such transients, whose detection is desired, would also be smoothed and filtered out. However, since the third stage of the procedure determines statistical parameters of both a preceding and a subsequent window and uses an AND operation to identify anomalous measurements, this third stage only filters out true outliers, while significantly altered measurements due to steep transients can be categorized as normal measurements. This further increases the reliability of the measurements and thus the assessment of the operational state of the technical system. The reliability of the measurements also leads to more reliable operation of the respective system.

[0121] The noise filter 20 is then used. Fig. 5 The noise filter 20 is provided with predetermined parameters 8d and the measured values ​​7. The predetermined parameters 8d comprise the parameters wb, kb,noise, wf, and kf,noise. The noise filter 20 performs a fourth processing stage. The fourth processing stage is identical to the third processing stage of the statistical filter 18. Therefore, identical steps are not described again. The fourth processing stage differs from the third processing stage in that the predetermined parameters 8d comprise the third and fourth predetermined comparison parameters kb,noise, kf,noise instead of the first and second comparison parameters kb, kf. These parameters are used in the fourth processing stage in the same way as the comparison parameters kb, kf in the third processing stage. They differ in that at least one of the parameters kb,noise, kf,noise is smaller than the corresponding parameter kb, kf.In this way, measured values ​​that do not meet the criteria for an outlier according to the fourth stage of the procedure, but nevertheless deviate significantly from the location parameters of the preceding and following windows, can be identified as noise.

[0122] The fourth stage of the noise filter 20 also differs from the third stage of the statistical filter 18 in that measured values ​​identified as noise are not linked and provided as anomalous measured values ​​11c, but are directly discarded. The cause of measured values ​​identified as noise is suspected to be, for example, physical disturbances that are not due to an anomaly or malfunction of the sensor.

[0123] The fourth stage of the procedure can be carried out after or before the third stage of the procedure, at least partially in parallel with the third stage of the procedure, or simultaneously with the third stage of the procedure.

[0124] The following describes the ACU 12 from Figur 4 described in more detail. The ACU 12 is provided with the anomalous measured values ​​11 linked to characterizing information from the ACU 12. In addition, even if this is in Figur 4 Although not shown, the ACU 12 can also access the measured values ​​10 categorized as normal and the raw measured values ​​7 in a suitable manner. The ACU 12 can be implemented as a separate unit. The ACU 12 performs anomaly classification and provides, for example, measured values ​​13 categorized as anomalous, which are linked to identifying information.

[0125] According to the second embodiment, the ACU 12 assigns a marker to an anomalous measured value xs,t indicating that the anomalous measured value is a time-correlated measured value (marker about a time correlation of the anomaly) if the temporally preceding measured value xs,t-1 or the temporally subsequent measured value xs,t+1 has also been categorized as an anomalous measured value.

[0126] According to the second embodiment, the ACU 12 adds a marker to an anomalous measurement of the second type, indicating that the anomalous measurement is a "severe anomaly" (marker indicating the severity of the anomaly) if the anomalous measurement deviates particularly far from the position parameter µv. This is considered fulfilled if, for the anomalous measurement xi,t in the second processing stage of the ADU 9, the condition of the following equation (5) or (6) is satisfied: x i , t − μ v t > AD + σ v t s t x i , t − μ v t μ v t > RD + σ v t s t μ v t with σ v (t): standard deviation of the measurements identified as normal for measurement time t and st: number of measurements identified as normal for measurement time t.

[0127] According to the second embodiment, the ACU 12 determines whether an anomalous third-type measurement is a severe anomaly by re-executing the third procedure stage of the ADU 9 with a predetermined parameter kb,severe and a predetermined parameter kf,severe instead of the predetermined parameters kb, kf, where kb,severe > kb and kf,severe > kf. According to a variant of the second embodiment, kb,severe = kf,severe = 4. If the anomalous third-type measurement is again identified as an anomalous measurement, this means that the anomalous third-type measurement deviates particularly strongly from the location parameters of the preceding and subsequent windows normalized with the dispersion parameters, and the ACU 12 assigns a marker to such an anomalous third-type measurement indicating that the measurement is a severe anomaly (anomaly severity marker).

[0128] According to the second embodiment, the ACU 12 links an anomalous measurement xs,t of the second or third type with a marker indicating that the anomalous measurement is due to an anomaly of a majority of sensors (marker indicating multiple occurrences of the anomaly in different measuring sensors for the same measurement time) if, from the totality of the measurement values ​​xi,t ; i=1...m for time t, fewer than three measurement values ​​were categorized as normal.

[0129] According to a further development of the second embodiment, the anomaly classification provided by the ACU 12 also includes a sensor-related quantitative summary. This sensor-related quantitative summary consists of a multitude of counters that indicate, for each of the m sensors MS1 to MSm, how many readings, or alternatively, what percentage of the readings provided by the sensor, were categorized as anomalous readings, as time-correlated anomalous readings, as serious anomalous readings, as anomalous readings due to an anomaly in a majority of sensors, and / or as one or more specific combinations thereof. Such a sensor-related quantitative summary allows for a simple graphical visualization of a sensor quality diagram, for example, as a bar or pie chart.

[0130] According to a further development of the second embodiment, it is also conceivable that the sensor-related quantitative summarization is used in a further process step to identify sensors in which one or more of the counters reach excessively high values ​​as unreliable and to deactivate them automatically. Deactivation can be achieved by disregarding the measured values ​​of the deactivated sensors in the process steps of the second embodiment or by switching off the sensors in question. A measuring sensor in which one or more of the counters reach an excessively high value may not be functioning correctly, so that even the measured values ​​of the sensor that are not identified as anomalous may not be reliable. Thus, by deactivating such sensors, the reliability of the provided normal measurement data 10 can be further improved and the amount of data to be considered can be reduced.Switching off such sensors also reduces the energy consumption of all the measuring sensors.

[0131] The ADU 9 and the ACU 12 and the filters 14, 16, 18 and 20 included by the ADU 9 can be implemented by a program-controlled device 4 which executes a computer program that initiates the execution of a method for providing measured values ​​according to the second embodiment, which realizes the functions of the ADU 9 and the ACU 12 and the filters 14, 16, 18 and 20 implemented by the ACU 9.

[0132] The effectiveness of the proposed method for providing measurement data from a technical plant according to the second embodiment was investigated by the applicant using field data obtained from a gas turbine.

[0133] In an initial application, six thermocouples MS1...MS6 were arranged as measuring sensors at the burner tip of a gas turbine 2 and provided measurement series of temperature Ts,t (s: sensor number, t: measurement time) downstream of the burner. These measurements are intended to detect successful ignition of the gas turbine based on the temperature rise in a process for operating the technical system.

[0134] Fig. 8 Shows a plot of raw measurements from the six thermocouples MS - MS6 over a period of 1,500 minutes. In Fig. 8 The horizontal axis represents time t in minutes, and the vertical axis represents a dimensionless temperature T. Measurement data were provided at 1-minute intervals. The graph shows three steady states (31, 32, 33) and three steep transients (34, 35, 36) in the plots of sensor readings MS2 to MS5. The readings from sensor MS1 alternate between correct values ​​(upper dotted area of ​​plot MS1) and clearly anomalous values ​​(lower area of ​​plot MS1). The readings from sensor MS6 do not exhibit this transient. If a simple average were calculated from the readings of the six sensors MS1 to MS6, it would be significantly distorted by the readings from sensors MS1 and MS6. Furthermore, minor outliers are visible at t = 450 (reference 37).

[0135] With the in Fig. 8 The raw measured values ​​shown were used in a procedure according to a second embodiment of the invention. The predetermined parameter AD for the majority decision filter 16 was selected to be AD = 35 °C.

[0136] Fig. 9 Figure 1 shows a plot of the normal, i.e., reliable, measured values ​​provided by the procedure. The unreliable measured values ​​from sensor MS6 were successfully identified by the second stage of the procedure and removed from the normal measured values. From the measurement data of sensor MS1, some reliable measured values ​​were retained between t = 800 and 1500. Furthermore, the outliers 37' were smoothed by the third stage of the procedure, i.e., some of the outliers 37 were removed. Transients 34, 35, and 36 were correctly categorized as normal and are accordingly shown in Figure 1. Fig. 9 still clearly visible.

[0137] In a second application, thirteen thermocouples MS - MS13 were installed as measuring sensors in the shaped section of the combustion chamber of a gas turbine 2 and provided measurement series of temperature Ts,t (s: sensor number, t: measurement time). Such measurements are used to detect flame ignition in a process for operating the gas turbine.

[0138] Fig. 10 This shows a plot of raw measurements from the thirteen thermocouples MS1-MS13 over a period of 1,500 minutes. Fig. 10 Two steady states 38 and 40 and one steep transient 39 can be identified in the plots of the measured values ​​from sensors MS1 to MS13. Outliers are visually apparent in areas 41. In the region of steady state 38, the plots of the measured values ​​from sensors MS1 and MS12 deviate significantly from the plots of the measured values ​​from the other sensors.

[0139] At the in Fig. 10 The raw measurement values ​​shown were used in a procedure according to a variant of the second embodiment of the invention. The predetermined parameter AD for the majority decision filter 16 was selected to be AD = 35 °C. For the statistical filter 18, a biweight mean was used as the statistical location parameter µ f,b and a biweight standard deviation as the statistical dispersion parameter σ f,b. The parameters kb and kf were both set to kb = kf = 3.

[0140] Fig. 11 Figure 1 shows a plot of the normal, i.e., reliable, measured values ​​provided by the procedure. The unreliable measurement data from sensors MS1 and MS12 were successfully identified by the second stage of the procedure and removed from the normal measured values. The strong outliers 41 ( Fig. 10 ) could be almost completely categorized as anomalous and removed by the third processing stage. Transient 39, however, was correctly categorized as normal and is therefore in Fig. 11 still clearly visible. The same applies to stationary areas 38 and 40.

[0141] Although the present invention has been described using exemplary embodiments, it is highly adaptable. Technical systems other than gas turbines can be controlled and operated. Furthermore, statistical parameters other than those mentioned in the examples are conceivable for assessing the normality or anomaly of the measured values. The proposed measures lead overall to improved measurement quality and thus to more reliable system control. In essence, the measures filter the raw measurement data to improve the overall measurement quality. Moreover, the particularly reliable and accurately provided measurement data facilitates a simplified evaluation of historical measurement data, enabling, for example, the identification of certain pathological operating conditions, such as fault conditions or critical operating modes. This simplifies the control and regulation of the system's operation.

Claims

1. A method for providing (S5) measured values of a technical installation (2) by means of a programme-controlled apparatus for performing the method, wherein measured values (7) of at least one measurement row are detected (S1), wherein a respective measured value is provided by a measurement sensor (MS1-MSm) for a respective physical measurement variable in the technical installation (2) for a respective measurement time (S0), and the measured values are categorised (S4) as normal measured values (10) or anormal measured values (11) with the help of a threshold comparison (S2) and at least one further method step (S3), characterised by: performing the threshold comparison (S2) as a first method step (S2) with the steps: detecting at least one measured value of the at least one measurement row, comparing the at least one measured value with a predetermined threshold (8a) for generating a comparison result, and identifying the at least one measured value as a normal measured value or as an anormal measured value of a first type as a function of the comparison result, and performing a second and a third method step (S3), wherein the second method step (S3) comprises: detecting several selected measured values of several measurement rows, wherein the several selected measured values are provided by different measurement sensors (MS1-MSm) for a same measurement variable and a same measurement time, calculating a statistical location parameter (µv) of the selected measured values, for at least one of the detected selected measured values: comparing the at least one measured value with the statistical location parameter (µv) and, if the at least one measured value deviates from the statistical location parameter (µv) by more than a predetermined relative (RD) or absolute (AD) deviation (8b), identifying the at least one measured value as an anormal measured value of a second type, wherein the third method step (S3) comprises: detecting a measurement row with measured values, wherein the measured values are provided by a measurement sensor for a same measurement variable and different measurement times and the measured values are arranged in time in the measurement row, and for at least one of the detected measured values of the measurement row: determining (S19) a first statistical location parameter (µb) and a first statistical dispersion parameter (σb) for a first predetermined number (wb) of measured values of the same measurement row which precede the at least one measured value of the measurement row in time, determining (S19) a second statistical location parameter (µf) and a second statistical dispersion parameter (σf) for a second predetermined number (wf) of measured values of the same measurement row which follow the at least one measured value of the measurement row in time, calculating a first quotient (Qb) from the amount of the difference between the at least one measured value and the first statistical location parameter (µb) and the first statistical dispersion parameter (σb), calculating a second quotient (Qf) from the amount of the difference between the at least one measured value and the second statistical location parameter (Qf) and the second statistical dispersion parameter (σf), identifying (S21, S22) the at least one measured value as an anormal measured value of a third type if the first quotient (Qb) is larger than or equal to a predetermined first comparison value (kb) and the second quotient (Qf) is larger than or equal to a predetermined second comparison value (kf), or as a normal measured value if the first quotient (Qb) is smaller than the predetermined first comparison value (kb) or the second quotient (Qf) is smaller than the predetermined second comparison value (kf), wherein identifying a measured value which is identified as an anormal measured value further comprises storing classified information on a severity of an anomaly, on a temporal correlation of the anomaly and / or on multiple occurrences of the anomaly with different measurement sensors for the same measurement time.

2. The method according to claim 1, wherein the statistical location parameter (µv) is a median value, a middle value or a biweight middle value, the first statistical location parameter (µb) and the second statistical location parameter (µf) each are a middle value, a median value or a biweight middle value, and the first statistical dispersion parameter (σb) and the second statistical dispersion parameter (σb) each are a standard deviation, a mean absolute deviation or a biweight standard deviation.

3. The method according to claim 2 or 3, wherein the third method step (S3) further comprises: identifying a respective measured value of the measurement row as a noise when the first quotient (Qf) is larger than or equal to a predetermined third comparison value (kb, noise) and the second quotient (Qb) is larger than or equal to a predetermined fourth comparison value (kf, noise), wherein the predetermined third comparison value (kb, noise) is smaller than or equal to the predetermined first comparison value (kb) and the predetermined fourth comparison value (kf, noise) is smaller than or equal to the predetermined second comparison value (kf).

4. The method according to any one of claims 1 to 3, wherein identifying a measured value further comprises changing the measured value if the measured value is not identified as a normal measured value.

5. The method according to claim 4, wherein changing a measured value comprises providing the measured value with an identifier and / or removing the measured value from the measurement row.

6. The method according to any one of claims 1 to 5, wherein the second method step further comprises: when a measured value is identified as an anormal measured value of a second type: removing the anormal measured value of a second type from the several selected measured values, and repeating the steps of the second method step for the remaining selected measured values if more than two selected measured values remain.

7. The method according to any one of claims 1 to 6, further comprising: determining the number of the measured values provided by a selected measurement sensor (MS1-MSm) and the number of the anormal measured values from the measured values provided by the selected measurement sensor (MS1-MSm); and deactivating the selected measurement sensor (MS1-MSm) when the proportion of the anormal measured values in the measured values provided by the selected measurement sensor (MS1-MSm) exceeds a predetermined acceptance value.

8. A technical system (1), comprising a technical installation (2), at least one measurement sensor (MS1-MSm) and a programme-controlled apparatus (4) which is adapted to perform the method according to any one of claims 1 to 7.

9. The technical system according to claim 8, wherein the technical installation (2) is a gas turbine.

10. The technical system according to claim 8 or 9, wherein the measurement sensor (MS1-MSm) is a temperature sensor, a pressure sensor, a movement sensor or a vibration sensor.

11. A method for operating a technical system (1) with a technical installation (2), at least one measurement sensor (MS1-MSm) and a programme-controlled apparatus (4), comprising: performing the method according to any one of claims 1 to 7 with measured values by the at least one measurement sensor (MS1-MSm) for at least one physical measurement variable in the technical installation (2) with the programme-controlled apparatus (4), and changing an operating parameter of the technical installation (2) as a function of the measured values provided by the method.

12. The method according to claim 11, wherein changing an operating parameter of the technical installation (2) comprises deactivating the technical installation (2) for maintenance purposes.

13. A computer programme product which causes performance of a method according to any one of claims 1 to 7, 11 and 12 on a programme-controlled apparatus (4).