Method for determining at least one point of change in time sequence of sensor values

Abstract

The invention relates to a method for determining at least one point of change in a time sequence of sensor values, wherein the method (1) comprises the following steps: providing a time sequence of sensor values and dividing the time sequence of sensor values into at least one evaluation window (2); determining, for each of the at least one evaluation window, at least two different characteristics (3) of the sensor values contained in the corresponding evaluation window; determining, for each of the at least one evaluation window, whether the sensor values contained in the corresponding evaluation window have at least one change point (4), respectively, on the basis of at least two different characteristics of the sensor values contained in the corresponding evaluation window; and-providing information (5) about the determined change point.

Description

Technical Field

[0001] The present invention relates to a method for determining at least one change point in a time series of sensor values, by which the change point in a continuous multidimensional time series of sensor values ​​can be determined, or a plurality of different types of change points can be determined simultaneously. Background Art

[0002] In this case, a change point is understood to be a significant or sudden change, in particular an unexpected change, in a signal, for example in a time series of sensor values. Such a change point can in this case, for example, indicate an anomaly, such as a corresponding anomaly in a manufacturing process.

[0003] A sensor - also called detector, (measurement variable or measurement) recorder or (measurement) probe - is a technical component which can detect certain physical or chemical properties and / or material attributes of its environment qualitatively or quantitatively as a measurement variable.

[0004] Such sensors are usually designed to continuously detect values, such as pressure, temperature, vibration or acceleration values ​​during the production process. The continuously detected values ​​form a time series of sensor values, or a corresponding time series is output at the output of the corresponding sensor in the form of an electrical signal or a digital signal.

[0005] For example, in order to be able to detect anomalies, in particular anomalies in manufacturing processes, corresponding time series of sensor values ​​are usually analyzed and it is determined whether they have change points. However, it is disadvantageous in this case that the known methods for change point detection are not designed to determine change points in a continuous multidimensional time series of sensor values ​​or to simultaneously evaluate a plurality of different features of the time series of sensor values, based on which in particular different types of change points can be respectively determined.

[0006] A method for evaluating a data-based sensor model to determine the instant of a change point in a sensor signal time series is known from the printed document DE 102022200284A1, wherein an evaluation signal time series is provided in an evaluation time window of the sensor signal time series, sensor signal segments which are time-shifted or offset from each other by a plurality of sampling steps are determined from the evaluation signal time series, wherein the sensor signal segments have a length which is smaller than the length of the evaluation signal time series, one or more frequency components are determined from the sensor signal segments, and wherein the frequency components are evaluated in a trained data-based sensor model to determine the instant of the change point within the evaluation time window. Summary of the invention

[0007] The object of the present invention is therefore to provide an improved method for identifying change points in a time series of sensor values.

[0008] This object is achieved by a method for identifying at least one change point in a time series of sensor values ​​according to the features of claim 1 .

[0009] Furthermore, the object is achieved by a system for identifying at least one change point in a time series of sensor values ​​according to the features of claim 8 .

[0010] According to one embodiment of the present invention, the task is solved by a method for determining at least one change point in a time series of sensor values, wherein a time series of sensor values ​​is provided and the time series of sensor values ​​is divided into at least one evaluation window, at least two different characteristics of the sensor value contained in the corresponding evaluation window are determined for each of the at least one evaluation window, and for each of the at least one evaluation window, it is determined based on the at least two different characteristics of the sensor value contained in the corresponding evaluation window whether the sensor value contained in the corresponding evaluation window has at least one change point, and information about the determined change point is provided.

[0011] In this case, an evaluation window is understood to be a finite and continuous or temporally related segment from a continuous time series.

[0012] Features are also understood to be characteristics or properties, in particular statistical properties, of the sensor values ​​contained in the corresponding section of the time series.

[0013] Different features are understood here in particular to mean that each of the features has different advantages in detecting a change point, in particular each of the features is suitable for detecting a different change point.

[0014] Determining and evaluating or analyzing at least two different characteristics, in particular at least two characteristics, each of which is suitable for detecting a further change point, has the advantage that change points or different types of change points in a continuous multidimensional time series of sensor values ​​can be determined simultaneously using a method. This ensures, for example, that anomalies in a manufacturing process can be detected at an early stage.

[0015] Furthermore, evaluating only the individual evaluation windows in each case has the advantage that the amount of data or information evaluated at the same time can be significantly reduced and thus the resources required for this, in particular storage and / or processing capacity, can also be significantly reduced.

[0016] In summary, an improved method for identifying change points in a time series of sensor values ​​is therefore proposed.

[0017] The step of determining for each of the at least one evaluation window whether the sensor value contained in the corresponding evaluation window has at least one change point may include comparing the information based on at least one of the at least two features with at least one limit value.

[0018] A limit value is understood here to be a value which can indicate a significant or sudden, in particular unexpected, change in the corresponding sensor signal.

[0019] The change point can therefore be determined in a simple manner by known methods or by comparison with limit values ​​without requiring a complex and resource-intensive adaptation.

[0020] In addition, the step of determining, for each of the at least one evaluation window, whether the sensor value contained in the corresponding evaluation window has at least one change point may include: applying a machine learning algorithm, which is trained to determine the change point in the information of at least one feature based on the sensor value based on the information of at least one feature of the at least two features.

[0021] Machine learning algorithms are based on the use of statistical methods to train a data processing device so that it can perform a specific task without being explicitly programmed for this purpose. The goal of machine learning is to build algorithms that can learn from data and make predictions. These algorithms create mathematical models with which data can be classified, for example.

[0022] This has the advantage that the change points can be determined in a simple manner, automatically, with high accuracy and independently of expert knowledge.

[0023] In one embodiment, the method further comprises the step of combining at least a portion of at least two different features into a common feature for each of the at least one evaluation window, and wherein the step of separately determining for each of the at least one evaluation window whether a sensor value contained in the corresponding evaluation window has at least one change point comprises: determining whether a sensor value contained in the corresponding evaluation window has at least one change point based on the common feature.

[0024] In this case, a common feature is understood to be the sum of at least parts of at least two different features, ie data or information representing parts of at least two different features, or an aggregation of parts of at least two different features.

[0025] The determination of the presence of at least one change point based on a common feature instead of based solely on at least two different features has the advantage that the corresponding analysis or evaluation is greatly simplified, in particular for a corresponding expert.

[0026] Furthermore, the step of determining for each of the at least one evaluation window whether the sensor value contained in the corresponding evaluation window has at least one change point may include applying a normalization method to the information based on at least one of the at least two features.

[0027] In this case, a normalization method is understood to be a standardization method which serves to unify or standardize values ​​or information representing corresponding features, thereby improving the comparability of corresponding features.

[0028] This has the advantage of further simplifying the identification of change points.

[0029] Furthermore, the time series of sensor values ​​may in particular be a time series of sensor values ​​detected during the execution of the manufacturing process which is representative of the manufacturing process.

[0030] In this context, a manufacturing process is understood to be a standardized workflow for manufacturing a product.

[0031] In particular, anomalies in the production process, such as anomalies in the operation of individual workstations or machines, can be identified based on the detected change points. Based on the corresponding identified anomalies, the correspondingly manufactured components can then be automatically discarded, for example.

[0032] In this case, the production process may be, for example, a workflow for producing assembled semiconductor wafers.

[0033] Using another embodiment of the present invention, a method for identifying anomalies in a manufacturing process is also described, wherein the method comprises: during the execution of the manufacturing process or during the operation of the manufacturing process, detecting a time series of sensor values ​​representing the manufacturing process, determining at least one change point in the time series of the detected sensor values ​​by the method described above for determining change points in the time series of sensor values, and checking at least one change point to identify anomalies in the manufacturing process.

[0034] Thus, a method for identifying anomalies in a manufacturing process is described, which method is based on an improved method for identifying change points in a time series of sensor values. Determining and evaluating or analyzing at least two different features, in particular at least two features, wherein each of the at least two features is suitable for detecting other change points, has the advantage that change points or different types of change points in a continuous multidimensional time series of sensor values ​​can be determined simultaneously using a method. This can ensure, for example, that anomalies in a manufacturing process can be identified as early as possible. Furthermore, evaluating only the individual evaluation windows separately has the advantage that the amount of data or information evaluated at the same time can be greatly reduced, and thus the resources required for this, in particular storage and / or processing power, can also be greatly reduced.

[0035] In addition, another embodiment of the present invention also describes a system for determining at least one change point in a time series of sensor values, wherein the system has at least one sensor for detecting the time series of sensor values ​​and at least one computing unit for processing the detected time series of sensor values, and wherein the system is configured to execute the method described above for determining at least one change point in a time series of sensor values.

[0036] Thus, an improved system for determining change points in a time series of sensor values ​​is described. The determination and evaluation or analysis of at least two different features, in particular at least two features, wherein each of the at least two features is suitable for detecting the other change point, has the advantage that change points or different types of change points in a continuous multidimensional time series of sensor values ​​can be determined simultaneously using a method. This can ensure, for example, that anomalies in a manufacturing process can be identified as early as possible. Furthermore, the evaluation of only individual evaluation windows has the advantage that the amount of data or information evaluated at the same time can be greatly reduced, and thus the resources required for this, in particular storage and / or processing power, can also be greatly reduced.

[0037] Using another embodiment of the present invention, a system for identifying anomalies in a manufacturing process is also described, wherein the system has: at least one sensor for detecting a time series of sensor data representing a manufacturing process during execution of the manufacturing process; a determination unit, which is configured to determine at least one change point in the time series of detected sensor values ​​by using the system described above for determining at least one change point in the time series of sensor values; and an inspection unit, which is configured to inspect at least one change point to identify anomalies in the manufacturing process.

[0038] A system for identifying anomalies in a manufacturing process is therefore described, which is based on change points identified by an improved system for identifying change points in a time series of sensors. Determining and evaluating or analyzing at least two different features, in particular at least two features, wherein each of the at least two features is suitable for detecting other change points, has the advantage that change points or different types of change points in a continuous multidimensional time series of sensor values ​​can be determined simultaneously using a method. This can ensure, for example, that anomalies in a manufacturing process can be identified as early as possible. Furthermore, evaluating only the individual evaluation windows separately has the advantage that the amount of data or information evaluated at the same time can be greatly reduced, and thus the resources required for this, in particular storage and / or processing power, can also be greatly reduced.

[0039] In summary, it can be determined that the present invention describes a method for determining at least one change point in a time series of sensor values, which method can be used to determine change points in a continuous multidimensional time series of sensor values, or multiple different types of change points can be determined simultaneously.

[0040] The described embodiments and developments can be combined with one another as desired.

[0041] Further possible embodiments, improvements and implementations of the invention also include combinations of features of the invention described above or below with respect to the exemplary embodiments that are not explicitly mentioned. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The attached drawings are intended to promote a further understanding of the embodiments of the present invention. These drawings illustrate the embodiments and, in conjunction with the description, are used to explain the principles and concepts of the present invention.

[0043] Further embodiments and many of the mentioned advantages can be derived with reference to the drawings. The elements shown in the drawings are not necessarily shown to scale with respect to one another.

[0044] Figure 1 A flow chart of a method for determining at least one change point in a time series of sensor values ​​according to an embodiment of the present invention is shown;

[0045] Figure 2 A part of a method for determining at least one change point in a time series of sensor values ​​according to an embodiment of the invention is shown; and

[0046] Figure 3 A schematic block diagram of a system for determining at least one change point in a time series of sensor values ​​according to an embodiment of the present invention is shown.

[0047] In the figures of the drawings, identical reference signs denote identical or functionally identical elements, components or parts, unless otherwise specified. DETAILED DESCRIPTION

[0048] Figure 1 The flow chart of the method for determining at least one change point in a time series of sensor values ​​according to an embodiment of the present invention is shown. Figure 1 .

[0049] Sensors are usually designed to continuously detect values, such as pressure, temperature, vibration or acceleration values ​​during the production process. The continuously detected values ​​form a time series of sensor values, or a corresponding time series is output at the output of the corresponding sensor in the form of an electrical signal or a digitized signal.

[0050] For example, in order to be able to detect anomalies, in particular anomalies in a manufacturing process, corresponding time series of sensor values ​​are usually analyzed and it is determined whether they have change points. However, it is disadvantageous in this case that known change point detection methods are not designed to determine change points in a continuous multidimensional time series of sensor values ​​or to simultaneously evaluate a plurality of different features of the time series of sensor values, based on which in particular different types of change points can be determined in each case.

[0051] Figure 1 In this case, a method 1 is shown, which includes step 2: providing a time series of sensor values ​​and dividing the time series of sensor values ​​into at least one evaluation window; step 3: for each of the at least one evaluation window, determining at least two different characteristics of the sensor value contained in the corresponding evaluation window; step 4: for each of the at least one evaluation window, determining whether the sensor value contained in the corresponding evaluation window has at least one change point based on at least two different characteristics of the sensor value contained in the corresponding evaluation window; and step 5: providing information about the determined change point.

[0052] Determining and evaluating or analyzing at least two different features, in particular at least two features, each of which is suitable for detecting a further change point, has the advantage that change points or different types of change points in a continuous multidimensional time series of sensor values ​​can be determined simultaneously using a method. This ensures, for example, that anomalies in a manufacturing process can be identified at an early stage.

[0053] Furthermore, evaluating only the individual evaluation windows in each case has the advantage that the amount of data or information evaluated at the same time can be significantly reduced and thus the resources required for this, in particular storage and / or processing capacity, can also be significantly reduced.

[0054] In summary, an improved method 1 for identifying change points in a time series of sensor values ​​is thus described.

[0055] In particular, a method 1 is described which is designed to determine in a simple manner and without complex and resource-intensive adaptations when there is a significant deviation from a normal behavior, ie the time of a change point.

[0056] Here, the method 1 is characterized by its flexibility, easy interpretability and effectiveness. Therefore, the method can be applied not only online but also offline to determine any type of changes. In addition, the features are very informative and can be evaluated in a simple way.

[0057] The time series can have a random, predefined length, wherein the individual evaluation windows can be broken down and analyzed directly after the end of the duration corresponding to the respective time window. In addition, the time series can also be divided into individual evaluation windows after the complete time series has been obtained, and then the individual evaluation windows can be analyzed separately. The individual time windows can also have the same length, but it is also possible for the individual evaluation windows to have different lengths.

[0058] Furthermore, no pre-setting or specification is required for the data distribution or the number of change points.

[0059] Depending on the application, the provision of sensor values ​​may also include oversampling and / or undersampling.

[0060] In this case, the at least two different features can be, in particular, statistical properties, such as a variance, a spectrum or a deviation from a mean value.

[0061] Here, a total value can be determined for the entire evaluation window, i.e., the corresponding values ​​in the evaluation window are summed. The value corresponding to the left boundary of the evaluation window and the value corresponding to the right boundary of the evaluation window can then be subtracted from the total value to determine the total change of the corresponding feature within the corresponding evaluation window.

[0062] according to Figure 1 In an embodiment, step 4 of determining for each of the at least one evaluation window whether the sensor value contained in the corresponding evaluation window has at least one change point comprises comparing information based on at least one of the at least two features with at least one limit value.

[0063] In particular, the determined total change of the corresponding feature within the corresponding evaluation window can be compared with at least one limit value.

[0064] In this case, the at least one limit value can be predetermined or can be based on a corresponding value from a previous evaluation window, for example a corresponding total change within a previous evaluation window.

[0065] The corresponding result, ie the position of the change point within the corresponding evaluation window, can also be output in the form of a Boolean vector.

[0066] according to Figure 1 In an implementation form, step 4 of determining for each of at least one evaluation window whether the sensor value contained in the corresponding evaluation window has at least one change point comprises: in addition, applying a machine learning algorithm to information based on at least one feature of the at least two features, the machine learning algorithm being trained to determine change points in information based on at least one feature of the sensor value.

[0067] The machine learning algorithm can be trained here, for example, based on corresponding labeled training data, for example training data representing the relationship between information representing a feature and a change point.

[0068] according to Figure 1 In an implementation form, the method further comprises step 6: for each of at least one evaluation window, at least a part of the at least two different features are combined into a common feature, wherein for each of at least one evaluation window, step 4 of determining whether the sensor value contained in the corresponding evaluation window has at least one change point comprises: determining whether the sensor value contained in the corresponding evaluation window has at least one change point based on the common feature.

[0069] In particular, in this case, depending on the application, individual features can be combined to form a common feature and other features can be evaluated individually.

[0070] Furthermore, the individual aggregated features can be weighted in each case during the combination, for example based on the respective importance or the respective influence on the respective application case.

[0071] according to Figure 1 In an embodiment of the present invention, the time series of sensor values ​​is furthermore a time series of sensor values ​​detected during the execution of the manufacturing process and representative of the manufacturing process.

[0072] In particular, this can be a time series of pressure, temperature, vibration or acceleration values ​​during a production process, for example of a workstation or a machine.

[0073] Based on the determined change points, it is then possible to check, for example, whether anomalies exist. If anomalies are detected, the corresponding manufactured products can be automatically discarded, for example. In addition, an alarm can also be automatically triggered to signal an expert to analyze the corresponding anomaly and its influence on the manufacturing process more precisely.

[0074] Figure 2 A part of a method for determining at least one change point in a time series of sensor values ​​according to an embodiment of the invention is shown.

[0075] In particular, Figure 2 Here, the temporal course of a time series of sensor values ​​10 is shown, wherein the horizontal axis represents time and the vertical axis represents the corresponding sensor value, and wherein, inter alia, change points 11 can also be identified.

[0076] also, Figure 2Different features 12, 13, 14, 15, 16 of a time series of sensor values ​​10 are shown, in particular a feature based on kernel interpolation 12, a feature based on regression analysis 13, a feature based on a spectrum 14, a feature based on deviations from the mean 15 and a feature based on deviations from the variance 16. Here again, the horizontal axis represents time and the vertical axis represents the corresponding value.

[0077] also, Figure 2 Also shown is the sum of the individual features 12 , 13 , 14 , 15 , 16 or an aggregation 17 of the individual properties 12 , 13 , 14 , 15 , 16 , wherein the horizontal axis again symbolizes time and the vertical axis symbolizes the corresponding values.

[0078] Figure 3 A schematic block diagram of a system 20 for determining at least one change point in a time series of sensor values ​​according to an embodiment of the present invention is shown.

[0079] like Figure 3 As shown, the system 20 comprises a sensor 21 for detecting a temporal sequence of sensor values, and a computing unit 22 for processing the temporal sequence of sensor values ​​detected by the sensor 21 .

[0080] like Figure 3 As further shown, the calculation unit 22 also has a division unit 23, which is configured to divide the time series of sensor values ​​into at least one evaluation window; a first determination unit 24, which is configured to determine at least two different characteristics of the sensor value contained in the corresponding evaluation window for each of the at least one evaluation window; a second determination unit 25, which is configured to determine whether the sensor value contained in the corresponding evaluation window has at least one change point based on the at least two different characteristics of the sensor value contained in the corresponding evaluation window for each of the at least one evaluation window; and a providing unit 26, which is configured to provide information about the determined change point.

[0081] The division unit, the first determination unit and the second determination unit can be implemented here, for example, respectively based on codes stored in a memory and executable by a processor. The providing unit can also be a transmitter, which is designed to transmit corresponding information or data.

[0082] according to Figure 3 In an embodiment of the present invention, the second determination unit 25 is designed to compare the information based on at least one of the at least two features with at least one limit value.

[0083] according to Figure 3In an embodiment, the second determination unit 25 is further configured to apply a machine learning algorithm to information based on at least one of the at least two features, the machine learning algorithm being trained to determine change points in information based on at least one feature of the sensor value.

[0084] like Figure 3 As shown, the system 20 shown also includes an aggregation unit 27, which is configured to combine at least a portion of at least two different features into a common feature, wherein the second determination unit 25 is configured to determine whether the sensor value contained in the corresponding evaluation window has at least one change point based on the common feature.

[0085] In this case, the aggregation unit can again be implemented, for example, based on a code stored in a memory and executable by a processor.

[0086] The determination unit 25 shown is also designed to apply a normalization method to the information based on at least one of the at least two features.

[0087] Furthermore, the time series of sensor values ​​is again a time series of sensor values ​​detected during the execution of the manufacturing process and representative of the manufacturing process.

[0088] Furthermore, the system 20 shown is configured to carry out the method described above for determining at least one change point in a time series of sensor values.

Claims

1. A method for determining at least one change point in a time series of sensor values, wherein the method (1) comprises the following steps: - providing a time series of sensor values ​​and dividing the time series of sensor values ​​into at least one evaluation window (2); - for each of the at least one evaluation window, respectively determining at least two different characteristics (3) of the sensor value contained in the corresponding evaluation window; - for each of the at least one evaluation window, determining whether the sensor value contained in the corresponding evaluation window has at least one change point (4) based on at least two different features of the sensor value contained in the corresponding evaluation window; and -Provide information about the determined change points (5).

2. The method (1) according to claim 1, wherein the step (4) of determining for each of the at least one evaluation window whether the sensor value contained in the corresponding evaluation window has at least one change point comprises: comparing information based on at least one feature of the at least two features with at least one boundary value.

3. The method (1) according to claim 1 or 2, wherein the step (4) of determining for each of the at least one evaluation window whether the sensor value contained in the corresponding evaluation window has at least one change point comprises: applying a machine learning algorithm to information based on at least one feature of the at least two features, the machine learning algorithm being trained to determine change points in information based on at least one feature of the sensor value.

4. The method (1) according to any one of claims 1 to 3, wherein the method further comprises the step of: for each of the at least one evaluation window, combining at least a portion of the at least two different features into a common characteristic (6), and wherein for each of the at least one evaluation window, the step (4) of determining whether the sensor value contained in the corresponding evaluation window has at least one change point comprises: determining whether the sensor value contained in the corresponding evaluation window has at least one change point based on the common characteristic.

5. The method (1) according to any one of claims 1 to 4, wherein the step (4) of determining for each of the at least one evaluation window whether the sensor value contained in the corresponding evaluation window has at least one change point comprises: applying a normalization method to information based on at least one of the at least two features.

6. The method (1) according to any one of claims 1 to 5, wherein the time series of sensor values ​​is a time series of sensor values ​​detected during the execution of the manufacturing process and representative of the manufacturing process.

7. A method for identifying anomalies in a manufacturing process, wherein the method comprises the following steps: - during the execution of a manufacturing process, detecting a time series of sensor values ​​representative of said manufacturing process; - determining at least one change point in a time series of detected sensor values ​​by means of a method for determining at least one change point in a time series of sensor values ​​according to any one of claims 1 to 6; and - Examining said at least one variation point to identify anomalies in said manufacturing process.

8. A system for determining at least one change point in a time series of sensor values, wherein the system (20) has at least one sensor (21) for detecting the time series of sensor values ​​and at least one computing unit (22) for processing the detected time series of sensor values, and wherein the system is configured to perform a method for determining at least one change point in a time series of sensor values ​​according to any one of claims 1 to 6.

9. A system for identifying anomalies in a manufacturing process, wherein the system comprises: at least one sensor for detecting a time series of sensor values ​​representing the manufacturing process during execution of the manufacturing process; a determination unit configured to determine at least one change point in the time series of the detected sensor values ​​by using a system for determining at least one change point in the time series of sensor values ​​according to claim 8; and an inspection unit configured to inspect at least one change point to identify anomalies in the manufacturing process.

10. A computer program having a program code so that when the computer program is executed on a computer, the method for determining at least one change point in a time series of sensor values ​​according to any one of claims 1 to 6 and / or the method for identifying anomalies in a manufacturing process according to claim 7 are executed.

Images