Computer-implemented method and system for identifying anomalies

Abstract

A computer-implemented method for identifying anomalies when controlling a construction device (TD) by means of a control mechanism (PLC), comprising the following steps: a) providing input data (IN), encoding the input data (IN) into encoded data (C) by means of an automatic encoder, and decoding the encoded data (C) into output data (OUT) by means of the automatic encoder, and determining and classifying reconstruction errors (RE), the reconstruction error (RE) exceeds a predetermined reconstruction error limit (RET), b) determining a target variable (TV) from the reconstruction error (RE) if the reconstruction error (RE) exceeds a predetermined reconstruction error limit (RET), c) providing an artificial intelligence-based model to the evaluation device (XPL), d) determining a prediction of the target variable (TV) by the evaluation device (XPL) on the basis of a classification of the model and the reconstruction error (RE) and providing the prediction to the analysis device (REA), e) determining a rule for improving the reconstruction error (RE) from the predicted and predetermined rule set by the evaluation device (REA), providing the rule to the control device (PLC), and actuating the engineering device (TD) by the control device (PLC) by means of the rule.

Description

Technical Field

[0001] The invention relates to a computer-implemented method and system for identifying anomalies when controlling an engineering system by means of a control mechanism.

[0002] The invention further relates to a computer program, an electronically readable data carrier and a data carrier signal. Background Art

[0003] In factories, systems based on artificial intelligence (“AI” for short) are often used, for example, in order to identify anomalies in manufacturing steps and to take corresponding measures in order to improve production quality and increase efficiency by keeping production downtimes low.

[0004] With the commonly used concept of "black box" AI models, it is difficult or even impossible to explain why the AI ​​made a specific decision. This reduces trust in AI-based systems and may prevent humans from further applying AI and taking advantage of its benefits.

[0005] This is especially important for systems that collect data from programmable controllers and perform recognition of anomalies. Here, a suitable artificial intelligence model is an autoencoder that learns the normal behavior or good state of the system and recognizes deviations from normal behavior during runtime. One indicator of the deviation is the so-called reconstruction error. If the indicator is high, it can be considered that there is an anomaly. The scheme is simple but effective. However, the user does not know: which collaboration of the input variables or which association of the features leads to high reconstruction errors. If the input variables are, for example, current, voltage, vibration, temperature, torque, drive speed or pressure, it is usually unclear: which state of the variable affects the high deviation about the reconstruction error. The reconstruction error is the sum of the deviations between the input variables and the reconstructed output variables. Therefore, it can be concluded that which input variables, such as torque or drive speed, are decisive for high reconstruction errors, but which states of the input variables are not decisive, such as high vibrations at low temperatures and low drive speeds.

[0006] In the prior art, simpler models are used in parallel with deep learning models such as decision trees (random forests, or RF), which support excellent interpretability but weak performance.

[0007] On the other hand, the concept of so-called "explainable artificial intelligence", or "XA" for short, is known. In this context, most methods state more generally which specific areas of the input data are taken into account during the model classification process. Summary of the invention

[0008] The object of the present invention is therefore to provide a method for applying artificial intelligence which, for example, provides a machine user with improved information about why errors occurred in a production step and which can be compensated accordingly, in order to reduce or eliminate the errors in further production.

[0009] In other words, a method should be created which better diagnoses the cause of a detected error in a system for detecting anomalies based on an autoencoder and preferably eliminates the cause automatically and accurately.

[0010] The object according to the invention is achieved by a method of the type mentioned at the outset by carrying out the following steps:

[0011] a) providing input data at a data input via a control device, the input data comprising sensor data, the sensor data being detected by a sensor device at the engineering device, encoding the input data into encoded data by means of an autoencoder, decoding the encoded data into output data by means of the autoencoder, and determining and classifying reconstruction errors,

[0012] b) if the reconstruction error exceeds a predetermined reconstruction error limit, determining a target variable from the reconstruction error,

[0013] c) providing an evaluation device with an artificial intelligence based model generated and trained with the aid of historical data of reconstruction errors and corresponding input features from an autoencoder,

[0014] d) determining, by means of an evaluation device, at least one prediction for the target variable and optionally for other variables related to the target variable based on the model and the classification of the reconstruction error, and providing the at least one prediction to an analysis device,

[0015] e) Determining, by means of the evaluation device, a rule for improving the reconstruction error from at least one predicted and predetermined rule set, providing the rule to the control device, and controlling the technical system by means of the control device.

[0016] This makes it possible to analyze reconstruction errors of the autoencoder in a simple manner and to identify the exact cause of detected errors.

[0017] The object according to the invention is achieved by a system of the type mentioned at the outset, which comprises a sensor arrangement for detecting sensor data relating to the technical installation and a memory for storing historical data and is designed to carry out the method according to the invention.

[0018] In a development of the invention, it is provided that the system is arranged at the edge of a client-server system.

[0019] Advantageously, efficient methods can be implemented at the edge of a distributed system, because availability is higher and communication costs can be reduced. However, edge computing devices usually do not have large computing and storage capacities, which is why only particularly efficient methods for identifying anomalies, such as the method according to the invention, can be used.

[0020] In a development of the invention, it is provided that the sensor device is a camera and the sensor data are camera images.

[0021] The method according to the invention is particularly effective when analyzing images in conjunction with an autoencoder.

[0022] The object according to the present invention is also achieved by a computer program, which comprises instructions, which, when executed by a computer, cause the computer to execute the method according to the present invention.

[0023] The object according to the invention is also achieved by an electronically readable data carrier having readable control information stored thereon, the control information comprising at least a computer program according to the invention and being designed such that the control information executes the method according to the invention when the data carrier is used in a computing device.

[0024] The object according to the invention is also achieved by a data carrier signal which transmits the computer program according to the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The present invention will be explained in more detail below based on the embodiments shown in the attached drawings. In the drawings:

[0026] Figure 1 An embodiment of the method according to the invention is shown in a general and simplified form,

[0027] Figure 2 An embodiment of the method according to the invention is shown in detail,

[0028] Figure 3 An embodiment of a flow chart of a method according to the invention is shown. DETAILED DESCRIPTION

[0029] Figure 1 An embodiment of the method according to the invention is shown in a general and simplified form.

[0030] The method is computer-implemented, ie, one or more steps can be performed on a computer.

[0031] In a method for detecting anomalies when controlling a technical device TD by a control unit PLC, it is proposed that input data IN are provided by the control unit PLC at a data input of an autoencoder.

[0032] The input data IN include sensor data detected by a sensor mechanism SM at the process device TD, for example image data from a camera which optically detects the production of the product and is used for optically checking the produced product.

[0033] The sensor device SM is preferably connected to the control device PLC and detects sensor data, which are provided to the autoencoder via the control device PLC.

[0034] The autoencoder encodes input data IN into coded data C by means of a coding device COD.

[0035] Furthermore, the autoencoder decodes the encoded data C into output data OUT by means of a decoding device DEC and also determines a reconstruction error RE.

[0036] A predetermined error limit value is assigned to the reconstruction error RE, within which the sensor value is permissible or impermissible.

[0037] The reconstruction error RE is now used to identify anomalies in the input data IN or the output data OUT.

[0038] The technical device TD is controlled by means of a control unit PLC.

[0039] The evaluation device XLP is now used to determine a prediction of the reconstruction error RE based on an artificial intelligence-based model.

[0040] This prediction is transmitted to an evaluation device REA which determines rules from a predetermined set of rules for improving the reconstruction error RE. Subsequent production should be improved as a result, since errors in production have been identified and causes derived therefrom in the form of predictions, and corresponding measures for improving product properties can be taken by means of the control device PLC by means of the rules to control the engineering device TD.

[0041] Figure 2 The method according to the present invention is shown in detail. Figure 1 An embodiment of the invention, wherein the following steps are provided:

[0042] a) providing input data IN at a data input by a control device PLC, said input data comprising sensor data which are detected by a sensor device SM at a technical device TD, encoding the input data IN into encoded data C by means of an autoencoder, and decoding the encoded data C into output data OUT by means of an autoencoder, and determining and classifying a reconstruction error RE,

[0043] b) if the reconstruction error RE exceeds a predetermined reconstruction error limit RET, determining a target variable TV from the reconstruction error RE,

[0044] c) providing the evaluation device XPL with a model based on artificial intelligence, which model is generated and trained with the help of the historical data DH of the reconstruction errors and the corresponding input features from the autoencoder,

[0045] d) determining, by means of the evaluation device XPL, at least one prediction for the target variable TV and optionally for other variables related to the target variable TV from the model and providing the at least one prediction to the analysis device REA,

[0046] e) determining, by means of the evaluation device REA, a rule for improving the reconstruction error RE from a predetermined set of rules, providing the rule to the control device PLC, and actuating the technical device TD by means of the control device PLC,

[0047] The evaluation device XPL (English “interpreter”) is generated and trained with the help of the collected historical input data, which are generated in the autoencoder by its model, in order to determine the reconstruction error RE. This can be implemented by means of random forest regression trees or classification trees.

[0048] The regression tree can learn: based on the input variable prediction reconstruction error RE as the target variable (English “target variable”).

[0049] Classification trees can predict categories (“labels” in English) based on input variables.

[0050] The evaluation device XPL can form the categories by automatically dividing the reconstruction errors into groups ("clusters" in English) during a specific time period. The groups can be used to derive the categories, and the input variables and their groups can be used to train a classification tree.

[0051] Therefore, the evaluation device XPL provides information about the importance of variables and the interactions between variables.

[0052] The information obtained from the evaluation device XPL is now used to define rules that should lead to an improvement in the reconstruction error RE. For example, such rules can be a semi-automatic or fully automatic adaptation of the operating variables of the technical device TD, or also the performance of a calibration or sensor cleaning.

[0053] When selecting rules, the analysis device REA can preferably take into account statistical variable correlations R, with the help of which the predictions can be weighted among the various parameters and features, for example with the help of variable importance (English "variable importance"), a technique for determining which features within a dedicated input vector make the greatest contribution to the output variables of the model.

[0054] The analysis device REA is preferably rule-based, but can optionally also use artificial intelligence models.

[0055] The "variable importance" is found with the help of the XPL solution discussed previously. The rule set is defined by the rule {English: "IF'condition'THEN'result'"}.

[0056] For each rule in the rule set, the relevance R is calculated by determining which input variables are identified by means of variable importance and which variables are processed in the rule set.

[0057] An example of a rule is

[0058] {IF 'vibration' == high AND 'temperature' == low AND 'speed' == low THEN 'action: increase speed'}

[0059] It is proposed that the prediction includes one or more parameters associated with the target variable TV.

[0060] The figure also shows an embodiment with a block diagram of a system S according to the invention.

[0061] The system S has engineering equipment TD.

[0062] Furthermore, a control unit PLC is included which provides sensor data as input data IN.

[0063] Here, the control device PLC is connected to a sensor device SM, which detects sensor data - related to the engineering device TD - such as image data of a camera, which optically detects the production of products and is used for optical inspection of the produced products, and provides the sensor data to the automatic encoder.

[0064] The autoencoder is implemented in a computing device EDGE having a memory and a processor, and may also comprise an evaluation device XPL and an analysis device REA.

[0065] Furthermore, history data DH formed from input data IN analyzed by the autoencoder as well as the model and reconstruction errors RE can be processed and stored in the computing device EDGE. The history data DH provide the basis for the model of the evaluation device XPL in step c) of the method.

[0066] Figure 3 An embodiment of a flow chart of the method according to the invention is shown.

[0067] List of reference numerals:

[0068] C coded data

[0069] DEC decoder equipment (English "decoder")

[0070] DH historical data

[0071] Computing devices at the edge of an EDGE client / server system

[0072] ENC encoding equipment (English "encoder")

[0073] IN Input data

[0074] OUT Output data

[0075] PLC Programmable Logic Controller

[0076] R Correlation

[0077] RE reconstruction error

[0078] RET reconstruction error threshold

[0079] REA analysis equipment (English "reasoning engine")

[0080] S system

[0081] SM Sensor mechanism, such as camera

[0082] t time

[0083] TD Technical Device

[0084] TV target variable

[0085] XPL evaluation equipment (English "explainer interpreter")

Claims

1. A computer-implemented method for identifying anomalies when controlling a technical device (TD) by a control mechanism (PLC), comprising the following steps: a) providing input data (IN) at a data input by the control device (PLC), the input data comprising sensor data, the sensor data being detected by a sensor device (SM) at the engineering device (TD), encoding the input data (IN) into coded data (C) by means of an autoencoder, and decoding the coded data (C) into output data (OUT) by means of the autoencoder, and determining and classifying a reconstruction error (RE), b) determining a target variable (TV) from said reconstruction error (RE) if said reconstruction error (RE) exceeds a predetermined reconstruction error limit (RET), c) providing an evaluation device (XPL) with an artificial intelligence based model, said model being generated and trained with the aid of historical data (DH) of reconstruction errors and corresponding input features from said autoencoder, d) determining, by means of the evaluation device (XPL) on the basis of the model and the classification of the reconstruction error (RE), at least one prediction for the target variable (TV) and optionally for other variables related to the target variable (TV), and providing the at least one prediction to an analysis device (REA), e) determining, by the analysis device (REA), a rule for improving the reconstruction error (RE) from the at least one predicted and predetermined rule set, providing the rule to the control device (PLC), and actuating the engineering device (TD) by the control device (PLC) with the aid of the rule.

2. A system (S) for identifying anomalies when controlling engineering equipment (TD) by a control mechanism (PLC), the system comprising a sensor mechanism (SM) for detecting sensor data related to the engineering equipment (TD) and a memory for storing historical data (DH), and the system (S) is designed to: perform a method according to any one of the preceding claims.

3. System according to the preceding claim, wherein the system is arranged at the edge of a client-server system.

4. System according to any one of claims 2 or 3, wherein the sensor means (SM) is a camera and the sensor data is a camera image.

5. A computer program comprising instructions, which, when executed by a computer, cause the computer to perform the method according to claim 1.

6. Electronically readable data carrier having stored thereon readable control information which at least comprises a computer program according to the preceding claim and is designed such that the control information executes the method according to claim 1 when the data carrier is used in a computing device.

7. Data carrier signal transmitting a computer program according to claim 1.

Images