COMPUTER-IMPLEMENTED METHOD AND SYSTEM FOR ANOMALY DETECTION

Description

[0001] The invention relates to a computer-implemented method and a system for anomaly detection in the control of a technical device by a control means.

[0002] Furthermore, the invention relates to a computer program, an electronically readable data carrier and a data carrier signal.

[0003] In factories, systems based on artificial intelligence (AI for short) are often used to detect anomalies in manufacturing steps and to initiate appropriate countermeasures in order to improve the quality of production and increase efficiency by minimizing downtime.

[0004] The commonly used concept of "black box" AI models makes it difficult or even impossible to explain why an AI arrived at a particular decision. This reduces trust in AI-based systems and can discourage people from further using AI and benefiting from its advantages.

[0005] This is particularly important for systems that collect data from programmable logic controllers (PLCs) and perform anomaly detection. A suitable AI model for this is an auto-encoder, which learns the normal behavior or good state of a system and detects deviations from this normal behavior during runtime. One indicator of a deviation is the so-called reconstruction error. If this error is high, it can be assumed that an anomaly is present. This approach is simple but effective. However, a user does not know which interplay of input variables or which relationships between the features led to the high reconstruction error. If the input variables are, for example, current, voltage, vibration, temperature, torque, drive speed, or pressure, etc., it is often not apparent which state of these variables influenced the high deviation with respect to the reconstruction error.The reconstruction error is the sum of the deviations between the input variables and the reconstructed output variables. Therefore, it is possible to determine which input variables, such as torque or drive speed, are primarily responsible for a high reconstruction error, but not which states of the input variables were decisive, such as high vibration at low temperature and low drive speed.

[0006] In current technology, simpler models are used in parallel with deep learning models such as decision trees (random forests, or RFs), which offer excellent interpretability but are less powerful.

[0007] On the other hand, the concept of so-called "explainable AI," or "XA" for short, is well-known. Most methods using this approach specify, rather generally, which particular areas of the input data the model considers during classification.

[0008] Publication EP 3 876 060 A1 discloses a method for analyzing the cause of an anomalous state of a machine in a model-based monitoring procedure that detects anomalous states of the machine by determining an anomaly measure, wherein the model models machine states by reconstructing at least two operating parameters of the machine, which serve as input values ​​of the model, as output values, wherein in the model the output values ​​exhibit deviations from the respective input values ​​and the model has been trained by machine learning to minimize the deviations.

[0009] Publication US 2020 / 310370 A1 describes a control system with a memory that stores a sequence of sensor data received from one or more sensors. The control system has a processor that processes the sensor data to calculate a sequence of derived sensor data values. An autoencoder receives the sequence of derived sensor data values ​​and calculates a forward prediction of the sequence. The autoencoder was trained by establishing a relationship between the positions of the derived sensor data values ​​encoded in a latent space of the autoencoder. A processor initiates the control of a device using the forward prediction.

[0010] It is therefore an object of the invention to provide a method for the application of artificial intelligence which, for example, provides a machine user with improved information as to why an error occurred in a production step and can accordingly counteract it in order to reduce or eliminate the error in further production.

[0011] In other words, a procedure should be created that better diagnoses the cause of a detected error in a system based on an auto-encoder for anomaly detection and preferably corrects it automatically and accurately.

[0012] The problem according to the invention is solved by a method of the type mentioned at the outset, in which the following steps are carried out: a) Providing input data via the control device, comprising sensor data acquired by a sensor on the technical device, to a data input; encoding the input data into coded data using an auto-encoder; decoding the coded data into output data using the auto-encoder; and determining and classifying a reconstruction error; b) Determining a target variable from the reconstruction error if the reconstruction error exceeds a predetermined reconstruction error limit; c) Providing an artificial intelligence-based model, generated and trained using historical data of reconstruction errors and corresponding input features from the auto-encoder, to an evaluation device; d) Determining at least one prediction for the target variable and optionally for other variables that depend on the target variable.based on the model and classification of the reconstruction error by the evaluation device, and providing the at least one prediction to an analysis device, e) determining a rule for improving the reconstruction error by the analysis device from the at least one prediction and a predetermined rule set, providing the rule to the control device and controlling the technical device using the rule by the control device.

[0013] This makes it possible to easily analyze the reconstruction error of an auto-encoder and to identify an accurate cause for the detected error.

[0014] The problem according to the invention is solved by a system of the type mentioned at the outset, comprising a sensor means for recording sensor data relating to the technical device, as well as a memory for storing historical data, an evaluation device and an analysis device, and the system is configured to carry out the method according to the invention.

[0015] In a further development of the invention, it is provided that the system is arranged at an edge of a client-server system.

[0016] It is advantageous if an efficient method can be executed at the edge of a distributed system, as availability is higher and communication costs can be reduced. However, an edge computing device usually does not have large computing and storage capacities, which is why only particularly efficient anomaly detection methods, such as the method according to the invention, can be used.

[0017] In a further development of the invention, it is provided that the sensor means is a camera and the sensor data are camera images.

[0018] The method according to the invention is particularly efficient in the analysis of images in combination with an auto-encoder.

[0019] The problem according to the invention is also solved by a computer program according to claim 5.

[0020] The problem according to the invention is also solved by an electronically readable data carrier with readable control information stored on it, which at least includes the computer program according to the invention and is designed in such a way that it carries out the method according to the invention when the data carrier is used in a computing device.

[0021] A data carrier signal transmits the computer program according to the invention.

[0022] The invention is explained in more detail below with reference to an embodiment illustrated in the accompanying drawings. The drawings show: Fig. 1 shows an embodiment of the method according to the invention in a general and simplified form, Fig. 2 shows the embodiment of the method according to the invention in detail, Fig. 3 shows an embodiment of a flowchart for the method according to the invention.

[0023] Fig. 1 Figure 1 shows an embodiment of the method according to the invention in a general and simplified form.

[0024] The process is computer-implemented, meaning that one or more steps can be performed on a computer.

[0025] In the procedure for anomaly detection in the control of a technical device TD by a control instrument PLC, it is provided that input data IN is supplied by the control instrument PLC to a data input of an auto-encoder.

[0026] The input data IN includes sensor data which is captured by a sensor medium SM on the technical device TD, for example image data from a camera which optically records the production of products and is used for optical inspection of manufactured products.

[0027] The sensor device SM is preferably connected to the control device PLC and acquires the sensor data, which is provided to the auto-encoder via the control device PLC.

[0028] The auto-encoder uses a coding device COD to encode the input data IN into encoded data C.

[0029] Furthermore, the auto-encoder decodes the encoded data C to output data OUT using a decoding device DEC and also determines a reconstruction error RE.

[0030] Predefined error limits are assigned to the reconstruction error RE, within which a sensor value is permissible or not.

[0031] The reconstruction error RE is now used to detect an anomaly in the input data IN or output data OUT.

[0032] The technical device TD is controlled using the PLC control system.

[0033] An evaluation device XLP is now used to determine a prediction for a reconstruction error RE based on an artificial intelligence-based model.

[0034] The prediction is transmitted to an analysis device (REA), which determines a rule for improving the reconstruction error (RE) from a predetermined set of rules. This is intended to lead to an improvement in subsequent production, since a production error has been detected and its cause derived in the form of the prediction, and a corresponding measure to improve product characteristics can be initiated using the PLC control unit by controlling the technical device (TD) according to the rule via the PLC control unit.

[0035] Fig. 2 shows the exemplary embodiment of the Fig. 1 for the method according to the invention in detail, wherein the following steps are provided: a) Providing input data IN via the PLC, comprising sensor data acquired by a sensor SM on the technical device TD, to a data input; encoding the input data IN to coded data C using an auto-encoder; decoding the coded data C to output data OUT using the auto-encoder; and determining and classifying a reconstruction error RE; 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, generated and trained using historical data DH of reconstruction errors and corresponding input characteristics from the auto-encoder, to an evaluation device XPL; d) determining at least one prediction for the target variable TV and optionally for other variables.which depend on the target variable TV, based on the model by the evaluation device XPL, and providing at least one prediction to an analysis device REA, e) Determining a rule to improve the reconstruction error RE by the analysis device REA from a predetermined rule set, and providing the rule to the control device PLC, and controlling the technical device TD using the rule by the control device PLC.

[0036] The evaluation device XPL (English for "explainer") is created and trained using collected historical input data, which is generated in the auto-encoder by its model, in order to determine the reconstruction error RE. This can be implemented using random forest regression trees or classification trees.

[0037] Regression trees can learn to predict the reconstruction error RE as a target variable based on input variables.

[0038] Classification trees can predict classes (labels) based on input variables.

[0039] The evaluation tool XPL can create these classes by automatically dividing the reconstruction error into groups (clusters) over a specific period. These groups can be used to derive the classes, and the input variables with their groups can be used to train classification trees.

[0040] The XPL evaluation device provides information regarding variable importance and interaction between variables.

[0041] 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. Such rules can, for example, include a semi- or fully automatic adjustment of operating parameters of the technical device TD, or the performance of a calibration or sensor cleaning.

[0042] The analysis device REA can preferentially consider a statistical measure relevance R when selecting rules, which can be used to weight the prediction in individual parameters or features, for example using variable importance, a technique for determining which features within a dedicated input vector contribute most to an output variable of a model.

[0043] The REA analysis device is preferably rule-based, but can optionally also apply an artificial intelligence model.

[0044] The "variable importance" is determined using the previously discussed XPL approach. A rule set is defined using rules {"IF 'condition' THEN 'result'"}.

[0045] For each rule in the rule set, a relevance R is calculated by determining which input variables have been identified using Variable Importance and which variables are dealt with in the rule set.

[0046] An example of a rule is IF ′ vibration ′ = = HIGH AND ′ temperature ′ = = LOW AND ′ drivespeed ′ = = LOW THEN ′ action : drivespeed − increase ′

[0047] The prediction is intended to include one or more parameters that depend on the target variable TV.

[0048] The figure also shows an embodiment with a block diagram for a system S according to the invention.

[0049] System S includes a technical device TD.

[0050] Furthermore, a PLC control device is included, which provides sensor data as input data IN.

[0051] The PLC control device is connected to a SM sensor device, which captures sensor data relating to the technical device TD, such as image data from a camera that optically records the production of products and is used for optical inspection of manufactured products, and provides this data to the auto-encoder.

[0052] The EDGE computing device, which includes a memory and a processor, incorporates the auto-encoder and can also include the XPL evaluation device and the REA analysis device.

[0053] Furthermore, the historical data DH, which is generated from input data IN already analyzed by the auto-encoder, as well as the model and reconstruction errors RE, can be processed and stored in the computing device EDGE. The historical data DH provides the basis for the model of the evaluation device XPL in step c) of the procedure.

[0054] Fig. 3 shows an exemplary embodiment of a flowchart for the method according to the invention. Reference symbol list:

[0055] C-coded data; DECDecoder; DHh-historical data; EDGERacuity at the edge of a client / server system; ENC-encoder; INinput data; OUToutput data; PLCprogrammable logic controller; R-relevance; RE-reconstruction error; RETReconstruction error threshold; REAreasoning engine; SSystem; SMSsensor, for example, a camera; ttime; TDtechnical device; TVtarget variable; XPLexplainer

Claims

1. Computer-implemented method for anomaly detection during the control of a technical device (TD) by a control means (PLC), comprising the following steps: a) provision of input data (IN) by the control means, comprising sensor data which is captured by a sensor means (SM) at the technical device (TD), at a data input, coding of the input data (IN) with the aid of an autoencoder to form coded data (C) and decoding of the coded data (C) with the aid of the autoencoder to form output data (OUT), and determination and classification of a reconstruction error (RE), b) determination of a target variable (TV) from the reconstruction error (RE), if the reconstruction error (RE) exceeds a predetermined reconstruction error threshold (RET), c) provision of a model on the basis of artificial intelligence to an evaluation device (XPL), said model having been generated and trained with the aid of historic data (DH) from reconstruction errors and corresponding input features from the autoencoder, d) determination by the evaluation device (XPL) of at least one prediction for the target variable (TV) and optionally for further variables which depend on the target variable (TV), using the model and the classification of the reconstruction error (RE), and provision of the at least one prediction to an analysis device (REA), e) determination by the analysis device (REA) of a rule for improving the reconstruction error (RE) from the at least one prediction and a predetermined rule set, provision of the rule to the control means (PLC), and control by the control means (PLC) of the technical device (TD) with the aid of the rule.

2. System (S) for anomaly detection during the control of a technical device (TD) by a control means (PLC), comprising a sensor means (SM) for capturing sensor data relating to the technical device (TD), and a memory for storing historic data (DH), an evaluation device (XPL) and an analysis device (REA), the system (S) being designed to execute the method according to the preceding claims.

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

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

5. Computer program, comprising commands which, when executed by a computer, cause the system according to one of the preceding claims to execute the method according to claim 1.

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