Computer-implemented method for quantifying importance of measured values in time series

Abstract

The invention relates to a computer-implemented method for quantifying the importance of measured values in a time sequence, the method comprising the following steps: providing a measured time sequence (S10), the measured time sequence comprising a sequence of measured values ordered in time; wherein the measured values have been acquired as process parameters of the industrial process; -providing a reference time sequence (S12); -determining a residual value for the measured time sequence by comparing the measured time sequence with a reference time sequence (S14); and determining an importance value (S16) for each measurement value of the measured time series from the residual values.

Description

TECHNICAL FIELD

[0001] The present invention relates to the field of monitoring and analysis of industrial processes, in particular data processing related thereto. BACKGROUND

[0002] Monitoring and analysis of industrial processes face increasing challenges due to the handling of large data volumes. This data abundance results from the implementation of sensors and loT devices in different areas of industry. While these technologies help to make processes more efficient and to predict failures, they are at the same time a complex problem.

[0003] Industrial facilities and machines produce extremely large amounts of data in real time. These data include parameters such as temperature, pressure, flow rate, acceleration due to motion or vibration, and many other parameters. It can be a challenge to simply collect and store these data, especially when they occur in large amounts and at high frequencies.

[0004] Furthermore, these data often have to be analyzed in real time to identify potential problems or anomalies. This requires advanced analysis methods such as machine learning and artificial intelligence. However, processing such large data volumes also requires huge computing resources and can lead to delays that can affect the ability for real-time monitoring and control. Depending on the detected anomaly, the respective production process can be intervened or controlled, for example, where the respective product is discarded or marked as abnormal or production is stopped.

[0005] To reduce the data volume when monitoring and analyzing industrial processes, various methods are used that aim to extract important information and minimize unnecessary data. One of the most famous methods is data aggregation. Here, the raw data is compiled in order to reduce the number of data sets to be stored. For example, this aggregation can be done on a temporal basis, in a way that data is compiled over a certain time interval, or it can be done spatially, in a way that data of multiple sensors or sources is combined into a single data set.

[0006] Another method is to reduce the sampling or down-sampling. Here, data points are selected at a certain frequency in order to reduce the data volume while keeping important information. This can be particularly useful when data is collected at a higher frequency than needed for the analysis. However, it is difficult to distinguish which information is important and which is not.

[0007] Additionally, techniques such as feature extraction are often applied in data analysis in order to extract important information from the raw data and only store or analyze the information. Here, specific features or characteristics of the data of interest for analysis and monitoring are identified and extracted. This enables to reduce the data volume, where however important information can be lost.

[0008] It is therefore the task on which the present invention is based to propose a method with which the amount of data in the monitoring and / or analysis of an industrial process can be reduced without loss of important information.

[0009] This task is solved by the subject matter of the independent claims. SUMMARY

[0010] According to a first aspect of the present invention, the task is solved by a computer-implemented method for quantifying the importance of measured values in a time series, wherein the method comprises the following steps:

[0011] - providing a measured time series, wherein the measured time series comprises a sequence of measured values ordered in time, wherein the measured values have been acquired as process parameters of an industrial process;

[0012] - providing a reference time series;

[0013] - determining residual values for the measured time series by comparing the measured time series with the reference time series; and

[0014] - determining an importance value for each measured value of the measured time series from the residual values.

[0015] An industrial process can be any commercially applicable process for which at least one machine is used. In the scope of the present invention, a system denotes a combination of at least one machine, preferably a plurality of machines. The machines can perform all possible industrial processes. Industrial processes include, without limitation: conveying, cutting, drilling, milling, grinding, stamping, casting and injection molding, assembling, welding, painting, packaging, positioning, etc.

[0016] The measured values in the monitoring or analysis of an industrial process can thus be all physical parameters that can be acquired from the industrial process. These include, for example, the temperature of workpieces or machines and machine parts, the pressure in hydraulic or pneumatic systems, the movement speed of movable machine parts, the voltage or current in electronic components, etc.

[0017] The reference time series represents an ideal or average process development, while the measured time series represents the real, if necessary live, process. The reference time series and the measured time series thus relate to the same process parameters and have the same time division.

[0018] The measured values or data points of the time series can be acquired, for example, in the microsecond range, in the millisecond range or in the second range. In principle, longer intervals are also possible, although these are more likely to be rarely automated in this way and monitored due to the problem of the amount of data.

[0019] The measured time series is compared with a reference time series to highlight differences. These differences reveal where the measured time series deviates from the reference time series. These deviations are important for the monitored process, as deviations from the industrial process specifications often indicate errors in the process. Deviations are understood as residual values ​​for each measured value.

[0020] For each measured value of the measured time series, a significance value is determined from the residual value, so that a vector is generated for the entire measured time series, which indicates the significance of each individual measured value for monitoring or analyzing the industrial process. Thus, a significance map can be generated for the entire process, which can be used for monitoring or analysis to reduce processing effort.

[0021] The measured time series can be completely or partially below or above the reference time series. Both deviations increase the significance value. Therefore, when determining the significance value, it is preferred to use the absolute value of the residual value.

[0022] Preferably, a significance value for the measured value can be determined as the measured time series is acquired, so that the significance value can be used directly in real-time monitoring of the industrial process.

[0023] In this way, significance values ​​can be determined for the measured values ​​of multiple measured time series. For example, a reference time series can be used for an entire series of industrial processes. This method can also identify gradual changes that would not be detectable from individual time series, or even from a small number of time series, if the reference time series does not change.

[0024] The importance value specifies which measured values ​​are of particular interest for monitoring or analysis by reflecting their deviation from a target or normal state. For non-critical subprocesses, which generally do not change between executions, the importance value is zero or almost zero, since these subprocesses barely differ from the reference.

[0025] For example, an industrial process can be a drilling process in which the forces acting on the drill bit are monitored. The process begins with the drill bit being positioned on the workpiece. Before the drill bit is positioned, no forces are effectively acting on it. This means that the measured forces do not differ from the reference forces, or only differ within the measurement inaccuracies of the corresponding sensors. Therefore, the significance value is very small. Forces also come into play when the drill bit begins drilling. For example, if irregularities occur in the workpiece, such as if the material is brittle, the workpiece is not positioned correctly for the process, etc., the force curve during drilling may differ from that of the reference drilling. If the measured forces deviate from the reference forces, the significance value for these measured values ​​increases. During monitoring, the machine can then directly initiate countermeasures and, for example, correct or stop the process.

[0026] Irrespective of how the measured time series are further processed, the processing of the time series can be limited to the measured values of the time series which have a particularly high importance value. Thereby, the data to be examined is reduced to the relevant minimum, thereby sparing computing resources and thus simplifying the overall process. The amount of measured values which are marked as important can be set, for example, by a threshold value. The invention thereby solves its task.

[0027] In one embodiment, the residual values are determined from a difference between the measured time series and the reference time series.

[0028] Forming a difference between the measured time series and the reference time series is a simple possibility for determining the residual values. For this, the monitoring system can have stored the reference time series in a working memory, for example. Upon acquisition of each measured value, a difference to the corresponding value from the reference time series can be formed directly and the result stored. In this way, a particularly simple form of the invention is realized in an advantageous manner.

[0029] In one embodiment, determining the residual values comprises:

[0030] - determining spline coefficients for the reference time series;

[0031] - determining spline coefficients for the measured time series;

[0032] - determining one or more differences between the spline coefficients of the reference time series and the measured time series, wherein the residual values are determined from the one or more differences between the spline coefficients;

[0033] - inverse transforming the differences of the spline coefficients into the time domain and assigning the residual values of the spline coefficients to the corresponding measured values of the measured time series.

[0034] In this embodiment, the determination of the residual values is slightly more complex. In this embodiment, the time series can be divided into a plurality of segments, to which segments a spline is assigned.

[0035] A spline, also known as a polynomial motion trajectory, is a mathematical construct used in curve approximation and interpolation. Here, a curve is approximated by a series of polynomial segments, which are connected to one another in order to form a smooth and continuous curve. Each polynomial segment is referred to as a segment of the spline and is usually represented by a low-order polynomial, for example a quadratic or cubic polynomial. Splines are used to describe complex shapes and curves. They enable flexible and efficient representation of curves, which can be easily adapted to different requirements. The coefficients of these polynomial segments are referred to as spline coefficients.

[0036] The transformation of the time series into a spline can be used to limit the data to be monitored or to be analyzed to a small number of parameters. Thereby the amount of data to be processed is reduced, which is particularly resource-saving, especially when using machine learning algorithms.

[0037] In an embodiment, the importance value corresponds to a normalized residual value.

[0038] In a particularly simple manner, the importance value can be a normalized residual value. Normalization is important for comparability. Although it appears that a high residual value initially has a particularly high importance. However, if the residual value is large because the values of the measured time series and the reference time series are also large, then at this point size is not necessarily a powerful force for importance.

[0039] In the normalization, the residual value is made proportional to the values of the time series, preferably to the reference time series. Thereby the determination accuracy of the importance of the individual measurement values can be improved for the subsequent plausibility of the measured time series.

[0040] In an embodiment, the reference time series is an average time series composed of a plurality of measured time series.

[0041] The reference time series can be determined and provided in various ways. One possibility is to use historical data and to use an average value. For example, a fixed number of measured time series can be stored in the middle, from which an average time series can then be formed, which in turn forms the reference time series.

[0042] For this, for example, 100 or 1,000 time series can be used. The more time series are used, the more representative the reference time series is. However, as the number of time series increases, the storage and computing requirements for determining the reference time series also increase. When the reference time series is formed from the past x real-time measurements, the computing resources and storage requirements can be a limiting resource, especially when continuously monitoring in real time.

[0043] In an embodiment, the reference time series has been determined with a model of the industrial process.

[0044] Instead of using historical data, it is also possible to use synthetic time series. For example, a synthetic time series can be generated by a model in such a way that the depicted industrial process is simulated and / or the change process of the monitored process variable is calculated.

[0045] The advantage of a synthetic reference time series is that it can be generated in advance and does not tie up additional computing resources when monitoring in real time. Depending on how accurately the model works, the error range of the measurement values can also be taken into account when generating the reference time series, so that the error can also be estimated even when calculating the residual value.

[0046] In one embodiment, the method further comprises:

[0047] - determining an anomaly in the measured time series, wherein an anomaly is one or more measurement values having an importance value above a defined threshold value.

[0048] Anomaly detection is a method of data analysis in which certain deviations from the norm or statistically expected behavior in an industrial process are automatically identified. Here, the goal is to determine whether a data point deviates from the normal or set pattern or behavior by analyzing the data. For this, a reference must first exist by which it is determined which behaviors can be considered normal.

[0049] If the importance value of a measurement value or a series of measurement values is particularly high, the measured time series deviates particularly strongly from the reference time series at this point. The sensitivity of the anomaly detection can be determined here by a threshold value. The smaller the threshold value, the more likely deviations are captured as anomalies. If the threshold value is higher, the difference between the measured time series and the reference time series must also be greater for an anomaly to be identified.

[0050] In one embodiment, the method further comprises:

[0051] - providing a further reference time series;

[0052] - providing a further measured time series, wherein the further measured time series comprises measurement values of a further process parameter of the industrial process;

[0053] - determining residual values for the further measured time series;

[0054] - determining an importance value from the residual values for each measurement value of the further measured time series; and

[0055] - determining important time points for the industrial process from the importance values of the measured time series.

[0056] When monitoring or analyzing an industrial process, several process parameters can be captured at the same time. For example, when drilling, the drilling depth and the pressure on the drill bit can be captured. If several parameters are captured, the parameters can also be monitored in parallel or evaluated sequentially.

[0057] When monitoring and / or analyzing several parameters, several reference time series must exist, i.e. at least one reference time series for each process parameter, so that the captured data can be compared with the respective reference values.

[0058] The importance values of all measured time series can be compiled into an importance time series for the entire process, wherein each of the monitored parameters provides a part of the importance values. From the importance time series, the most important time points can be determined, which represent the most important, that is to say, in particular, the process steps that deviate most from the norm.

[0059] In another aspect, the present application relates to a computer-implemented method for training a machine learning algorithm, wherein the training comprises the following steps:

[0060] - providing a training data set, wherein the training data set comprises a plurality of measured time series, wherein for each measurement value of a time series an importance value is determined with the method according to any of the preceding claims;

[0061] - inputting the training data set into the machine learning algorithm for training the machine learning algorithm; and

[0062] - providing the trained machine learning algorithm.

[0063] The above-described method can be used to train a machine learning algorithm in order to prepare training data for the training. In the pre-processing, the measured time series can be combined with the importance values in order to show the machine learning algorithm to be trained which parts of the respectively input time series are most important for the evaluation.

[0064] By marking the most important time points in an industrial process, the training of a machine learning algorithm can be improved, so that the trained machine learning algorithm can better solve its task.

[0065] Furthermore, the machine learning algorithm can be trained with time series of a plurality of parameters. Here, for each time series of each parameter, an own importance value is provided, so that the machine learning algorithm can independently process the time series and can be adapted to the importance values.

[0066] In an embodiment, the machine learning algorithm comprises a plurality of models, wherein the measured time series each comprise at least one range. Each model is assigned to one or more ranges and is configured to process the measurement values of the one or more ranges assigned to it,

[0067] The ranges are determined with the following steps:

[0068] - determining measurement values in the measured time series, the importance values of which are local maxima or exceed a defined first threshold value,

[0069] - determining a range around the found measurement values, in which the importance values are above a defined second threshold value.

[0070] The reasons for the deviation of the measured time curve from the corresponding reference time sequence can be manifold. In correspondence therewith, it can be advantageous to use different models for monitoring and / or analysis for different process steps.

[0071] According to the importance values, the measured time sequences can be divided into different ranges. The data from these ranges can then be input into different models, which are selected respectively according to the proposed task. Thereby, the results of the monitoring or analysis can be improved.

[0072] The models can comprise, for example, but not limited to, linear models, decision trees, support vector machines, neural networks and more others.

[0073] The ranges to be selected can be set by means of a first threshold value and a second threshold value. The lower the threshold values, the larger the ranges, and vice versa.

[0074] In an embodiment, the machine learning algorithm is furthermore trained to process range parameters, wherein the range parameters of at least one range comprise a start of the range, an end of the range, a mean value and / or a median value of the measured time sequences in the range, a standard deviation of the measured time sequences in the range, a maximum importance value in the range, an average importance value in the range and / or other values characterizing the range.

[0075] The mentioned parameters can be determined for each range in a simple manner. By selecting the range parameters, the amount of data to be processed by the machine learning algorithm is significantly reduced, so that the required resources for the processing can also be reduced. Depending on the model, the model can even be simplified.

[0076] In an embodiment, each of the measurement values is assigned a weight for training the machine learning algorithm, wherein the weight is related to the importance of the respective measurement value.

[0077] In some models, the training data can be equipped with weights in order to reflect the importance of the respective data points or data packets for the training. Advantageously, the importance values can be used directly or indirectly as weights.

[0078] In another aspect, the present application relates to a computer program having a program code for performing the method as described above when the computer program is executed on a computer.

[0079] In another aspect, the present application relates to a computer-readable data carrier having a program code of a computer program for performing the method as described above when the computer program is executed on a computer.

[0080] In another aspect, the present application relates to a system for quantifying the importance of measured values in a time series, wherein the system is configured to perform the method as described above.

[0081] In general it can be determined that with the present application a method for quantifying the importance of measured values in a time series, a method for training a machine learning algorithm, a computer program, a computer readable data carrier having program code and a system for quantifying the importance of measured values in a time series are explained.

[0082] The described design solutions and refinements can be arbitrarily combined with each other.

[0083] Further possible design solutions, refinements and implementation forms of the present application include combinations of the features of the present application, which have not explicitly been mentioned before or below with respect to the embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0084] The accompanying drawings are included to provide a further understanding of embodiments of the application. The drawings illustrate an embodiment and together with the description serve to explain the principles and operation of the application.

[0085] Many of the advantages mentioned and other embodiments are derived from the drawings. The elements shown in the drawings are not necessarily to scale with each other.

[0086] Figure 1 The flow of the method according to the embodiments is schematically illustrated. DETAILED DESCRIPTION

[0087] In the drawings of the figures, like reference numerals indicate like or functionally similar elements, components or assemblies, wherever possible, unless otherwise indicated.

[0088] Figure 1 The flow of the method according to the embodiments is schematically illustrated.

[0089] The method starts in step S10 with providing a measured time series. The time series can have been collected or generated by any sensor or measuring device. Further, the time series can also represent process parameters read out from an industrial machine. For example, the time series can represent a voltage variation process in an electric drive. Thus, the time series can be provided by a sensor, a measuring device, etc. or loaded from an intermediate storage of a computing system.

[0090] In any case, the time series represents a physical quantity that has been collected at the time of an industrial process. The measured time series can be represented as f(t).

[0091] In an embodiment, a plurality of time series can be provided, which reflect the same physical quantity each. This can be the case, in particular, in the case of repeatable industrial processes, in particular work routines which are to be carried out identically by a work machine at all times. Even when workpieces are manufactured which have properties which remain constant at all times, a measured time series can be provided for each workpiece.

[0092] In addition to the measured time series, a reference time series is provided in step S12. The reference time series states a target value for each point in time of the industrial process. The reference time series can thus be used as an orientation of the industrial process which should be run. The reference time series can be represented as f ref (t).

[0093] If the measured time series map a plurality of process parameters of the same industrial process, that is to say for example when the plurality of process parameters has been captured in parallel or partially in parallel at the same facility, one reference time series must be provided for each process parameter.

[0094] The time series are compared with one another in step S14. In the comparison, residual values are produced which state to what extent the measured time series deviates from the reference time series. The residual values can preferably be formed in two ways.

[0095] The first possibility consists in subtracting the measured time series from the reference time series and using the absolute value of the result as the residual value. The residual value can be described as a function with:

[0096] f res (t) = |f(t) - f ref (t) |.

[0097] In the second possibility, the measured time series and the reference time series are approximated by splines. The absolute value of the difference of these spline coefficients produces a new function in the time domain, from which the residual value can be read.

[0098] The approximation can be represented by an n-th spline polynomial, for example, with

[0099]

[0100] For the residual value, for example, it is then possible to derive

[0101]

[0102] where

[0103]

[0104] In a last step S16, for each measurement value an importance value is determined from the residual value assigned thereto. Preferably, the importance value can be determined by normalizing the residual value. Normalization advantageously leads to the fact that importance values of different time series, even of different dimensions, can be compared with each other. For example, it is thereby possible to compare the importance when measuring temperature with the importance of the measurement of force with each other.

[0105] The importance value can be calculated, for example, with:

[0106]

Claims

1. A computer-implemented method for quantifying the importance of measurements in a time series, The method comprises the following steps: - providing a measured time series (S10), wherein the measured time series comprises a temporally ordered sequence of measurement values, wherein the measurement values ​​have been acquired as process parameters of the industrial process; - providing a reference time series (S12); - determining a residual value for the measured time series by comparing the measured time series with a reference time series (S14); and - Determining a significance value for each measured value of the measured time series from the residual value (S16). 2 . The computer-implemented method of claim 1 , wherein the residual value is determined from a difference between the measured time series and a reference time series.

3. The computer-implemented method of claim 1 , wherein determining the residual value (S14) comprises: - Determine the spline coefficients for the reference time series; - determining the spline coefficients for the measured time series; - determining a difference or differences between the spline coefficients of the reference time series and the measured time series, wherein a residual value is determined from the difference or differences of the spline coefficients; The differences of the spline coefficients are transformed back into the time domain and the residual values ​​of the spline coefficients are assigned to the corresponding measured values ​​of the measured time series.

4. A computer-implemented method according to any one of the preceding claims, wherein the importance value corresponds to a normalized residual value.

5. The computer-implemented method of any of the preceding claims, wherein the reference time series is an average time series composed of a plurality of measured time series, or wherein the reference time series has been determined using a model of the industrial process.

6. The computer-implemented method of any preceding claim, wherein the method further comprises: - determining anomalies in the measured time series, wherein an anomaly is one or more measurement values ​​having a significance value above a defined threshold.

7. The computer-implemented method of any one of the preceding claims, wherein the method further comprises: - provide another reference time series; - providing a further measured time series, wherein the further measured time series comprises measured values ​​of a further process parameter of the industrial process; - determining a residual value for said further measured time series; - determining a significance value for each measured value of another measured time series from the residual value; and - determining important time points for the industrial process from the importance values ​​of the measured time series.

8. A computer-implemented method for training a machine learning algorithm, wherein the training comprises the following steps: - provide training datasets, The training data set includes multiple measured time series. wherein for each measured value of the time series, a significance value is determined using a method according to any of the preceding claims; - Input the training data set into the machine learning algorithm to train the machine learning algorithm; and -Provide trained machine learning algorithms.

9. The computer-implemented method of claim 8 , wherein the machine learning algorithm comprises a plurality of models, wherein the measured time series each comprise at least one range, wherein each model is assigned to one or more ranges and is configured to process the measured values ​​of the one or more ranges assigned thereto, The range is determined using the following steps: - determining a measured value in the measured time series, the significance value of which is a local maximum or exceeds a defined first threshold value, - determining a range around the found measured value in which the significance value lies above a defined second threshold value.

10. A computer-implemented method according to any one of claims 8 or 9, wherein the machine learning algorithm is further trained to process range parameters, wherein the range parameters of at least one range include the start of the range, the end of the range, the average and / or median of the time series measured in the range, the standard deviation of the time series measured in the range, the maximum importance value in the range, the average importance value in the range and / or other values ​​characterizing the range.

11. A computer-implemented method according to any one of claims 8 or 10, wherein each of the measurements is assigned a weight for training a machine learning algorithm, wherein the weight is related to the importance of the corresponding measurement. 12 . A computer program comprising a program code for executing the method according to claim 1 , when the computer program is executed on a computer.

13. A computer-readable data carrier having a program code of a computer program for carrying out the method according to any one of claims 1 to 7 when the computer program is executed on a computer. 14 . A system for quantifying the importance of measured values ​​in a time series, wherein the system is designed to carry out the method according to claim 1 .

Images