Determination of the cause of abnormal events
By identifying the time interval and periodicity of abnormal events in an automated system, using Fourier transform and autocorrelation technology, the problem that the existing technology cannot automatically identify the cause of abnormalities is solved, and automatic cause analysis and system management of abnormal events are realized Improved efficiency.
Patent Information
- Application Number
- CN201980025153.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2018-04-11
- Filing Date
- 2019-03-06
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2039-05-02
AI Technical Summary
The existing exception recognition methods cannot automatically identify the cause of abnormal events, resulting in the inability to automatically identify and resolve the cause.
By identifying the time interval between abnormal events in an automated system, and using technologies such as Fourier transform and autocorrelation to determine the periodicity of abnormal events, we can then identify possible causes, and realize causal analysis of abnormal events and automatic cause representation of abnormal events. .
It effectively identifies the potential causes of abnormal events, improves the operation and management efficiency of the automation system, and can be implemented in facilities or in cloud infrastructure, so as to achieve automatic correction and control of abnormalities.
Smart Images

Figure CN111954857B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the identification of abnormal events and their causal relationships in detected or measured data, and in particular to a method, system and computer program and computer program product for identifying abnormal events and their causal relationships within detected or measured data. Background Art
[0002] A method for detecting anomalies in data provided by each of a plurality of related sensors is known from the publication WO 2017086963 A1. Here, a time series data sequence is received by each sensor, and a digital representation is determined for each of the time series data sequences, and finally an abnormal value is determined for each of the time series data sequences under the condition of the digital representation determined for the time series data sequence, and a distribution of the determined abnormal values is obtained under normal conditions.
[0003] It is known from patent document US7310590B1 that the time series of multiple functions are analyzed simultaneously in order to identify the anomalies of the data points in the time series. The data point values are predicted by multiple functions. When the actual data points in the queue are obviously different from the predicted values of the data points generated by the functions, an anomaly occurs. When a statistically sufficient model determines the anomaly of the data point, an abnormal event is generated there. The set of functions can include different types of functions, the same function type configured with different constants, or their combination. In addition, a method for anomaly identification is also known from US 20120041575A1.
[0004] These known methods for anomaly identification do not identify or do not automatically identify the underlying problem, ie, the cause of one or more abnormal events, so that cause identification is not possible or cannot be automatically achieved. Summary of the invention
[0005] It is therefore an object of the present invention to identify one or more potential causes of an anomaly or one or more abnormal events and to propose improvements in anomaly identification.
[0006] This object is achieved by a method, an apparatus, a computer program, a computer program product and a data flow.
[0007] In a first embodiment, the object is achieved by a method for evaluating at least one data set of at least one component of an automation system. The method comprises and / or causes the following steps: determining a time interval between two abnormal events in a first data set comprising data about at least one first component of the automation system, and determining repeated events spaced apart according to the time interval in a second data set comprising data about at least one second component of the automation system.
[0008] The method is a computer-implemented method. In addition, the method can be applied in facilities of process technology or automation technology. Here, the method can be used to identify repeated process effects that are particularly (for example, in a second data set) encoded as events. Starting from this, the method can be used for operation management in a control and / or monitoring facility. Preferably, the method is applied in a facility with one or more repeated processing steps. For example, wear of a machine tool or a transmission can be identified, or changes in the settings of one or more devices of the facility for implementing processing steps can be identified. Other applications are, for example, changes in the lifting process and / or transportation process in a press. Other processes with cyclic processing steps are, for example, in the semiconductor industry. Here, the cyclic processing steps can exist periodically with the rhythm of the processing steps (for example, in installation, especially in the automotive industry). Subsequently, for example, the controller of the facility can be affected so that one or more of the repeated process effects are shifted or blocked, wherein one or more interventions in the controller of the facility are implemented. Here, the method can be implemented in the facility or on a remote computing unit, for example, in a cloud infrastructure. In addition, one or more steps of the method can be implemented in the facility, and one or more steps of the method can be implemented in a remote computing unit, for example, in a cloud infrastructure.
[0009] In a second specific embodiment, this object is achieved by a device for carrying out the method according to the first specific embodiment.
[0010] In a third embodiment, the object is achieved by a computer program having program code means for executing the method according to the first embodiment. Here, the computer program comprises program code means, and when the program code means is executed, the program code means is used to implement one or more method steps of the method according to the first embodiment.
[0011] In a fourth embodiment, the object is achieved by a computer program product having program code means. Here, the computer program product comprises program code means, which, when executed, are used to implement one or more method steps of the method according to the first embodiment.
[0012] In a fifth embodiment, the object is achieved by a data stream representing the computer program according to the third embodiment. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] The foregoing and other features of the present invention are apparent from the following detailed description taken in conjunction with the accompanying drawings.
[0014] Figure 1 A schematic diagram showing an automation system connected to a cloud infrastructure,
[0015] Figure 2A schematic diagram showing a first specific embodiment of a device for evaluating at least one data set, in particular for determining the cause of an abnormal event,
[0016] Figure 3 A schematic diagram showing a second embodiment of a device for evaluating at least one data set, in particular for determining the cause of an abnormal event,
[0017] Figure 4 A schematic diagram showing an embodiment comprising a plurality of method steps for determining the cause of an abnormal event by determining the time interval between the abnormal events,
[0018] Figure 5 A schematic diagram showing an embodiment comprising a plurality of method steps for determining a cause of an abnormal event by performing an autocorrelation based on a first data set,
[0019] Figure 6 A schematic diagram showing an embodiment comprising a plurality of method steps for determining a cause of an abnormal event by performing a cross-correlation based on a first data set and a second data set,
[0020] Figure 7 A schematic diagram showing an embodiment comprising a plurality of method steps for determining the cause of an abnormal event by compensating for changes in the time intervals between abnormal events,
[0021] Figure 8 A schematic diagram shows an embodiment comprising a plurality of method steps for determining the cause of an abnormal event by taking into account different operating states of an automation system,
[0022] Fig. 9 A schematic diagram showing an embodiment including a plurality of method steps for determining a cause of an abnormal event by determining various abnormal indicators,
[0023] Fig.10 A schematic diagram showing an embodiment comprising a plurality of method steps for determining a cause of an abnormal event by determining an additional abnormal indicator,
[0024] Fig.11 A schematic diagram showing an embodiment comprising a plurality of method steps for determining the cause of an abnormal event by determining an m-dimensional transformation,
[0025] Fig.12 A schematic diagram of an embodiment comprising a plurality of method steps for determining the cause of an abnormal event by further evaluation of an m-dimensional transformation is shown. DETAILED DESCRIPTION
[0026] In e.g. Figure 1In the automation system shown, multiple components, such as field devices, are used to detect and / or influence process variables. For detecting process variables, measuring devices such as fill level measuring devices, flow measuring devices, pressure and temperature measuring devices, pH measuring devices, conductivity measuring devices, etc. can be used, which detect the corresponding process variables fill level, flow, pressure, temperature, pH value or conductivity. For influencing process variables, actuators such as valves or pumps can be used, via which, for example, the flow of a liquid in a pipeline or the fill level of a medium in a container is changed. The term "component" used includes all types of measuring devices and actuators. In addition, all devices that are used in the vicinity of the process and provide or process process-related information are also considered components. In addition to the measuring devices / sensors and / or actuators mentioned above, components generally also include such units as remote I / O s , gateways, connection devices and wireless adapters, PLCs, etc., which are connected to a bus, such as a field bus, and are used to communicate with a superior unit. At the field level, one or more controllers PLC A and PLC B can be arranged according to the size of the facility. The communication between the controllers PLC A, PLC B and the components coupled to the controllers PLC A and PLC B is realized via at least one of the field buses commonly used in automation technology. At the control level, the controllers PLC A and PLC B provide the data collected from the components, such as measured values, and / or further processed measured values to the superior control unit (such as SCADA). The components of the automation system are designed to be networkable and interconnected via a network. Here, each component can correspond to a single address in the network. Communication can be realized by means of a network protocol. Therefore, the current controller is also understood as a component. In addition, one or more components can also be implemented purely in the form of software, such as in the form of a so-called (software) agent.
[0027] Figure 1 An automation system is shown which is connected to a cloud infrastructure 5. In order to establish a connection to the cloud infrastructure, a cloud interface is provided in the automation network 3 of the automation system. The cloud interface can be part of the communication device 1. The cloud interface can also be installed directly in the automation network 3, for example in one of the components of the automation system, for example as a (software) agent. The automation system can be, for example, Figure 1 One or more components 2a, 2b, 2c, 2d, 2e are shown.
[0028] In this case, the cloud infrastructure can have one or more servers, one or more platforms and / or one or more applications, i.e. computer programs, which run or can run on one or more platforms. The cloud infrastructure can be, for example, the so-called MindSphere of Siemens. In this case, the cloud infrastructure is implemented, for example, by one or more computing units, which preferably provide one or more cloud services.
[0029] Thus, the cloud infrastructure can be communicatively connected to the automation system 4. This connection can be set up, for example, via a so-called communication device, for example the so-called MindConnect Nano from Siemens. Thus, the communication device 1 can be used to receive data via one or more protocols and to enable the transmission of this data to the cloud infrastructure. In this case, the communication device 1 can support the transmission of encrypted data via a secure network connection in order to enable cloud-based applications and services.
[0030] However, large amounts of data may accumulate within the automation system and / or when data is transferred to the cloud infrastructure. These data are generated, for example, from multiple components (such as actuators, internal or external sensors, distance measuring devices and telemetry devices), which are usually monitored at a high frequency. The individually monitored components can be considered as dimensions, and therefore the collection of readings detected at a certain point in time can be considered as multidimensional points. Therefore, the method for generating anomaly scores for each given point can use calculations that take into account the density of the points, such as the Mahalanobis distance or K nearest neighbors (KNN), or reduce the dimensionality of the multidimensional points by means of principal component analysis or self-organizing maps, as shown in the published patent application US20130060524A1.
[0031] In this case, the data can represent one or more interrelated physical and / or chemical variables. These variables can be provided by components, such as sensors, actuators and / or controllers. The components can be, for example, components of the same type or components of different types. For example, they can be sensors that perform the same type or different types of measurements of one or more physical and / or chemical variables (such as temperature) and are preferably relatively close to each other (for example in industrial equipment).
[0032] An anomaly is generally defined as at least one data point that differs in its actual detected or measured value significantly enough from the detected or measured values of the rest of the data points in a group, template, chain, or data queue (referred to as a data set) to be marked as at least problematic. That is, for historical reasons or otherwise, the detected or measured value proposes an expected "normal" value or range of normal values for the sampled data, and the anomaly is a data value that does not match or does not match the normal value or range of normal values for the data accurately enough. Other common names for anomalies are outliers, deviations, anomalies, accidents, interferences, exceptions, etc. The group of data points that are sampled and often lead to anomalies can be called a time series, which is a sequence or template of data measured over a time interval in which each data point corresponds to a discrete point or detected value in that time (e.g., a data point detected every second over a period of one hour).
[0033] When an anomaly is detected or identified, the anomaly often triggers a subsequent process, such as identifying the cause of the anomaly and / or preventing the anomaly from causing damage to an automation system (e.g., a process control system or process) that includes or applies the data. The subsequent process can correct the problem in the automation system caused by the identified anomaly. Therefore, anomaly identification generally involves the identification of a template for a given data set, which template is inconsistent with an established, expected, or normal behavior data template. Generally, it is desirable to identify the anomaly as early as possible or as quickly as possible before the anomaly causes damage to the underlying data processing system and / or automation system. Here, the cause of the anomaly can be a change in a process running in the automation system.
[0034] In general, the role of technology is constantly becoming more important and new applications and application areas of existing technologies are being discovered every day. One of these areas is the use of sensors for detecting environments and for detecting control devices, for example in industrial applications and in everyday public use. Here, environmental sensors located outdoors can include, for example, temperature sensors arranged in different spaces of a house and a plurality of sensors of various types, for example in cars, trains, offices, factories and computer networks. One of the main purposes of the sensor detection mode is to identify and prevent faults in the control of equipment by identifying anomalies in the measurement data provided by the sensors as quickly as possible. As mentioned above, there are methods that can locate or determine anomalies in time series data.
[0035] In this context, in the industrial sector, for example in automation systems, process influences, in particular repetitive process influences, which are not recognized as a fundamental, industry-specific operation management (e.g. reliance on irregular pause times, etc.), can hinder the identification of causes, for example in transfer processes between stations in the semiconductor industry or in stamping plants.
[0036] In a first step, the identification of one or more anomalies or one or more abnormal events in one or more data sets can be suitably performed, and the data sets are provided, for example, by one or more components representing related variables. For this, for example, it is possible to realize the acceptance of one or more data sets in the form of one or more time series of each of a plurality of data sources (for example, one or more components, such as sensors, actuators or other process-oriented components). The data set can include a plurality of data, which are detected by the components at a sampling frequency. For example, it can be the raw data of one or more components. In addition, derived data can also be included in the data set. For example, a data set, for example a first and / or second data set, can include raw data and / or data calculated based on the raw data.
[0037] On the other hand, one or more abnormal events can be determined by the user, for example. This can be achieved, for example, by selecting one or more events from the data set. In addition, this can be achieved by selecting data in the data set, for example, via a display unit and / or an operating unit.
[0038] Here, the method (step) described in the scope of the current open document can be implemented by means of a device, which, for example, detects anomalies in one or more data sets provided by one or more components, and the device includes at least one processor, which is connected to one or more memories. The processor can be configured to receive, for example, a corresponding data set in the form of a time series from each of one, more or more components, wherein each data set preferably includes multiple data. The processor can also be configured, for example, to determine a digital representation for each of the time series, so as to determine an abnormal value for each time series when applying a determined number for each time series, so as to determine the distribution of the measured abnormal scores under normal conditions. A corresponding suitable system is described, for example, in WO2017086963A1.
[0039] A corresponding device can be used, for example, Figure 1In the embodiment shown, it is physically arranged in the automation system. The device can be, for example, a communication device 5. The device can be configured to receive data of each of one, more or more components 2a, 2b, 2c, 2d, 2e. Here, the device can communicate with each of the components 2a, 2b, 2c, 2d, 2e wirelessly or wired. Each component 2a, 2b, 2c, 2d, 2e can provide its data to the device, for example, in the form of a time series, so that each component can provide its data at a discrete time point (for example, once per second, once per minute, once per hour, etc.). This can be realized by, for example, each component 2a, 2b, 2c, 2d, 2e continuously providing its data and then the device periodically reading the data of each component in the desired time interval (for example, once per second, once per minute, once per hour, etc.). When the sampling frequency is not the same for all components 2a, 2b, 2c, 2d, 2e, it is possible to calculate or determine a statistical vector for each time series, such as a maximum value, a minimum value, an average value, a standard deviation, a high-order moment, etc. This enables direct comparison of the data for each time series.
[0040] In addition, as described above, an abnormality score can be determined in the step for each of the time series of the corresponding component (which can be a temperature sensor, for example) under the condition of applying the numerical representation calculated for each time series. For example, under the condition of applying the determined numerical data representation, the average distance (e.g., Euclidean, Manhattan or weighted) of the detected data about each component to other components is calculated or determined.
[0041] In an exemplary embodiment, the detected or measured data can then be processed to determine the presence of a single anomaly or multiple anomalies or a single abnormal event or multiple abnormal events, for example within a template or time series of the data. When one or more anomalies or abnormal events are determined, corrective action can be taken to determine the cause of the anomaly and / or prevent the underlying process control system.
[0042] exist Figure 2 Another embodiment of at least one data set for evaluating at least one component of an automation system is shown in . Here, one or more steps of the method can be executed by a computing unit or on site in the automation system. The computing unit can be communicatively connected to one or more components (e.g., a controller, a management system and / or an operation management system). In addition, further components not shown can be connected to the computing unit. Thus, one or more of these components can transmit data to the computing unit. Subsequently, these data can be deposited or stored in a database, which is preferably also connected to the computing unit.
[0043] Thus, in order to identify an anomaly or one or more abnormal events, to determine the time interval between two abnormal events in the first data set and / or to determine events in the second data set that are repeated and spaced apart according to a time interval, data from the database and / or (directly) data from the component can be used. Here, one or more events correspond to data from the first and / or second data set.
[0044] In addition, for example, one or more events found in the second data set can be displayed or found on a display connected to the computing unit in a communicative manner in order to initiate one or more corrective measures. Here, the display can be part of an application running in a browser of a user device, for example.
[0045] The proposed method can be implemented on-site on a computer unit in an automation system. For example, the events in the second data set that are repeated and spaced apart according to time intervals can be transmitted in the communication device according to Figure 1 In this case, for example, only the discovered events and / or the findings derived therefrom can then be transmitted to a cloud infrastructure and / or for display.
[0046] On the other hand, it is also possible Figure 3 It is possible to carry out the evaluation by means of a cloud infrastructure and / or by means of an application running in the cloud infrastructure. For this purpose, the data detected by one or more components can be processed by the application running in the cloud infrastructure. In addition, the embodiment in which the application is run in the cloud infrastructure has the advantage that a plurality of automation systems or subsystems of automation systems can be connected to the application in a communicative manner and the application can therefore run centrally or can process data and / or data sets from a plurality of automation systems or subsystems. For example, in Figure 3 As shown in FIG. 1 , the first (sub) automation system can include a first controller, a first management system and / or a first operation management system. The second (sub) automation system can include a second controller, a second management system and / or a second operation management system. The two automation systems can communicate with each other for the purpose of data transmission of one or more of their components according to Figure 3 The embodiment of the invention is connected to a computing unit arranged in the cloud infrastructure. The computing unit can deposit or store the data received by the first and / or second automation system in a database, which can also be part of the cloud infrastructure, for example.
[0047] In addition, an application running in a computing unit or on another device can be configured by means of a display unit and / or an operating unit for evaluating one or more data sets. For example, data can be identified in the form of one or more data sets and / or one or more events in a data set can be identified as abnormal events.
[0048] Subsequently, the evaluated event or events can be displayed on a display, for example so as to be made known to a user.
[0049] Therefore, for a more efficient interpretation of anomalies, a method for identifying recurring associations is proposed, which runs on a computing unit locally in the facility or in a cloud infrastructure, evaluates data from controllers, management systems and operational management systems and, if possible, makes the results available to a user, such as a data analyst, on a display unit.
[0050] Figure 4 An embodiment of the proposed method is shown. First, in step S0, the first and / or second data set is detected or, for example, selected by a user. As described above, this can be performed automatically, for example, by continuously determining one or more anomalies of the data (set) in time, for example, by directly using the data set or by manual input by the user. A data set is understood to be a combination of multiple data. For example, the first data set and the second data set can be detected by loading the data set into an application (i.e., a computer program). Here, the first data set and the second data set can be detected and / or loaded separately. For example, the first data set can be detected and / or loaded first and then the second data set.
[0051] Subsequently, in step S1, provision can be made for performing a Fourier transformation of the data comprising the first data set. Particularly preferably, the data corresponding to the abnormal event are subjected to a Fourier transformation. Subsequently, in step S2, a Fourier transformation can be performed on the data of the second data set.
[0052] The first or second data set can represent, for example, individual variables of one or more components of the automation system. These data sets can be subjected to a Fourier transformation, for example a fast Fourier transformation.
[0053] In step S3, the time interval between two abnormal events in the first data set can be determined. The abnormal events can be one or more abnormal events of the same abnormality or abnormal events of different abnormalities. For example, the abnormality is restored in the Fourier transform by a peak, i.e., multiple occurrences of a certain type or multiple types of events. These events are called abnormal events. Multiple such peaks can appear in the Fourier transform. One or more peaks can correspond to different abnormalities.
[0054] Thus, the abnormal abnormal events are mutually spaced apart in the original space, i.e. there is a time interval, for example a cycle, between two (e.g. repeated) abnormal events. In the image space, the abnormal events form one or more peaks. Thus, the abnormal abnormal events (in the original space) are all equally spaced apart from each other. In the case of considering different abnormal abnormal events, the first abnormal abnormal event is therefore also spaced apart from the second abnormal abnormal event, i.e. there is also here a time interval (in some cases different from the first time interval) between the two abnormal events. If this time interval is determined, then it is possible to find associations, i.e. correlations and causal relationships of one or more abnormal events.
[0055] In step S4, events that are repeated and spaced apart according to time intervals in the second data set can be determined. Here, one or more events in the second data set are preferably determined according to the Fourier transform of the second data set. These events can be determined by determining events spaced apart according to time intervals (determined by means of one or more abnormal events in the first data set). Therefore, the determined periodic abnormal events in the first data set can be matched with events with the same periodicity in the second data set. The events in the second data set can be data that are repeated and spaced apart according to time intervals (determined, for example, by means of the first data set). Therefore, the data in the data set (for example, the second data set) corresponds to the event, or the event can be determined by means of the data. For example, the association can be identified by means of the mutual relationship of the Fourier transform implemented. Therefore, a cause study can be performed during anomaly detection, and automatic cause characterization can be achieved. Here, the periodicity can also be determined according to the extreme value observation of the data or events in the first and / or second data set. In addition, one or more time intervals can be determined according to the modeling of approximations, for example, by one or more periodic functions (for example, sine and / or cosine) of the data in the first or second data set and according to the parameters of the function considering the period, and one or more events that cause one or more anomalies can be determined according to the one or more time intervals.
[0056] Abnormal events determined in the first data set and / or events determined in the second data set (spaced apart from one another according to a time interval) can be identified in step S4 and subsequently made available to the user. For this purpose, these events can be stored and / or displayed.
[0057] Figure 5 Further steps are shown, which can be implemented according to the embodiment and according to Figure 4 The steps of the embodiments may be combined or performed independently of each other.
[0058] In step S5, the combination can be performed Figure 4In step S6, an autocorrelation can be performed on the basis of the first data set. In this case, preferably, a Fourier transformation of the first data set is applied. Preferably, for this purpose, the first data set includes data from one or more components of the automation system and detected over different time periods. As a result, correlations and / or causal relationships of events, in particular in the form of periodically recurring events in the first data set, can be determined as described in conjunction with the previous figures.
[0059] Figure 6 In step S7, a first data set (eg having a period T) can be determined. k ). The period can be a time interval described, which represents the (temporally) spaced-apart abnormal events in the first data set. In step S8, it can be determined that the second data set (e.g., with a period T j ) periodicity. Therefore, in this embodiment, events in the first and second data sets are determined, which in some cases do not have the same periodicity, but have different periodicities. In step S9, a cross-correlation based on the first data set and the second data set can be performed. Preferably, the cross-correlation can also be applied to the Fourier transform of the first data set and the Fourier transform of the second data set. Here, a cross-correlation can be performed between an abnormal event with a period Tk in the first data set and a repeated event of the second data set that is spaced apart according to a time interval or periodicity Tj. Therefore, the cause can be determined when there is an abnormality or one or more cyclically occurring abnormal events. For example, single and multi-dimensional variables and / or mutual correlations for abnormalities can be identified.
[0060] Although the identification of repetitions in data has typically been applied to all time-related data fi(t) (KPI, process data, text, process guidance, events, MES, logistics, weather ...), according to the current disclosure, the one-sided Fourier transform is applied only as a second step after identifying the relevant data with anomalies, that is, the data set including abnormal events, for identifying candidates for periodic repetition of anomalies. When directly applying time-related signals, Fourier transform can be used according to the disclosure for the identification of periodic anomalies. In the time range f(t), periodic repetition of anomalies (such as so-called Montagautos) can remain undetected due to background noise, but these anomalies are clearly identified in the spectrum range F(s) by the increase of amplitude (such as in weekly frequencies). In the case of the display of each two spectrum j and k overlapping, the overall periodicity can be identified according to the increased amplitude at the same frequency s. The subsequent mutual relationship of the signal can be identified according to the increased amplitude of the correlation function, which signals appear again in the same cycle. This additional information of the repetition of the cycle of the signal with frequency s helps to identify the candidate as anomaly. Autocorrelation (k=j) can be used to identify one-dimensional periodicity (correlation length) Tk. In the subsequent cross-correlation (k≠j), the data is restricted to the correlation length in order to reduce computational complexity and to hit quickly.
[0061] Figure 7 An embodiment including further steps is shown. In step S10, abnormal events, for example, exceeding a predetermined threshold value can be determined. For this purpose, for example, data from the first data set, the second data set and / or the further data set can be applied. In step S11, one and / or more abnormalities in the first data set can be determined. For this purpose, for example, the determined abnormalities can be selected by the user. Additionally or alternatively, for example, an abnormality with a determined minimum number of abnormal events can be determined. Here, the threshold value can be predetermined by the user. In addition, the abnormal event can also be predetermined, i.e., limited, by the user. In addition, the abnormal event can be a data sequence with an increasing value in a template, for example, a data set.
[0062] In step S12, a (Fourier) transformation of the abnormal events and / or a cross-correlation of the (Fourier) transformed data sets can be implemented. It is already possible here to determine the correlation or causal relationship between the abnormalities that occur. In step S13, the time interval between two abnormal events of one or more abnormalities can be determined. In step S14, the change of the time interval between two abnormal events of one or more abnormalities can be determined. In step S15, the periodic change of the time interval between two abnormal events of one or more abnormalities can be compensated. Thus, for example, once the correlation or causal relationship is identified, they can be tracked, and in particular the change of the time interval between the differences, for example by interruptions (for example one or more pauses), the change of the time interval between the abnormalities caused by the operation of the automation system can be taken into account. In particular, the relationship of the automation system with repetitions can be identified, for example in the case of repeated process influences, in the case of irregular pause times or transmission processes between workstations.
[0063] The periodicity can be determined not only over time but also relative to any other variable fj, wherein the Fourier transform is calculated step by step via the available data. In this case, the frequency analysis of uneven sampling is taken into account when applying the Fourier transform, since the variables fi are usually not present at uniform intervals. This analysis identifies repeated changes of abnormal indicators relative to other variables, so that, for example, cyclic deviations can be identified under different operating conditions (for example, multiple power or speed). Different operating conditions can be balanced by preprocessing, wherein, for example, the variable flj(t)=Bl(t)fj(t) is used only when the first desired operating condition Bl(t)=1, t=1 and Bl(t)=0, t≠1. In addition, non-existent values for spectral analysis but not for correlation analysis can be supplemented, for example, by means of so-called zero filling. In addition, interpolation can be achieved by upsampling with an interpolation function (polynomial, spline curve, etc.) or resampling (filtering and decimation) as downsampling of "fast" variables. If no values are present at certain locations of a plurality of sensors, the values at these locations can be removed from all signals in order to prevent false correlations by setting the values to 0. If the functional correlation of the abnormality indicator to the operating condition is known, this functional correlation is encoded in a wavelet and a wavelet transform is performed instead of a Fourier transform.
[0064] If one or more anomalies are identified, their periodicity can be determined simply by applying a Fourier transform to the anomaly indicator a(t). Cyclic changes in the anomaly indicator can show indications on regularly repeated learning and not yet mature anomaly indicators. In particular, indicators of periodic changes occur in cyclically changing operating conditions that have not yet been taken into account in the anomaly indicator (e.g. shift operation, annual production fluctuations, repeated process steps, employees, batches, production steps). The periodicity of the frequency ω found by the Fourier transform can be compensated from the anomaly indicator using a'(t)=a(t) / (aω·sin(ωt)).
[0065] Figure 8 Another embodiment is shown, which can be used with Figures 4 to 7 On the other hand, it is possible to independently Figures 4 to 7 Steps S16 to S18 are implemented according to the embodiment described. Thus, in step S16, a first, preferably repetitive, operating state (Bk(t)) of the automation system can be determined. In step S17, a first data set belonging to the first operating state can be determined, which first data set includes data about at least one first component of the automation system. In step S18, a second data set belonging to the first operating state can be determined, which second data set includes data about at least one second component of the automation system.
[0066] Fig. 9 Another embodiment comprising steps S19 to S24 is shown. In step S19, a first transformation based on the first data set can be performed. In step S20, a first abnormal event can be determined by determining the number of abnormal events for the first transformation. The first abnormal indicator can correspond to the above-mentioned abnormal score, for example. In step S21, a second transformation based on the second data set can be performed. In step S22, different sampling rates of the first data set and the second data set can be compensated. Here, step S22 of compensating for different sampling rates can be included in one of steps S21 and / or S19.
[0067] In step S23, a second anomaly indicator can be determined, for example, by determining the number of abnormal events for the second transformation. Here, the second anomaly indicator can correspond, for example, to the above-mentioned anomaly score. Preferably, an anomaly indicator of the same type as the first anomaly indicator is used for the second anomaly indicator. In step S24, the first anomaly indicator can be compared with the second anomaly indicator to determine one or more anomalies during the operation of the automation system.
[0068] When a suitable transformation (Hilbert, Wiegner, Weiler, etc.) is selected, the relationship between any repeated functions is detected. In contrast to the mutual relationship, scaling and expansion of the transformation of the basis functions for determining the relationship between the eigenvalues is now also considered.
[0069] The data set j which has the most anomalies after the transformation is examined more closely as a candidate for anomaly compared to the other data sets k, in particular after balancing one or more operating conditions. It can be determined via a criterion, such as a predetermined threshold value, whether this number of anomalies is significant with respect to the random occurrence of anomalies. In order to reduce the necessary computing power, the transformation is limited to the two data sets in question. For industrial applications, a sequence is selected in which the anomaly detection on a part of the process data is supplemented step by step with event-, location-, function- (electrical, mechanical, process technology, logic, sound, hydraulic, pneumatic, thermal)-related data in order to gradually take into account different time scales and the periodicity of the system under consideration.
[0070] Fig.10 Another embodiment comprising steps S25 to S27 is shown. In step S25, a third data set belonging to the first operating state can be determined. In step S26, a first transformation based on the first data set can be performed. In step S27, the third abnormal indicator can be compared with the first abnormal indicator and / or with the second abnormal indicator. Therefore, it can be determined which variables have changed in the same way. In addition, an abnormality or one or more abnormal events can be associated with multiple or all variables. This can be achieved, for example, by calculating a correlation function of the current Fourier transform and the Fourier transform used as a fingerprint. As mentioned above, a fast Fourier transform can be used for this.
[0071] Fig.11 Another embodiment comprising steps S28 to S30 is shown. In step S28, a plurality of data sets can be determined, each of which includes data about at least one component of the automation system. In step S29, at least one first group and a second group can be formed, each of which includes one or more data sets. In step S30, an m-dimensional linear transformation of the first group and the second group can be determined. Fig.12Another embodiment including steps S31 to S33 is shown. In step S31, the process state can be determined based on the envelope of the m-dimensional transformation (for example, the transformation obtained from step S30). In step S32, multiple data sets can be determined, each of which includes data about at least one component of the automation system. The envelope is used to group cycle-related data or events and filter out non-cyclic data derived only from static data or events. For this purpose, for example, the time below the peak of the envelope can be grouped as a process state or taken into account for further observation. In order to select the corresponding peak value, a threshold value can be determined. For example, the threshold value can be formed based on the average value or maximum value of one or more peak values of the envelope and the average value of one or more secondary peak values of the envelope.
[0072] Even after applying a template recognition method (e.g., Kalman filter, correlation analysis, SVM, neural network, decision tree analysis, K-NN, self-organizing map, hierarchical matrix model, association network, ...) according to the present invention, the recognition of the period is applied as follows. All the methods mentioned are the same, namely determining the corresponding function zm (f1, ... fn) to the mth cluster as a function of the fn data. The internal periodic structure of the cluster is found by applying a multidimensional Fourier transform on the corresponding function. Uneven sampling can be considered by using a frequency analysis of uneven sampling. The inverse of the frequency of the high amplitude of the Fourier transform indicates the spacing between two potential subclusters. The spectrum latitude involved indicates another separation data dimension suitable for the cluster. A more accurate characterization of the transformation is achieved based on the shape of the amplitude that appears. The peak amplitude indicates a clear period, while the flat amplitude indicates a periodic fluctuation, which can be explained, for example, by fluctuations in shift time and production time. Periodic peaks, such as sidebands, indicate relevant periodicity, that is, for example, abnormalities occur every Monday or on the first Monday of every month. This information is helpful in the identification of causes, when, for example, typical multiple transformations occur, which can be identified by the correlation of the transformations. The autocorrelation (k=j) allows the identification of the periodicity (correlation length) Tj of one dimension j. Here, pauses, waiting times, etc. can be taken into account by the above-mentioned preprocessing and used in turn for the identification of the periodicity of the operating conditions and the selection of the iteration of B1 for the preprocessing. The cross-correlation with other variables (k≠j) is carried out after resampling (up-sampling / down-sampling), and it is also particularly promising to fill in the correlation with this correlation length Tj in the correlation of other dimensions, so that computing time can be saved. The evaluation of the phase gives when the periodic anomaly begins relative to the start of the recording. The earliest periodic anomaly at the beginning indicates periodic learning, while the anomaly that starts later is more attractive because it involves current changes. Because it is expected that the anomaly is accompanied by a change in the correlation, that is, the correlation changes over time, the trend display of the correlation that changes particularly greatly over time is particularly advantageous. In particular, it is necessary to check whether the correlation that rises significantly over time is related to the anomaly.
[0073] Here, a plurality of or all of the steps described can be exchanged in their order and / or automatically performed in sequence. In addition, it is possible to combine a plurality of steps into one step.
[0074] For the above steps, a systematic method for characterizing anomalies and finding possible operating or boundary conditions that may be the cause of the anomalies is advantageous. In addition, even in complex automation systems, connections can be identified using temporally extended and stretched operating sequences as well as interrupted operating sequences. In addition, the specific (process) cycle times and operating conditions of the respective automation system can be taken into account.
[0075] In addition, a computer-readable storage medium is proposed. A computer-readable storage medium can be a tangible device that can store instructions (for use by an instruction execution device). A computer-readable storage medium can be, for example, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or some suitable combination of the foregoing, but this is not limited to this. A non-exhaustive list of specific examples of computer-readable storage media includes: a portable computer floppy disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static access memory (SRAM), a portable CD-ROM memory (CD-ROM), a DVD (Digital Versatile Disk, Digital Versatile Disk), a memory stick, a floppy disk, a machine-readable perforated paper, or a protruding structure of a groove for recording instructions and each suitable combination of the foregoing. A computer-readable storage medium should not be understood as a transient signal, such as a radio wave or other electromagnetic wave that propagates freely through a wave conductor or other transmission medium (such as a light pulse through an optical fiber cable) or an electrical signal transmitted through a wire.
[0076] The computer-readable program instructions described herein can be transmitted from a computer-readable storage medium or an external computer or an external storage device via a network (e.g., the Internet, a local area network, a wide area network, and / or a wireless network). The network can include copper transmission cables, optical transmission fibers, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface of each computer / processing device receives the computer-readable program instructions for further storage in a computer-readable storage medium in the corresponding computer processing device.
[0077] Here, various aspects of the present invention are described with reference to the flowchart and / or block diagram of the method, device (system) and computer program product according to the embodiment of the present invention. It is understood that each frame of the flowchart and / or block diagram and the combination of the frames in the flowchart and / or block diagram can be executed by computer-readable program instructions. These computer-readable program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to produce a machine, thereby executing instructions via the processor of the computer or other programmable data processing system. Other devices provide components for executing the functions / actions specified in the flowchart and / or block diagram frame or frame. These computer-readable program instructions can also be stored in a computer-readable storage medium, such as a computer, a programmable data processing device and / or other devices. In the computer-readable storage medium that stores instructions or can include these instructions, these aspects correspond to the functions / actions described in detail in the flowchart by means of block diagrams and frames. The flowcharts and block diagrams in the accompanying drawings vividly illustrate the possible execution structures, functions and operations of systems, methods and computer program products according to various embodiments of the present invention. In this respect, each frame in the flow chart or block diagram represents a part of a module, a section or an instruction, and the instruction includes one or more executable instructions for performing a dedicated logical function. In some alternative executions, the function given in the frame can not appear in the order given in the figure. For example, the two frames shown in sequence can actually be implemented substantially at the same time, or these frames can sometimes be implemented in reverse order, depending on the function involved. It should also be noted that each frame of the block diagram and / or flow chart and the combination of the frames in the block diagram and / or flow chart can be performed by a dedicated hardware-based system, which implements a dedicated function or process. The example of a hardware-based system is an optical lens or a filter bank as a hardware implementation for Fourier transform, which can also be implemented as a micro-electromechanical system machine, and the computer instruction performs correlation and limit value monitoring.
[0078] Also described consistent with other embodiments is a computer program product for detecting anomalies in data, the data being provided, for example, by each of a plurality of data sources. The computer program product can include one or more computer-readable storage media having computer-implementable instructions executed thereon. The computer-readable storage medium includes instructions for receiving a corresponding time series data sequence for each of the plurality of data sources, wherein each data sequence represents a plurality of data values, the data values being detected by a corresponding one of the plurality of data sources at a certain sampling frequency, and each of the data values of each data sequence being sampled in a time series data order at a determined point in time.
Claims
1. A method for evaluating at least one data set of at least one component of an automation system (4), characterized in that Includes and / or causes the following steps: determining a time interval between two abnormal events in a first data set, the first data set comprising data about at least one first component (2a, 2b, 2c, 2d, 2e) of the automation system, and Determining events in a second data set that are repeated and spaced apart according to the time interval enables determining periodic abnormal events in the first data set to be associated with events with the same periodicity in the second data set, the second data set comprising data about at least one second component (2a, 2b, 2c, 2d, 2e) of the automation system (4).
2. The method according to the preceding claim, characterized in that Includes and / or causes the following steps: performing a Fourier transform of data comprising or based on the first data set, and A Fourier transform of data including or based on the second data set is performed.
3. The method according to claim 2, characterized in that Includes and / or causes the following steps: An autocorrelation based on the first data set is performed.
4. The method according to claim 3, characterized in that Includes and / or causes the following steps: determining periodic abnormal events in the first data set, and / or Periodic events in the second data set are determined.
5. The method according to claim 4, characterized in that Includes and / or causes the following steps: A cross-correlation based on the first data set and the second data set is performed.
6. The method according to claim 5 for evaluating data of an automation system, characterized in that Includes and / or causes the following steps: One and / or more anomalies in the first data set are determined by determining an abnormal event, and a time interval between two abnormal events of one or more anomalies is determined.
7. The method according to claim 6, characterized in that Includes and / or causes the following steps: Compensating for a change in the periodicity of the time interval between two abnormal events of one or more abnormalities, wherein, in the event of a change in the periodicity of the time interval between two abnormal events, a periodicity is applied for balancing the change in the time interval between two abnormal events.
8. The method according to claim 7 for evaluating data of an automation system, characterized in that Includes and / or causes the following steps: determining a first operating state of the automation system, and determining a first data set belonging to the first operating state, the first data set comprising data about at least one of the first components of the automation system, A second data set is determined, which belongs to the first operating state and includes data about at least one second component of the automation system.
9. The method according to claim 8 for evaluating data, characterized in that Includes and / or causes the following steps: performing a first transformation based on the first data set, and determining a first anomaly indicator for the first transformation by determining a number of the anomaly events, and performing a second transformation based on the second data set, determining a second anomaly indicator for the second transformation by determining a number of the anomaly events, and The first anomaly indicator and the second anomaly indicator are compared to determine one or more anomalies in operation of the automation system.
10. The method according to claim 9 for evaluating data, characterized in that Includes and / or causes the following steps: The different sampling rates of the first and second data sets are compensated for by compensating for the sampling rate of the application of a Fourier transform.
11. The method according to claim 10 for evaluating data, characterized in that Includes and / or causes the following steps: determining a third data set belonging to the first operating state, the third data set comprising data about at least one third component of the automation system during the first operating state, performing a third transformation based on the third data set, determining a third anomaly indicator for the third transformation by determining the number of the anomaly events, and / or The third exception indicator is compared with the first exception indicator and / or the second exception indicator.
12. A device for carrying out the method according to any one of the preceding claims, wherein: The device is configured to perform one or more steps of the method according to any one of claims 1 to 11.
13. A computer-readable storage medium having a computer program with program code means for implementing one or more steps of the method according to any one of claims 1 to 11 when the program code means are executed.
