Method for monitoring measured variables in process equipment using measuring instruments

Abstract

The present invention relates to a method for monitoring a measured variable measured in a process plant using a measuring instrument. The measured variable changes over time during the process and is continuously measured by a set of measuring instruments used on the process plant during repeated execution of a predetermined dynamic process, during which measurement data are continuously acquired, including measured values of the measured variable, training data measured during a training period, and monitoring data measured after the training period. Vectors are continuously determined based on the measurement data, whose vector components include rates of change of the measured values of the measured variable determined for a series of k consecutive time points based on the measurement data, each series of time points covering a time window of a predetermined duration, and the time windows of the consecutive vectors are shifted relative to each other by a predetermined time difference. The vectors determined based on the training data are detected and stored in the form of a reference cluster. A monitoring method is performed: the vectors determined based on the monitoring data are compared with the reference cluster, an anomaly is determined when at least one vector is outside the reference cluster, and an anomaly information is made available.

Description

Technical Field

[0001] The invention relates to a method, in particular a computer-implemented method, for monitoring a measured variable, which changes over time during a process and is measured continuously by a set of measuring instruments used on a device during the repeated execution of a predetermined dynamic process. Background Art

[0002] Process equipment for repeatedly executing predetermined dynamic processes is used in a variety of industries, such as the chemical and food industries, and in water and wastewater treatment. These include, but are not limited to, production, processing, and treatment equipment. Such equipment typically uses measuring instruments that measure at least one measurand related to the process being executed and / or the process products obtainable by the process. Examples of such measuring instruments include pressure gauges, level gauges, density meters, thermometers, flow meters, and pH meters, such as those sold by the Endress+Hauser Group. In measurement and control technology and process automation, the measured values of such measuring instruments are used, for example, to monitor, control, and / or regulate the process being repeatedly executed by the equipment.

[0003] Abnormalities that occur during the ongoing operation of such systems, such as malfunctions and / or damage to the system, faulty control and / or regulation of the system, and, of course, faulty measured values of the measured variables due to malfunctions of the measuring instruments, can in some cases lead to significant damage. This damage ranges from the production of faulty process products to hazards to humans and the environment resulting from the system and / or the faulty process products.

[0004] To ensure the proper functioning of equipment and measuring instruments, measuring instruments often undergo regular service measurements. Such service measurements include, for example, verification, calibration, and maintenance. Due to the costs and interruption of ongoing measurement operations associated with such service measurements, the time intervals between consecutive service measurements are preferably measured in such a way that, on the one hand, the intervals are short enough to ensure a sufficient degree of safety, and, on the other hand, they are long enough to keep costs and effort within reasonable limits. To ensure that the operability of a measuring instrument can be checked even during the time periods between chronologically consecutive service measurements, individual measuring instruments can be equipped with additional functions, such as self-monitoring or diagnostic functions, which allow the respective measuring instrument to automatically monitor its functionality. An example of this is the measuring instruments offered by the Endress+Hauser Group, which are equipped with a heartbeat function that allows the measuring instrument to monitor itself. However, such measuring instruments are generally more expensive than those without this function and, for cost reasons, are generally used only in safety-related areas.

[0005] Alternatively, a model created for this purpose can be used to monitor the function of a specific measuring instrument, which measures a measurand that is described as a function of one or more measurands that can be measured with the aid of other measuring instruments. However, the implementation of such a model generally requires detailed knowledge of the process to be performed by the device and the correlations between the individual measurands, which may arise from the process and often have physical causes.

[0006] Furthermore, EP 3 002 651 A1 describes a method for monitoring a process step of a process that can be performed using a device. The process step is monitored based on the measured values of a measured variable measured on the device during the execution of the process step and the associated measurement times, which are related to the start time of the respective execution of the process step. For this purpose, a model is created based on training data, which includes parameters that can be determined based on the measured values and measurement times. For a plurality of measuring instruments, each parameter includes:

[0007] a) coefficients of a basis function which represents the measured value as a function of time and which is determined on the basis of the measured values of the corresponding measuring instrument, and

[0008] b) The distribution of the deviations between the measured values and the function given by the previously determined coefficients and basis functions. In addition, they include the correlation between the measured values of pairs of measuring instruments that provide correlated measured values during the execution of the process step.

[0009] Furthermore, for each parameter, the model includes an associated reference range, determined based on the training data, within which the corresponding parameter is expected when the process step is performed flawlessly. Thus, if at least one of the parameters determined during monitoring is outside the associated reference range, an anomaly is detected by the monitoring method.

[0010] This method offers the advantage that the model, including the reference range, and the monitoring of the process steps can also be performed based on already existing measured values. However, it is disadvantageous that the method must be performed independently for each individual process step to be monitored, that the measured data must be acquired in conjunction with the start time of the respective process step, and, importantly, that the creation of the model is relatively complex. Summary of the Invention

[0011] The object of the present invention is to specify an alternative usable method for monitoring a measured variable which changes over time during a process and which is measured continuously at a process device during the repeated execution of a predetermined dynamic process by a set of measuring instruments used on the device, the method being executable in a simple manner without detailed process knowledge.

[0012] This object is achieved by a method, in particular a computer-implemented method, for monitoring a measured variable, which changes over time in a process and is measured continuously at a process device during repeated execution of a predetermined dynamic process by a set of measuring instruments used at the device, wherein:

[0013] During the repeated execution of the dynamic process, measurement data are continuously acquired, the measurement data comprising the measured values of the measured variable measured by means of the measuring instrument and the associated measurement times when the measured values are measured,

[0014] The measurement data includes training data measured during the training period and monitoring data measured after the training period.

[0015] continuously determining a vector on the basis of the measurement data, the vector components of which comprise the respective rates of change of the measured values of the respective measured variables determined for a series of k consecutive time points on the basis of the measurement data, wherein each series of time points covers a time window of a predetermined duration and wherein the time windows of the consecutive vectors are in each case offset relative to one another by a predetermined time difference,

[0016] Vectors determined based on the training data are detected and stored in the form of reference clusters, and

[0017] Execute a monitoring method, where:

[0018] The vectors determined from the monitoring data are compared with the reference clusters respectively,

[0019] If at least one of the vectors determined based on the monitoring data is outside the reference cluster, an anomaly is determined, and

[0020] Provides information about detected anomalies.

[0021] The method according to the present invention offers the advantage that it can be performed in a simpler manner, directly based on the measured values of the measured variables, without requiring detailed knowledge of the processes. Another advantage is that the reaction speed of the measured values of all cooperating measured variables is monitored via vectors within a period corresponding to the duration of a time window, which is coupled chronologically via the processes. Due to the sliding time window of the continuous vectors, monitoring can occur over the entire operating period of the system. Another advantage is that anomalies can be directly identified based on a reference cluster, and the reference cluster can be determined very simply based on training data. This offers the advantage that complex models and correlations between the individual measured variables do not need to be determined in order to perform the method.

[0022] An alternative embodiment comprises a measuring instrument according to the invention, wherein

[0023] Acquiring measurement data such that the measured values of the measured variables and the associated measurement times are transmitted to a data processing device and at least temporarily stored in a memory associated with the data processing device, and

[0024] The data processing device is designed to perform the determination of the vector, the determination of the reference cluster and the monitoring method based on the measurement data, and the data processing device is designed to output information about the detected anomaly via an interface connected to the data processing device and / or is designed to provide this information in a form that can be read out and / or retrieved via the interface.

[0025] A first refinement comprises a method in which the predetermined duration of the time window is:

[0026] Less than or equal to the process duration of the process

[0027] is greater than or equal to the duration of at least one dynamic process event occurring during each execution of the process, by means of which at least two of the measured variables vary with time,

[0028] is greater than or equal to at least one reaction time respectively associated with one of the measured variables, by which the measured value of the corresponding measured variable changes in response to the dynamic process event, and / or

[0029] Greater than or equal to at least one time scale respectively associated with one of the measured variables, at which the measured value of the corresponding measured variable changes over time.

[0030] A second refinement comprises a method in which the predetermined duration of the time window is determined based on training data, wherein:

[0031] determining, based on the rate of change of the measured value of one of the measured variables, a start time at which the rate of change of the measured value of the measured variable exceeds a predetermined limit value,

[0032] determining, based on the measured values of the other measured variables, for each of the other measured variables, a respective measured variable-specific end time at which the rate of change of the measured value of the respective other measured variable exceeds a predetermined limit value for the first time after the start time,

[0033] For each measurand, determine the difference between the measurand-specific end and start times, and

[0034] The duration of the time window is determined such that it is greater than or equal to at least one of these differences.

[0035] A third refinement comprises a method in which the predetermined duration of the time window is determined based on training data, wherein:

[0036] For at least one of the measurands, in each case:

[0037] determining a time period over which the measured values of the respective measured variables change over time and an average value of the rate of change occurring within such a time period based on the rate of change of the measured values of the corresponding measured variables, and

[0038] determining a time scale corresponding to the period during which the measured value of the measured variable changes at a rate of change corresponding to the mean value by a predetermined proportion of the measuring range of the measuring instrument measuring the corresponding measured variable, a proportion greater than or equal to 40% of the measuring range, or a proportion greater than or equal to 60% of the measuring range, and

[0039] The duration of the time window is determined such that it is greater than or equal to at least one of the time scales.

[0040] A fourth refinement includes a method wherein time intervals between consecutive time points within the time window are equal to a predetermined constant discretization time.

[0041] According to a development of the fourth development, the discretization time is dimensioned such that it is small enough to be able to resolve the process dynamics of the process.

[0042] According to another improvement of the fourth improvement, the discretization time is determined based on training data, wherein:

[0043] For at least one of the measurands, in each case:

[0044] determining a time period over which the measured value changes over time and an average value of the rate of change occurring within the time period based on the rate of change of the measured value of the corresponding measured variable, and

[0045] determining a time scale corresponding to the period of time during which the measured value of the measured variable changes at a rate of change corresponding to the mean value by a predetermined percentage of the measuring range of the measuring instrument measuring the corresponding measured variable, a percentage less than or equal to 10% of the measuring range, or a percentage less than or equal to 5% of the measuring range, and

[0046] The discretization time is determined such that it is less than or equal to at least one of the time scales.

[0047] A fifth refinement comprises a method in which the time windows of successive vectors are shifted relative to one another over time by a time difference which is greater than the discretization time and / or less than or equal to the duration of the time window.

[0048] A second embodiment comprises a method in which the cause of an anomaly established by a monitoring method is determined and / or resolved by corresponding countermeasures.

[0049] A sixth refinement includes a method in which at least one cause of an anomaly is determined for at least one vector identified as being outside of the reference cluster using the monitoring method based on the distance of the vector from the reference cluster, wherein:

[0050] For each measurand, determine the proportional contribution of the rate of change of the measured value of the corresponding measurand with respect to the distance of the vector from the reference cluster, and

[0051] The measured value of the measurand whose proportional contribution has a maximum value and / or exceeds a predetermined limit value is determined to be the cause of the anomaly, and

[0052] Provides information about the cause of the exception.

[0053] According to a further improvement of the sixth improvement, the cause of the abnormality determined by means of the monitoring method is determined based on at least one of the causes of the abnormality by checking whether the abnormality detected by the monitoring method is caused by a temporal change of at least one of the measured variables identified via the cause of the abnormality previously determined, which deviates from the expected behavior, or is caused by functional damage to the measuring instrument measuring the corresponding measured variable, wherein the cause of the abnormality determined by means of the monitoring method is checked.

[0054] According to the third embodiment, training data is acquired in a training period in which error-free operation of the device and the measuring instrument can be assumed and the process is performed continuously several times. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The present invention and its advantages will now be explained in detail using the figures in the accompanying drawings, which show exemplary embodiments. Identical elements are indicated by the same reference numerals in the figures.

[0056] Figure 1 An apparatus for repeatedly performing a dynamic process is shown;

[0057] Figure 2 The measured values of three different measurands are shown;

[0058] Figure 3 Shown Figure 2 The rate of change of the measured value shown in ; and

[0059] Figure 4 A reference cluster and vectors outside the reference cluster are shown. DETAILED DESCRIPTION

[0060] The invention relates to a method for monitoring a measured variable Mj, in particular a computer-implemented method, wherein the measured variable Mj varies over time in a process and is measured continuously on a process device during repeated execution of a predetermined dynamic process using a set of measuring instruments MDj used on the device.

[0061] Apparatus known from the prior art for repeatedly performing predetermined dynamic processes is suitable as process equipment. Examples include industrial equipment, such as production, processing, and treatment equipment, such as is currently used in various industries, such as the chemical industry, the food industry, and water and / or wastewater treatment. In particular, processes comprising one or more process steps that are performed continuously over time are suitable as dynamic processes that can be repeatedly performed using the apparatus. Examples include production, processing, and / or treatment processes that can be performed using the apparatus.

[0062] As an example, Figure 1 A process plant for producing a process product is shown. The plant comprises a container 1 which can be filled with various feed media via a feed line 3, which are processed in the container 1 by means of a stirrer 5 to form the process product, and which are then removed from the container 1 via a withdrawal line 7. This process is repeated continuously in the plant, with a predetermined amount of process product being produced each time the process is carried out and being removed from the container 1.

[0063] Suitable measuring instruments MDj for use on the device are measuring instruments known from the prior art for measuring at least one measured variable. These include, for example, measuring sensors containing sensors, as well as measuring instruments equipped with at least one sensor and / or measuring transducer, which are designed for metrological detection of at least one measured variable Mj. Examples of these include pressure gauges, level gauges, density meters, thermometers, flow meters, pH meters, etc., such as those sold by the Endress+Hauser Group. As possible examples, Figure 1 The set of measuring instruments MDj shown comprises:

[0064] a level meter L for measuring the filling level h of the filling material 9 located in the container 1,

[0065] at least one flow meter F1, F2, F3, respectively inserted into the feed line 3 or into one of the extraction lines 7, for measuring the flow rate f1, f2, f3 through the respective line, and

[0066] - a pressure gauge P for measuring the hydrostatic pressure ph exerted by the filling material 9 in the container 1 y d.

[0067] Of course, depending on the type of device and / or the corresponding process to be repeatedly performed with the aid of the device, other measuring devices MDj can also be used.

[0068] During operation, the device repeatedly executes a predetermined dynamic process. Simultaneously, the measured variable Mj is continuously measured using the measuring instrument MDj. Measurement data D is continuously acquired, comprising the measured values mj(ti) of the measured variable Mj measured by the measuring instrument MDj during the device's operation, as well as the associated measurement times ti at which the corresponding measured values mj(ti) are measured.

[0069] The continuously acquired measurement data D includes training data DT acquired during a training period at the start of the method and monitoring data DU acquired after the training period. In this case, the training period is a period during which error-free operation of the device and measuring instrument MDj can be assumed, and the training period is dimensioned such that the process is performed multiple times continuously during the training period.

[0070] Acquired measurement data D, for example consisting in the measured values mj(ti) of the measured variables Mj and the associated measurement times ti, are transmitted to the data processing device 11 and at least temporarily stored in a memory 13 associated with the data processing device 11 .

[0071] Depending on the design of the measuring instrument Mj and the device, the transmission of the measured value mj(ti) and the associated measurement time ti of at least one measuring instrument Mj can in each case occur via a direct connection between the corresponding measuring instrument Mj and the data processing device 11 or via a connection B1, B2; C1, C2 which continues from the corresponding measuring instrument Mj via at least one intermediate station to the data processing device 11. Figure 1 The direct connection A shown in FIG is shown using the example of a level gauge L. For example, an edge device 15 and / or a superordinated unit 17 arranged on or near the device are suitable as intermediate stations. A suitable superordinated unit 17 is, for example, a unit such as a process control system or a programmable logic controller, which monitors, controls and / or regulates the process to be performed by means of the device based on the measured values mj(ti) of the measuring instrument MDj. Figure 1 In the embodiment shown, the superordinate unit 17 can be designed, for example, such that it controls the measured hydrostatic pressure p of the measured filling level h via corresponding control of the controllable valves SV1, SV2, SV3 inserted into the feed line 3 and into the withdrawal line 7. hyd and / or at least one of the measured flow rates f1, f2, f3, to control or regulate the inflow of the incoming medium and / or the extraction of the process product.

[0072] Wired and / or wireless connections A, B1, B2, C1, C2 and / or communication protocols known in the art are suitable, for example, for transmitting the measurement data D. Examples include LAN, WLAN, fieldbus, Profibus, HART, Bluetooth, and near-field communication. Optionally, at least one of the measuring instrument Mj, the edge device 15, and / or the superordinated unit 17 is connected to the data processing device 11 via the Internet, for example via a communication network such as TCP / IP.

[0073] For example, a single computer, a mainframe computer, or other data processing device may be used as data processing device 11. Alternatively, data processing device 11 may be a device located in the cloud. In this case, the term "cloud computing" is used. Cloud computing describes a method of providing an abstract IT infrastructure such as hardware, computing power, data storage, network capabilities, or even software that dynamically adapts to demand via a network (e.g., wireless or wired Internet).

[0074] The continuously acquired measurement data D represent the temporal progression of the measured values mj(ti) of the measured variables Mj measured by the measuring instruments MDj. Based on the measurement data D, the change rates vj(ti) of the measured values mj(ti) of the individual measured variables Mj are continuously determined. Figure 2 As an example, a short section of the temporal development of the measured values m1 (t), m2 (t), m3 (t) of three different measured variables M1, M2, M3 that change over time is shown. Figure 3 Shown Figure 2 The rates of change of the measured values v1(t), v2(t), v3(t) are shown in FIG.

[0075] In the case of a deterministic, error-free process that runs repeatedly in the same manner and error-free measurement of the measured variables Mj, a velocity vector v(t) formed by the rates of change vj(t) (where v(t):=(v1(t), ..., vn(t))) describes a trajectory in n-dimensional space as a function of time t, which closes to form a loop, with the dimension n of the space corresponding to the number of measured variables Mj. This trajectory is traversed again each time the process is executed and reproduces the temporal coupling between the rates of change vj(t) of the measured values mj(ti) of the measured variables Mj, which are predetermined by the process, the reaction speed of the individual measured variables Mj to the process dynamics, and the measuring instrument-dependent reaction speed of the measured values mj(ti) to changes in the measured variables Mj.

[0076] Based on the rates of change vj(ti), vectors VR are continuously determined, whose vector components each comprise the rates of change vj(ti) of the measured values mj(ti) of the respective measured variable Mj(ti) determined for a series of k consecutive time points [tr1, ..., trk]. Each such vector VR thus comprises the vector components: VR:=(m1(t r1 ),…,m1(t rk ),…,m n (t r1 ),…,m n (t rk )) where n is equal to the number of measured variables Mj and k is equal to the number of consecutive time points [tr1, ..., trk] of the series. Independently of the number of measured variables Mj and the number k of time points [tr1, ..., trk], each series of time points [tr1, ..., trk] in each case covers a time window of predetermined duration ΔT. As an example, such a time window is Figure 3 The time windows are designed as sliding windows. This means that the time windows of consecutive vectors (VR) are in each case offset relative to each other by a predetermined time difference, as Figure 3 As shown by the displacement arrow S extending parallel to the time axis in . Therefore, each such vector VR detects the time curve of the change rate vj(ti) of the measured value mj(ti) of the cooperating measured variable Mj in the trajectory segment described by the velocity vector v(t).

[0077] In a first step, vectors VR determined based on the training data TD are detected and stored in the form of a reference cluster RC. Subsequently, a monitoring method is performed, in which the vectors VR successively determined based on the monitoring data DU are compared with the reference cluster RC previously determined based on the training data DT. Based on this comparison, a check is performed to determine whether at least one of these vectors VR is outside the reference cluster RC. If this is the case, an anomaly is detected, and information regarding the detected anomaly is provided.

[0078] To this end, Figure 4 A schematic diagram is shown in which vectors VR determined based on training data DT and forming a reference cluster RC are represented as points, and in which vectors VR determined based on monitoring data DU and located outside the reference cluster RC are represented as arrows originating from the origin of the n*k dimensional space coordinate system.

[0079] The determination of the vector VR, the determination of the reference cluster RC, and the monitoring performed after the determination of the reference cluster RC based on the vector VR are preferably performed with the aid of the data processing device 11. In this case, the data processing device 11 is preferably designed to output information about the detected anomaly via an interface 19 connected to the data processing device 11 and / or to make it available to the data processing device 11 in a readable and / or retrievable form via the interface 19. If the information is output, this can occur, for example, in the form of an alarm, an output signal, or in the form of an email or text message to a predefined recipient, such as a predefined email address, a smartphone, or a tablet.

[0080] The monitoring method is based on the fact that the measured values mj(ti) of all measured variables Mj are coupled in a temporal sequence via the process. During error-free operation of the system, this results in each dynamic process event causing the measured variable Mj affected by the process event to vary over time, with a reaction time that depends on the process and a reaction amplitude that depends on the process. In the event of error-free operation of the measuring instrument MDj, the measured values mj(ti) of the measured variables Mj also vary accordingly in each case with a reaction time and a reaction amplitude that depend on the reaction time and reaction amplitude of the measured variables Mj and on the measurement characteristics of the measuring instrument Mj. Thus, anomalies are detected via comparison with a reference cluster RC, which result in a temporal variation of the measured value mj(ti) of at least one measured variable Mj relative to the measured value mj(ti) of another measured variable Mj, with the reaction time and / or reaction amplitude deviating from the reference cluster RC. This offers the advantage of performing only one unified monitoring method based on the reference cluster RC, identifying both process-related and equipment-related anomalies, which result in at least one measured variable Mj monitored based on the measured value mj(ti) varying over time in a manner that deviates from the reference cluster RC relative to the measured values mj(ti) of other measured variables Mj, and identifying anomalies where the measured values mj(ti) of at least one measuring instrument MDj vary over time in a manner that deviates from the reference cluster RC due to a functional impairment of the corresponding measuring instrument MDj relative to the measured values mj(ti) of another measuring instrument MDj. Examples of process-related anomalies include faulty process execution, for example, caused by faulty control or regulation of the process. Examples of equipment-related anomalies include a hole in the container 1, which results in a time-delayed increase in the fill level h and / or a relatively small increase in the amount. Examples of functional impairments of the measuring instrument MDj that can be identified as abnormal with the aid of this method are functional impairments such as range errors, which affect the amplitude of the rate of change vj(ti) of the measured values mj(ti); and functional impairments that affect the reaction time of the measured values mj(ti) of the measuring instrument MDj to changes in the measured variable.

[0081] The method has the advantages mentioned at the outset. Optionally, the individual method steps may have different embodiments, which may be used individually or in combination with one another. Examples of this are described below.

[0082] For example, the duration ΔT of the time window can be determined in various ways. One option is to determine the duration ΔT of the time window based on the process duration of a process repeatedly executed on the device, such that the duration ΔT is less than or equal to the process duration of the executed process. For very short process durations, such as, for example, one or several minutes, the duration ΔT of the time window can, for example, be substantially equal to the process duration. For longer process durations, the duration ΔT of the time window can be determined, for example, such that it corresponds to a predetermined percentage of the process duration, such as a proportion of 5% to 10% of the process duration.

[0083] Alternatively or additionally, the duration ΔT of the time window is dimensioned, for example, such that it is greater than or equal to the duration of at least one dynamic process event occurring during each execution of the process, by means of which at least two measured variables Mj change over time. Examples of dynamic process events are Figure 1 The filling and emptying processes of the device shown are carried out, wherein in each case the flow rates f1, f2, f3 through the associated feed line 3 or withdrawal line 7, the filling level h in the container 1 and the hydrostatic pressure p in the container 1 hyd change.

[0084] Alternatively or additionally, the duration ΔT of the time window can be predetermined such that it is greater than or equal to at least one reaction time associated with each measured variable Mj, wherein the measured value mj(ti) of the respective measured variable Mj changes in response to a dynamic process event. For example, such reaction times can be estimated based on training data DT. This can be accomplished, for example, by determining a start time ta based on the rate of change vj(ti) of the measured value mj(ti) of one measured variable Mj, at which the rate of change vj(ti) of the measured value mj(ti) of such measured variable Mj exceeds a predetermined limit value, and determining a measured variable-specific end time tej based on the measured value mj(ti) of another measured variable Mj, at which the rate of change vj(ti) of the measured value mj(ti) of the respective measured variable Mj first exceeds the predetermined limit value after the start time ta. For each measured variable Mj, a difference Δtj:=tej-ta is then determined between the measured variable-specific end time tej and the start time ta. In this case, each of the difference values Δtj respectively forms an estimated value for the reaction time, based on which the duration ΔT of the time window is determined such that it is greater than or equal to at least one of these estimated values.

[0085] Alternatively or additionally, the duration ΔT of the time window can be predetermined such that it is greater than or equal to at least one corresponding time scale determined for one of the measured variables Mj, over which the measured value mj(ti) of the corresponding measured variable Mj varies over time. This time scale can be determined for each measured variable Mj based on training data TD. In this case, for example, a time period is established based on the rate of change vj(ti) of the measured value mj(ti) of the corresponding measured variable Mj, in which the measured value mj(ti) varies over time, and an average value of the rate of change vj(ti) occurring within this time period is determined. This average value is used to determine the time scale. Alternatively, this can be accomplished by setting the time scale equal to the time interval in which the measured value mj(ti) varies by a predetermined proportion of the measuring range of the corresponding measuring instrument MDj, such as a proportion greater than or equal to 40% or greater than or equal to 60%, at a rate of change vj(ti) corresponding to the average value.

[0086] Whether the duration ΔT of the time window can be determined based on the process duration, the duration of at least one dynamic process event occurring during each execution of the process, at least one reaction time associated with each measured variable Mj, and / or based on at least one time scale associated with each measured variable Mj, the time intervals between consecutive time points tr1, ..., trk within the time window can also be determined or predetermined in various ways. Alternatively, for example, a constant discretization time can be set. In this case, the time intervals between directly consecutive time points tr1, ..., trk are equal to the predetermined constant discretization time.

[0087] The discretization time is preferably measured so that it is small enough to still be able to interpret the process dynamics. For this purpose, the discretization time can be set, for example, to an estimated value determined based on the process. Alternatively or additionally, the discretization time can be determined based on at least one time scale, which can be established based on the training data DT in the manner previously described. In this case, the discretization time is determined, for example, so that it is less than or equal to the period in which the measured value mj(ti) of the corresponding measured variable Mj changes at a rate of change vj(ti) corresponding to the mean value for a predetermined percentage of the measured value range, such as, for example, less than or equal to 10% of the measuring range of the corresponding measuring instrument MDj, or less than or equal to 5% of this measuring range. This percentage is, of course, set to be significantly lower than the proportion of the measuring area used to determine the duration ΔT of the time window.

[0088] exist Figure 3The time difference indicated by the displacement arrow S in , by which the time windows of consecutive vectors VR are shifted in time relative to each other, is for example dimensioned such that it is greater than the discretization time and / or less than or equal to the duration ΔT of the time window.

[0089] If an anomaly is detected by the monitoring method, its cause is preferably determined and analyzed through corresponding countermeasures. Regarding the determination of the cause, it is optionally performed so that, for example, for at least one vector VR identified as being outside the reference cluster RC based on the monitoring method, at least one cause of the anomaly is determined based on the distance of the corresponding vector VR from the reference cluster RC. To this end, for each measured variable Mj, the proportional contribution of the rate of change vj(ti) of the measured value mj(ti) of the corresponding measured variable Mj relative to the distance of the vector VR from the reference cluster RC is determined. The measured value Mj(ti) of at least one of the measured variables Mj is then identified as the cause of the anomaly. In this regard, the measured value mj(ti) of the measured variable Mj with the maximum proportional contribution is determined as the cause of the anomaly. Alternatively or additionally, the measured values mj(ti) of those measured variables Mj whose proportional contribution exceeds a predetermined limit value are determined as the cause of the anomaly.

[0090] The determination of the cause of the anomaly is also performed, for example, by means of the data processing device 11, and information about the determined cause of the anomaly is made available. The latter is output, for example, via the interface 19 connected to the data processing device 11 and / or is made available to the data processing device 11 in a form that is readable and / or retrievable via the interface 19.

[0091] The cause of the anomaly makes it easier for the operator of the device to determine the reason behind the anomaly and to take corresponding countermeasures. For this purpose, it is optionally checked whether the anomaly detected, for example, by a monitoring method, is caused by a change in at least one measured variable Mj that deviates from the expected behavior over time or by a functional damage of the measuring instrument MDj that measures the corresponding measured variable Mj. For this purpose, for example, a method is used that first checks the measurement accuracy of the measuring instrument MDj identified via the cause of the anomaly. For example, this can be done based on a reference measurement performed during ongoing operation of the device or within the framework of a calibration method. If a functional damage of one of the measuring instruments MDj is detected in this case, the corresponding measuring instrument MDj is recalibrated or replaced. If the measuring instrument MDj is not functionally damaged, the cause is preferably determined by first checking the device area and the device control area and / or the adjustment area that have an impact on the measured variable Mj identified via the cause of the anomaly.

[0092] Reference Signs List

[0093] 1 container 13 storage

[0094] 3 Feed Lines 15 Edge Devices

[0095] 5 Mixer 17 Parent Unit

[0096] 7 Extraction pipeline 19 interface

[0097] 9 Filling material

[0098] 11 Data processing device

Claims

1. A method for monitoring a measured variable that varies with time during a process and is measured continuously at a process device by a set of measuring instruments used on the device during repeated execution of a predetermined dynamic process, wherein: continuously acquiring, during repeated execution of the dynamic process, measurement data (D), said measurement data (D) comprising measured values (mj(ti)) of said measured variables (Mj) measured by means of said measuring instruments (MDj) and associated measurement times (ti) at which said measured values (mj(ti)) were measured, The measurement data (D) includes training data (DT) measured during a training period and monitoring data (DU) measured after the training period. continuously determining a vector (VR) based on the measurement data (D), the vector components of which comprise the respective rates of change of the measured values of the respective measured variables determined based on the measurement data (D) for a series of k consecutive time points ([tr1, ..., trk]), wherein each series of time points ([tr1, ..., trk]) covers a time window of a predetermined duration (ΔT), and the time windows of consecutive vectors (VR) are in each case offset relative to one another by a predetermined time difference, Vectors (VR) determined based on said training data (DT) are detected and stored in the form of reference clusters (RC), and Execute a monitoring method, where: comparing the vectors (VR) determined based on the monitoring data (DU) with the reference clusters (RC), respectively, If at least one of said vectors (VR) determined based on said monitoring data (DU) is outside said reference cluster (RC), an abnormality is determined, and Information about detected anomalies is made available.

2. The method according to claim 1, wherein: Acquiring the measurement data (D) such that the measured values (mj(ti)) of the measured variables (Mj) and the associated measurement times (ti) are transmitted to a data processing device (11) and are at least temporarily stored in a memory (13) associated with the data processing device (11), and The data processing device (11) is designed to perform the determination of the vector (VR), the determination of the reference cluster (RC) and the monitoring method based on the measurement data (D), and the data processing device (11) is designed to output the information about the detected anomaly via an interface (19) connected to the data processing device (11) and / or to provide the information in a form that can be read out and / or retrieved via the interface (19).

3. The method according to claim 1, wherein The predetermined duration (ΔT) of the time window: is less than or equal to the process duration of the process, is greater than or equal to the duration of at least one dynamic process event occurring during each execution of said process, whereby at least two of said measured variables (Mj) vary with time, is greater than or equal to at least one reaction time respectively associated with one of said measured variables (Mj), by which the measured value (mj(ti)) of the respective measured variable (Mj) changes in response to a dynamic process event, and / or is greater than or equal to at least one time scale respectively associated with one of the measured variables (Mj), on which at least one time scale the measured value (mj(ti)) of the respective measured variable (Mj) varies over time.

4. The method according to any one of claims 1 to 3, wherein: The predetermined duration (ΔT) of the time window is determined based on the training data (DT), wherein: determining, based on the rate of change (vj(ti)) of the measured value (mj(ti)) of one of the measured variables (Mj), a start time Ta at which the rate of change (vj(ti)) of the measured value (mj(ti)) of the measured variable (Mj) exceeds a predetermined limit value, determining, based on the measured values (mj(ti)) of the other measured variables (Mj), for each of said other measured variables (Mj), a respective measured variable-specific end time (tej), at which end time (tej) the rate of change (vj(ti)) of the measured value (mj(ti)) of the respective other measured variable (Mj) exceeds a predetermined limit value for the first time after said start time (ta), For each measured variable (Mj), the difference (Δtj:=tej-ta) between the measured variable-specific end point (tej) and the start time (ta) is determined, The duration (ΔT) of the time window is determined such that it is greater than or equal to at least one of these differences (Δtj:=tej-ta).

5. The method according to any one of claims 1 to 3, wherein: The predetermined duration (ΔT) of the time window is determined based on the training data (DT), wherein: For at least one of the measurands (Mj), in each case: determining a time period during which the measured value (mj(ti)) of the corresponding measured variable (Mj) changes over time and an average value of the change rate (vj(ti)) occurring within the time period based on the change rate (vj(ti)) of the measured value (mj(ti)) of the corresponding measured variable (Mj), and determining a time scale corresponding to a period of time during which the measured value (mj(ti)) of the measured variable (Mj) changes at a rate of change (vj(ti)) corresponding to the mean value by a predetermined proportion of the measuring range of the measuring instrument (MDj) measuring the corresponding measured variable (Mj), a proportion greater than or equal to 40% of the measuring range, or a proportion greater than or equal to 60% of the measuring range, and The duration (ΔT) of the time window is determined such that the duration (ΔT) is greater than or equal to at least one of the time scales.

6. The method according to any one of claims 1 to 3, wherein: The time intervals between the consecutive time points (tr1, . . . , trk) within the time window are all equal to a predetermined constant discretization time.

7. The method according to claim 6, wherein: The discretization time is scaled such that it is small enough to resolve process dynamics of the process.

8. The method according to claim 6, wherein: The discretized time is determined based on the training data (DT), wherein: For at least one of the measured variables (Mj), in each case: determining, based on the rate of change (vj(ti)) of the measured value (mj(ti)) of the corresponding measured variable (Mj), a time period during which the measured value (mj(ti)) changes over time and an average value of the rate of change (vj(ti)) occurring within the time period, and determining a time scale corresponding to the period of time during which the measured value (mj(ti)) of the measured variable (Mj) changes at a rate of change (vj(ti)) corresponding to the mean value by reaching a predetermined percentage of the measuring range of the measuring instrument (MDj) measuring the corresponding measured variable (Mj), a percentage less than or equal to 10% of the measuring range, or a percentage less than or equal to 5% of the measuring range, and The discretization time is determined such that the discretization time is less than or equal to at least one of the time scales.

9. The method according to claim 6, wherein: The time windows of consecutive vectors (VR) are time-shifted relative to each other by a time difference that is greater than the discretization time and / or less than or equal to the duration (ΔT) of the time window.

10. The method according to any one of claims 1 to 3, wherein: The causes of the anomalies established by the monitoring method are determined and / or resolved by corresponding countermeasures.

11. The method according to any one of claims 1 to 3, wherein: At least one cause of anomaly is determined for at least one vector (VR) identified as being outside the reference cluster (RC) using the monitoring method based on a distance from the reference cluster (RC), wherein: determining for each measurand (Mj) the proportional contribution of the rate of change (vj(ti)) of the measured value (mj(ti)) of the corresponding measurand (Mj) relative to the distance of said vector (VR) from said reference cluster (RC), and determining as the cause of the anomaly the measured value (mj(ti)) of the measurand (Mj) whose proportional contribution has a maximum value and / or exceeds a predetermined limit value, and Provides information about the stated cause of the exception.

12. The method according to claim 11, wherein By checking whether the anomaly detected by the monitoring method is caused by a change over time of at least one of the measured variables (Mj) identified via a previously determined cause of the anomaly that deviates from the expected behavior or by a functional damage of the measuring instrument (MDj) measuring the corresponding measured variable (Mj), the cause of the anomaly determined by means of the monitoring method is determined based on at least one of the causes of the anomaly, wherein the measurement accuracy of the measuring instrument (MDj) identified via the cause of the anomaly is checked.

13. The method according to any one of claims 1 to 3, wherein: The training data (DT) are detected in a training period in which error-free operation of the device and the measuring instrument (MDj) can be assumed and the process is performed several times in succession.

Images