Anomaly detection in a pneumatic system

Abstract

Fault detection module (FM) for the detection and evaluation of anomalies in a pneumatic automation system, comprising: - A read interface (I1) for reading digital signals from the automation system (AA), wherein the signals originate from at least two different digital sensors (S) and a switching command and represent times of two limit switches on a cylinder of the pneumatic system and a valve switching time. i ; - A first processor unit (P1) designed to execute a detection algorithm (S2) to calculate an anomaly score for the automation system (AA) based on the set of input signals; - A second processor unit (P2), which - if the anomaly score calculated with the first processor unit (P1) indicates an anomaly - is configured to execute a machine localization procedure (S34) to locate the fault, wherein the machine localization procedure (S34) has been trained in a training phase to calculate (S3, S4) and provide (S5) probabilities for possible causes of faults with respect to individual components (K) of the automation system based on a captured circuit diagram of the automation system (AA) and the calculated anomaly score, wherein the fault detection module (FM) is configured to calculate four time intervals from the three digital signals: - Reaction time when extending the cylinder; - Travel time when extending the cylinder; - Reaction time when retracting the cylinder; - Travel time when retracting the cylinder".

Description

[0001] The present invention relates to the technical fault detection and localization in a pneumatic automation system, e.g. in a production plant with actuators and sensors, and relates in particular to a fault detection module, a fault detection system, a method and a computer program.

[0002] Components and field devices in various types of automation systems are subject to stringent requirements regarding quality, robustness, and availability. A failure or malfunction of a field device in a process can incur extremely high costs, especially if it results in a production stoppage. Therefore, significant technical effort is invested in field devices to substantially reduce the risk of malfunctions or to enable them to independently detect and report defects. Functions are integrated into the field device with multiple redundancies, and measurement results are continuously monitored and verified internally. The requirements for the reliability of the field device increase with its application (e.g., in nuclear power plants). Monitoring field devices must therefore ensure that the devices involved function flawlessly and that a failure is detected as early as possible, ideally before a malfunction occurs.

[0003] For this monitoring and analysis task, predictive maintenance methods are used, which analyze a large amount of sensor data from field devices. These methods are often based on predictive maintenance algorithms. The quality of these algorithms correlates with the amount of available sensor data from continuously monitored systems. However, if only very little sensor data is available, these approaches often do not lead to satisfactory results.

[0004] In the field of automatic decision support, approaches using machine learning and neural networks are still well-known.

[0005] However, if systems with only minimal sensors, such as a pneumatic system with only two end-position sensors, need to be monitored for faults, the known approaches cannot be applied. These systems, however, still need to be monitored for faults.

[0006] Based on this, the present invention is based on the technical problem of presenting an approach with which a statement can be made about the faultiness of components of a pneumatic automation system in particular. Monitoring is thus to be improved and the automation system made more reliable overall. A statement about faultiness should be made at least at the component level.

[0007] This task is solved by a fault detection module according to claim 1. In one possible embodiment, it can be configured for the detection and evaluation of anomalies in automation systems, in particular in a pneumatic automation system, comprising: - A read interface, e.g. a digital OPC-UA interface, for reading digital signals from the automation system; in particular, only three digital signals may be available (time signals from two limit switches and the time of the valve switching command); - A first processor unit designed to execute a detection algorithm for calculating an anomaly score for the automation system based on the set of input signals; - A second processor unit, which – if the anomaly score calculated by the first processor unit indicates an anomaly – is trained to execute a machine localization procedure to locate the fault, in order to calculate and provide probabilities for possible fault causes with respect to individual components of the automation system based on the anomaly score. In an advantageous further development, the result can even be provided with respect to sub-components of the components – and thus in even greater detail.

[0008] The invention has the technical advantage that fault localization is possible directly in relation to the components of the automation system, even when only a few sensors, in particular only two limit position sensors, are installed. This enables fault localization based on only three digital signal values: the times at which the two limit position switches on a cylinder are activated and the time at which the valve switching signal is received (the valve switching signal represents the technical process when the controller instructs the valve with the command "SWITCH NOW" and can therefore also be referred to as the valve switching command).

[0009] For configuring or training the machine localization method, an additional (e.g., third) processor unit can be implemented. This additional processor unit comprises: - A circuit diagram import interface for importing a circuit diagram for the automation system; this serves the purpose of training a model to be generated for fault localization, preferably once, to import the digitized circuit diagram.

[0010] In a preferred embodiment of the invention, the first processor unit (which may be assigned the functionality of the detection algorithm) is implemented on a different device than the second processor unit (with the functionality of the machine localization method for locating the fault in the case of an increased anomaly score) and is, in particular, configured on a control unit. This allows the fault detection and localization system to be adapted very flexibly to the respective hardware, so that computationally intensive processes can be offloaded to high-performance hardware (e.g., cloud servers).

[0011] In an alternative, more preferred embodiment of the invention, the fault detection module comprises a configuration interface as a front end for configuring and training the model. This allows, for example, the user or operator of the system to configure the construction of the decision tree in a simple and quick manner.

[0012] In another preferred embodiment of the fault detection module, it is used for automation systems with a specific architecture or typical configuration. The pneumatic system comprises one or more pneumatic actuators, each connected to at least one valve. Multiple valves can be arranged on a valve manifold and / or multiple valve manifolds can be connected to a single supply unit. Multiple actuators can also be connected to a single valve simultaneously. The architecture is represented in the electronic circuit diagram, which is read by the system and used for calculations. Other embodiments may employ a different architecture.This is possible because the machine localization method takes the respective circuit diagram into account and automatically recognizes patterns of activity and deviations from patterns, and can locate possible errors based on the recorded circuit logic.

[0013] In another aspect, the invention relates to a fault detection system for recognizing and evaluating anomalies in a pneumatic system as claimed. In one possible embodiment, the system can be configured with: - A fault detection module as described above; - A gateway (to the internet, e.g., Edge Computer) and - A cloud-based server connected to the fault detection module via a web interface. The first and second processor units can be deployed (implemented and provided) on different units (controller, gateway, and / or server) as a distributed system. They can also be configured on the same unit.

[0014] The solution to the problem has been described above with reference to the devices (fault detection module, system). Features, advantages, or alternative embodiments mentioned in this context are also transferable to the other claimed items, and vice versa. In other words, the method and the computer program can also be further developed with the features described and / or claimed in connection with the module or system. The corresponding functional features of the method are thereby implemented by corresponding physical modules, in particular hardware modules or microprocessor modules, of the system or product, and vice versa.

[0015] In another aspect, the invention relates to a method as claimed. The method can be configured in an embodiment for the detection and evaluation of anomalies in a pneumatic automation system, comprising the following process steps: - Reading in - preferably two - digital signals from the respective (pneumatic) drive and from the digital switching command for the valve of the automation system via a reading interface; the outputs or senders of the signals are also referred to as 'sensors'; - Execution of a detection algorithm to calculate an anomaly score for the automation system based on the set of input signals; in an advantageous further development, the anomaly score is not only calculated overall for the entire automation system, but also broken down and dedicated to its individual drives. This significantly improves the informative value and provides more detailed information. - If the calculated anomaly score indicates an anomaly and, in particular, exceeds a preconfigurable threshold: triggering a machine localization procedure to locate the fault, wherein the machine localization procedure has been trained in a training phase to calculate, based on a captured circuit diagram of the automation system and the calculated anomaly score, probabilities for possible causes of faults with regard to individual components of the automation system or with regard to sub-components (parts) of the components and to provide them as a result.

[0016] The circuit diagram is advantageously read from a file during the commissioning training phase to configure data relationships and dependencies. Alternatively, the circuit diagram can also be programmed locally on the fault detection module or entered manually.

[0017] In an advantageous embodiment of the invention, the machine learning method (or the second processor unit) can be configured not only to output a result with the calculated error probabilities per component of the system, but also in a more detailed form, namely for each sub-component of a given component. This allows the result to be provided in an even more granular and specific way for individual parts or elements of a component.

[0018] In a preferred embodiment, a pattern recognition algorithm is used as the detection algorithm for calculating the anomaly score. Alternatively, the anomaly score can be calculated by accessing a memory containing a trained detection model. The model can be created using automatic classification methods, in particular a k-means algorithm. A training phase is provided for this purpose, during which further configurations can be created and the model is trained. The model serves to classify or differentiate between two classes: a first class with a normal response pattern of the pneumatic system and a second class with a deviant or abnormal response pattern. It should be noted that the detection algorithm preferably processes real-time signals or data generated during the operation of the automation system.This means that the detection algorithm preferably refers to the current state of the system.

[0019] According to the invention, the signals from at least two different digital sensors and the switching signal for the valve are read in and thus represent the times of two limit switches on a cylinder (tensioner) of the pneumatic system and the valve switching time. The following four time intervals are calculated from the three digital signals: - Reaction time when extending the cylinder (time interval from the switching point / valve until the current end position is left); - Travel time when extending the cylinder (time interval from leaving one end position to reaching the other end position); - Reaction time when retracting the cylinder (time interval from the switching point / valve until the current end position is left); - Travel time when retracting the cylinder.

[0020] This aspect has the advantage that, based on only three digital signals (or binary signals, on / off), four statements can be derived that have a significant influence on fault detection and, if necessary, fault localization. This allows fault detection to be applied even to existing systems that are not yet equipped with extensive sensor technology.

[0021] In an advantageous embodiment of the invention, in addition to the minimal sensor setup (with the three digital signals) sufficient to perform the detection algorithm and fault localization, an additional sensor can be integrated into the valve to detect whether and when the valve has switched. This signal can be described as the valve switching time. This additional digital signal provides an additional time indication from which more detailed information can be obtained. For example, if the time between "valve now switching" and "valve has switched" is constant, but a change in the reaction time is also detected, this change is not due to the valve. The localization method will therefore point to another possible source or cause of the fault.

[0022] In a further advantageous embodiment of the invention, in addition to the minimal sensor system, pressure sensors can be provided at the two working ports of each valve. This pressure sensor system is implemented, for example, in the applicant's motion terminal (designated VTEM) and can be used to provide further information for calculating the anomaly score and for fault localization, thus enabling a more detailed localization result. In this embodiment of the invention, a pressure signal is therefore also considered as a signal for calculating the anomaly score and for fault localization.

[0023] In a further advantageous embodiment of the invention, in addition to the minimal sensor technology, a pressure and / or flow sensor technology can be provided which can monitor several valve manifolds in order to also provide further information for the calculation of the anomaly score and for fault localization and thus a more detailed localization result.

[0024] In a further, preferred embodiment of the invention, the detection algorithm, after calculating the reaction time and travel time during extension and retraction of the cylinder, performs at least one of the following processing steps: - Feature extraction; this step serves to reduce the data volume. This makes the process faster; - Z-score normalization; this step serves for standardization and relates to the transformation of a random variable. This increases generalizability and comparability. The advantage lies in scaling the physical quantities to normalized, equally weighted values. - Principal Component Analysis (PCA); this step serves to structure and simplify the extensive sensor-acquired datasets by approximating a large number of statistical variables with a smaller number of highly informative linear combinations (principal components). This reduces computation time; - Classification, especially using K-means or comparable methods; - Logistic function where the result of K-means is mapped to values ​​between '0' and '1', thus normalizing the anomaly score to values ​​in an interval [0,...,1]; and / or - Smoothing; noisy sensor data is smoothed only at the end of processing. This allows sensitivity and specificity to be adjusted.

[0025] In a further preferred embodiment of the invention, the detection algorithm outputs an anomaly score in the range [0,...,1] and a sensor relevance value as an intermediate result of the method. This intermediate result can then be used in a subsequent step to apply the machine localization method.

[0026] In a further preferred embodiment of the invention, the machine localization method is based on a decision tree, wherein the decision tree is calculated based on the acquired circuit diagram. The circuit diagram can be read from a file, such as EPLAN, FluidDraw, or an AutomationML file, or files in similar formats (e.g., XML-based). Alternatively, other machine learning methods can be used. In particular, an artificial neural network can be trained in a preliminary training phase to locate the fault.

[0027] In a further, preferred embodiment of the invention, the machine localization method extracts data relations between the data records from the captured circuit diagram and from the read-in signals, wherein the data relations serve to locate the fault.

[0028] In an advantageous further development of the method, the result of the machine localization procedure includes a probability of error value for preferably all—or alternatively, for components of the pneumatic system selected as relevant—and / or for sub-components within a component. Other further developments may include the following processing steps: - Aggregating all error probability values ​​of all components; - Access to a memory containing a set of rules for locating the fault in relation to individual components of the automation system.

[0029] The machine localization process comprises two stages for fault localization. In the first stage, the system calculates in which component of the automation system the fault is located. Fault localization in the first stage is therefore performed at the component level. The result might be, for example: "Clamp X is jammed" or "Valve Y is defective." In the second stage, the system calculates precisely where the fault can be located within the component. Fault localization in the second stage is therefore performed at the sub-component level. The result might be, for example: "Friction on the cylinder," "Leakage at cylinder chamber A," "Hose B has a leak," "Throttle D is clogged," etc.

[0030] In the automated localization method, the probability is first determined for preferably all components (clamp components). In an advantageous alternative embodiment of the invention, the probability can be determined only for components identified as relevant (e.g., in a configuration phase) to reduce the computational load and potentially provide the result more quickly. Subsequently, it is deduced whether the fault occurs in the identified clamp or whether all clamps of a valve are affected. If the latter is the case, the rule set is used to conclude that there is a problem with the valve. If all valves of a valve manifold exhibit an anomaly, the rule set indicates that the problem lies at the valve manifold level. The fault localization can thus be increasingly refined by accessing the rule set and narrowing down to specific components of the system.

[0031] Another solution involves a computer program with computer program code to carry out all the steps of the procedure described above, when the computer program is executed on a computer. It is also possible for the computer program to be stored on a computer-readable medium.

[0032] Another solution involves a computer program product containing computer program code to execute all the steps of the procedure described above when the computer program is run on a computer. The computer program product can be, for example, a saved, executable file, possibly with additional components (such as libraries, drivers, etc.), or an electronic unit (microprocessor, computer) with the computer program already installed.

[0033] The following section provides a more detailed explanation of the terminology used in this application.

[0034] The machine localization method is a purely computer-implemented procedure. It serves to predict faults occurring in specific components of the system. For this purpose, a decision tree can be constructed, representing a model. This model can be stored in memory. At runtime, the decision tree is used to assign objects (here: the individual components of the system, such as the valve, a valve group, the compressed air supply, the electrical supply, etc.) to fault classes. Probabilities can be assigned in this process.

[0035] For example, a Bayesian network or another decision logic can be applied. The basic idea is to deduce the probability of component-based and sub-component-based error sources from observing the three digital signals of the pneumatic system over its operating time. If a joint probability distribution of a larger number of variables is to be managed, explicit representation by specifying a probability per state combination quickly reaches resource limits (waiting time, processor capacity, etc.). For example, in the case of 20 binary variables, i.e., 20 variables with two states each, already 2 20= 1,048,576 individual values ​​can be specified. By exploiting (conditional) independence between variables of the domain to be modeled, the required number of values ​​to be specified can often be reduced to a manageable size. Bayesian networks represent such an approach. A Bayesian network of random variables consists of two parts: 1. A directed acyclic graph whose nodes correspond to the random variables and whose edges encode the conditional independencies between the variables. 2. Tables of conditional probabilities associated with the variables.

[0036] The decision tree is built during a training or learning phase and then traversed top-down when used for prediction or fault localization. Alternative techniques for machine localization include, for example, neural networks, Naive Bayes classifiers, k-nearest neighbor methods, or support vector machines.

[0037] The detection algorithm is a computer-implemented method for grouping or classifying data sets representing pneumatic system states (normal / anomalous) based on the detected signal combinations. For example, a k-means algorithm can be used. The goal of the k-means algorithm is to divide the data set into k (especially 2) partitions such that the sum of the squared deviations from the cluster centers is minimized. In extended embodiments of the invention, the k-median algorithm, the k-means++ algorithm, or comparable classification algorithms can also be used.

[0038] The input interface is a digital interface. It is used to read digital data and can be operated according to the OPC Unified Architecture (OPC UA) protocol. OPC UA is an industrial machine-to-machine communication protocol for ensuring interoperability. Data from fieldbuses, such as Profinet, can also be read.

[0039] The signals are digital (on / off) that can be directly processed digitally by the processor units. Preferably, directly digital sensors are used. With a digital sensor, the electrical signal is directly converted to digital (sensor-internal A / D conversion). Subsequent calculations (e.g., error compensation) can take place in a microprocessor. Alternatively, analog sensors can be used, whose signal is transformed into a digital signal in an external or separate A / D converter. The digital signal is then available as a numerical value and can be output via any digital protocol such as USB, CANopen, or Profibus. During further transmission, the digital pressure signal is immune to interference that could impair its accuracy.

[0040] A fault detection module is an electronic module that can be distributed across multiple components and is designed to prevent and locate faults in pneumatic systems. Specifically, the fault detection module, which can be implemented locally on devices within the automation system, should be able to access centrally executed, and especially cloud-based, calculations. The fault detection module is configured to implement control and / or diagnostic measures when the maintenance software detects a potential failure of an automation system component at an early stage. Defective components that could soon lead to system shutdown are thus identified independently of regular maintenance schedules and can be replaced before actual damage occurs.This allows for cost savings compared to routine or time-based preventive maintenance, as tasks are only performed when actually necessary. Within the scope of this invention, it is preferred that the analysis be carried out in parallel with the operation of the system in order to avoid downtime.

[0041] The gateway (node) is a computer-based unit, can be configured as an edge computer close to the field, and has a cloud-based interface (web interface) to the server. The gateway calculates the anomaly score and processes it further as part of the fault localization process. The result can be forwarded to a server and / or at the field level (e.g., PLC).

[0042] A component is a field device, and thus a technical piece of equipment in the field of automation technology that is directly related to a production process. In automation technology, "field" refers to the area outside of control cabinets or control rooms. Field devices can therefore be both actuators (actuators, valves, etc.) and sensors (transmitters) in factory and process automation. The components are connected to a control system, usually via a fieldbus. The components can be equipped with sensors to acquire, generate, or aggregate sensor data so that the data can be evaluated and used for regulation, control, and further processing. The components are part of an automation system, which may include other devices (e.g., industrial robots).

[0043] A control unit is an electronic module used to control and / or regulate a machine or automation system with a group of field devices and is programmed digitally. It can be, in particular, a programmable logic controller (PLC). In its simplest form, a control unit has inputs, outputs, an operating system (firmware), and an interface through which the user program can be loaded. The user program defines how the outputs are to be switched depending on the inputs. The operating system ensures that the user program always has access to the current state of the sensors. Based on this information, the user program can switch the outputs so that the machine or automation system functions as desired.The control unit is connected to the automation system with its field devices using sensors and actuators.

[0044] The following detailed description of the figures discusses exemplary embodiments, which are not to be understood as restrictive, along with their features and further advantages, using the drawing as an example. Brief overview of the characters

[0045] The following detailed description of the figures discusses exemplary embodiments, which are not to be understood as limiting, along with their features and further advantages, based on the drawing. This drawing shows: Fig. 1 an overview of the fault detection system according to the invention with a fault detection module; Fig. 2 one for representation in Fig. 1 alternative design of an error detection module; Fig. 3 another schematic representation of a fault detection module with a cloud-based server and other components; Fig. 4 an alternative schematic representation of the fault detection module; Fig. 5 a flowchart of process steps of a fault detection method according to a preferred embodiment of the invention and Fig. 6 a schematic representation of a fault detection system with further components according to a preferred embodiment of the invention. Detailed description of exemplary implementations using the figures

[0046] The invention serves for the technical monitoring of a pneumatic system as an example of an automation system or plant with various field devices (hereinafter also referred to as components) that are controlled by a control unit (e.g., PLC). In particular, faults are to be detected early and preferably at a point in time before the respective component fails or causes a fault in the plant. For this purpose, a fault detection module is to be used, which is described in more detail below with regard to Fig. 1 is explained.

[0047] The invention offers the advantage of enabling early fault detection in complex, multi-component automation systems—preferably pneumatic—even when very little measurement data is available and the systems can be operated with minimal sensors. In particular, it allows for fault localization by using only two digital sensors and a switching command, specifically for recording the times of two end-position sensors on a cylinder and one sensor for recording the valve switching time. This has the advantage of enabling anomaly detection even in systems where only the actuator is equipped with sensors (e.g., end-position sensors). The method presented here is based on a model that takes at least these signals into account.Optionally, additional signals, such as pressure and / or flow signals or other signals from valve-internal sensors, can be considered. These signals are acquired in the pressure supply and / or the valve itself. Using the detection algorithm, deviations or changes from the correct or typical response behavior of the pneumatic system can be automatically and in real time detected. Examples include the time between "valve switching" and "leaving end position 1" and the travel time (end position 1 to end position 2). Furthermore, the time between sending the control command and the physical switching of the valve can, in principle, be measured and learned. In an advantageous further development, an additional valve-internal sensor can be incorporated to detect when the valve has switched. The same applies to the valve's return movement. The measured values ​​and the resulting patterns are learned during "good" operation (i.e., during error-free operation).Fault patterns exhibit characteristic characteristics that, according to the invention, are used for anomaly detection and fault localization. The circuit diagram of the pneumatic system is also available in a digital pneumatic circuit diagram, which is imported, for example, from a Fluid Draw, Eplan, or Automation ML file, and used to build a decision logic. If a deviation from the GUT pattern is detected by the detection algorithm, fault localization can be provided in a second step by applying a machine localization method. For this purpose, a logic circuit with implemented decision logic can be used, for example, using a decision tree, Bayesian networks, or other machine learning methods.

[0048] The background to the proposed solution is that the timing behavior of a clamping system (e.g., in automotive manufacturing, bodywork) – consisting of a valve, hose system, and clamps – changes with increasing wear. A test setup was created to identify whether and how manipulations of the pneumatic system affect this timing behavior. Specific variations and manipulations were performed on the pneumatic system. This included adjusting friction and leakage at the clamp and valve, as well as changing the lever arm length, the hose length between the valve and clamp, and varying the supply pressure. The closing time and deceleration time were recorded during the cylinder's opening and closing phases. The applicant's tests show that changes in friction, leakage, and the clamp's supply pressure affect the deceleration and closing times derived from the limit switch signals.The results from the test setup are incorporated into the configuration of the fault localization model. In the first stage, the fault is localized with respect to individual components of the system, and in the second stage, it is localized with respect to individual sub-components of the component. This allows for the unambiguous identification of the type of fault. Consequently, the fault can be narrowed down and, in particular, localized based on the (three) digital signals.

[0049] In Fig. Figure 1 shows a schematic representation of the fault detection module FM. On the automation system AA side, it comprises the components K – preferably pneumatic – such as valve manifolds or valve discs. Each valve manifold, in turn, comprises several valves with clamping / cylinder units and / or further pneumatic actuators (e.g., pneumatic drives, etc.) and sensors, as well as a pressure supply. A controller is also provided, which can be, for example, a programmable logic controller (PLC). The components K are equipped with sensors S, which serve to detect digital signals or switching commands to a valve. A first component K1 comprises at least one sensor unit S1 for detecting three digital signals, a second component K2 in turn comprises a sensor unit S2 for detecting at least three digital signals, and so on.

[0050] As in Fig. As shown in Figure 1, additional sensors S3 can also send signals (e.g., pressure signals) to the PLC. The PLC receives the digital signals via a read interface 11 and is further equipped with a first processor unit P1, which serves to execute a detection algorithm based on the detected or read signals. The detection algorithm is used to calculate an anomaly score for the automation system AA based on the number of detected or read signals. The calculated anomaly scores can be transferred to an IoT gateway GW via a data interface (e.g., OPC UA). The calculated anomaly scores and / or the detected signals are transmitted via a second interface 12 to a second processor unit P2, which—if the anomaly score calculated with the first processor unit indicates an anomaly—may be configured to implement a machine localization method S34 (described in more detail below with reference to Figure 1). Fig. 5 is described) to perform the localization of the fault in order to calculate and provide as a result the anomaly score, probabilities for possible causes of faults with regard to individual components K of the automation system AA.

[0051] In the Fig. In the example shown, the first processor unit P1 is implemented on a different device than the second processor unit P2. The first processor unit P1 can be implemented on the PLC (Programmable Logic Controller), and the second processor unit P2 can be implemented, for example, on a gateway node GW (also referred to simply as a gateway). To execute the machine localization procedure, the second processor unit P2 accesses a memory MEM (Memory Memory) in which a trained model is stored. The second processor unit P2 receives a circuit diagram of the pneumatic system AA via a circuit diagram input interface 13. The circuit diagram is in digital form and contains information about the structure of the system AA and its functionality (in particular, the switching times of the valves, etc.).

[0052] In the Fig. In the embodiment shown in Figure 1, a separate gateway GW is provided, which serves as an intermediary between, on the one hand, the plant AA with its components K and programmable logic controller (PLC), and, on the other hand, the server SV. The gateway GW can, for example, be implemented in a higher-level control system of the plant AA and / or be assigned to the plant AA (e.g., in the same security domain as the plant). A third processor unit P3 can be implemented on the server SV to, for example, execute the machine localization procedure on a cloud-based server.

[0053] As in Fig. As schematically indicated in Figure 1, it is fundamentally possible for the first processor unit P1 to send the locally calculated anomaly scores to the second processor unit P2 as an intermediate result (solid arrow). Alternatively or cumulatively, the acquired signals can also be transmitted to the second processor unit P2. This can be done either directly from the sensor S and / or from component K (both are shown in Figure 1). Fig. 1 (shown with a dashed line) and / or by the PLC control system.

[0054] Fig. Figure 2 shows an alternative embodiment in which the gateway GW comprises both the second and the first processor units P2, P1. The components K send their three digital signals to the PLC, which then forwards the signals via the network connection (second interface 12) to the second processor unit P2 for processing. Alternatively, the components can send the locally acquired signals directly to the second processor unit P2 (without going through the PLC). It is even conceivable that the sensors themselves could be equipped with an additional network interface to transmit the data.

[0055] In Fig. Figure 3 shows an embodiment using a cloud-based server SV. The sensor data is again acquired on the components K of the pneumatic system AA. The first processor unit P1 can now be implemented either locally in the PLC or on an IoT gateway node GW assigned to the system, which can be configured as an edge computer. The gateway GW exchanges data with the server SV, on which the second processor unit P2 is implemented, via an internet protocol-based data connection (e.g., HTTPS). The second processor unit P2 is configured to execute the machine localization procedure. The learned model can be stored in the memory MEM of the server SV. This makes it possible to utilize the server's higher computing (and memory) resources for fault localization and for calculating the result.

[0056] As the preceding examples are intended to show, the functionality of the FM fault detection module can be combined with the two aspects: 1. Detection algorithm S2 and 2. The machine localization method S34 can also be implemented in a distributed manner.

[0057] In other words, the first processor unit P1 and the second processor unit P2 can be implemented on different computer-based entities. It is also possible to implement a further processor unit dedicated to configuring the model or training the localization algorithm using training data. This training data can include patterns of signal combinations in GUT (Good Under Trust) scenarios.

[0058] As in Fig. As shown in Figure 4, it is preferred that the detection algorithm S2 be executed as locally as possible, near the generated data, preferably in the PLC controller, and that the machine localization procedure S34 be executed on an entity that provides sufficient resources, preferably on the server SV. Only one client for model verification for the machine localization procedure S34 can then be installed on the gateway GW, so that the computationally intensive processes can be executed on the server SV and only the result is output to configurable entities, in particular to the gateway GW and, if necessary, to the components K of the plant AA and / or to the PLC controller. The output can be provided via an output interface AS.

[0059] In Fig. Figure 5 shows a flowchart of a fault detection procedure. After starting, the digital signals are read in step S1. In step S2, the detection algorithm is executed on or with the read signals. This calculates an anomaly score and a sensor relevance value as an intermediate result. The intermediate result thus represents whether an anomaly exists in the system AA or not. Depending on the result, as shown in Fig. As shown in section 5, the process branches into different calculation scenarios. If no anomaly is present, the system appears to be functioning "as usual"—that is, without errors. The process can be terminated or restarted with an EXIT command. Otherwise (if an anomaly or deviation is detected), a machine localization procedure is executed in step S34. This procedure, which has been trained in a training phase, uses a digitally or manually recorded circuit diagram of the automation system AA to calculate probabilities for possible causes of the error based on the calculated anomaly score. The machine localization procedure can comprise two stages. In the first step, S3, the fault is localized at the component level (e.g., fault in clamp X or valve Y), and in the second step, S4, the fault is localized at the sub-component level.In the second step, S4, the location of the fault within the identified component is analyzed. The machine localization procedure can be implemented as an algorithm that is executed taking into account the information from the captured circuit diagram (circuit layout, architecture, and structure, as well as switching points). As shown above, the functionality of the algorithm can also be implemented on other devices or servers (SV).

[0060] Fig. Figure 6 is another structural architecture diagram of a fault detection system with a first processor unit P1, implemented here on the PLC, and a second processor unit P2, implemented on the server SV, which exchanges data with the gateway GW via a data connection. Additionally, a configuration interface Config-Ul can be provided, which allows the configuration of the machine localization procedure and, in particular, the algorithms S3 and S4. The configuration interface Config-Ul is preferably cloud-based or can also be provided locally as a computer program. The configuration interface Config-Ul can include user interface elements such as dashboards. A version of the learned model (e.g., a constructed decision tree) can also be installed here, with a training master as an application for configuring the learning phase for the model.The decision tree is generated using a scoring master application to calculate the anomaly score, according to another option. A suite of applications for fault detection and localization can be installed on the server SV (e.g., an industrial PC). Specifically, a runtime environment (e.g., a Java Runtime Environment) for the trained model is implemented, synchronized with the configuration interface Config-Ul, and interacts with the gateway GW, preferably via HTTPS / REST upload requests. The received signals are then sent via the gateway GW to the server for fault detection and localization.

[0061] In one embodiment, another processor unit, which is in Fig.The third processor unit, designated P3, is provided for generating the model for the machine localization procedure S34. Users can configure settings via the Config-Ul interface. The model generation functionality can also be implemented on the SV server.

[0062] In this embodiment, the IoT gateway node (GW) can be configured with a client for the machine localization process. The client / gateway can be positioned in the field near the equipment. The gateway node (GW) can have a browse function that allows users to scroll through and review the anomaly scores transmitted by the PLC. Furthermore, the gateway node (GW) can provide a proxy for the algorithm, which can be run in the cloud (e.g., on the server SV), and a proxy for an automation suite with additional applications and programs as a PC application. The functionality of the automation suite is similar to that of the cloud application.Furthermore, the gateway (GW) can have a ring buffer for temporary data storage and a lite version of the trained model (for executing the machine localization process) for persistence, configuration, license management, and other functionalities related to the machine localization process. Depending on the configuration, other programs can also be installed on the gateway (GW), which may run in the background and provide certain services. User interaction preferably occurs only indirectly, for example, via signals, pipes, and especially (network) sockets.

[0063] In an experiment, six pneumatic clamps were operated continuously for an extended period until they reached the point of wear, with increased operating time or a shortened cycle time compared to normal operation. Signs of wear were detectable in the data for all clamps two weeks prior to failure. According to the invention, failures and induced faults can be detected using the machine localization method or a trained model, enabling automated process monitoring.

[0064] Finally, it should be noted that the description of the invention and the exemplary embodiments are not to be understood as limiting with regard to a specific physical realization of the invention. All features explained and shown in connection with individual embodiments of the invention can be provided in different combinations in the subject matter of the invention in order to simultaneously realize their advantageous effects.

[0065] The scope of protection of the present invention is defined by the following claims and is not limited by the features explained in the description or shown in the figures.

[0066] It is particularly obvious to a person skilled in the art that the invention can be applied not only to pneumatic systems, but also to hydraulic systems or other fluid power systems or electric axes. Furthermore, the components of the fault detection module can be implemented distributed across several physical products.

Claims

[1] Fault detection module (FM) for the detection and evaluation of anomalies in a pneumatic automation system, comprising: - A read interface (I1) for reading digital signals from the automation system (AA), wherein the signals originate from at least two different digital sensors (S) and a switching command and represent times of two limit switches on a cylinder of the pneumatic system and a valve switching time. i ; - A first processor unit (P1) designed to execute a detection algorithm (S2) to calculate an anomaly score for the automation system (AA) based on the set of input signals; - A second processor unit (P2), which - if the anomaly score calculated with the first processor unit (P1) indicates an anomaly - is configured to execute a machine localization procedure (S34) to locate the fault, wherein the machine localization procedure (S34) has been trained in a training phase to calculate (S3, S4) and provide (S5) probabilities for possible causes of faults with respect to individual components (K) of the automation system based on a captured circuit diagram of the automation system (AA) and the calculated anomaly score, wherein the fault detection module (FM) is configured to calculate four time intervals from the three digital signals: - Reaction time when extending the cylinder; - Travel time when extending the cylinder; - Reaction time when retracting the cylinder; - Travel time when retracting the cylinder". [2] Fault detection module (FM) according to claim 1, wherein the first processor unit (P1) is implemented on a different device than the second processor unit (P2) and in particular on a control unit (PLC). [3] Fault detection module (FM) according to at least one of the preceding claims, wherein the second processor unit (P2) or a further processor unit (P3) configured to generate a model comprises a circuit diagram input interface (I3) for inputting a circuit diagram for the automation system (AA) in digital form. [4] Fault detection module (FM) according to at least one of the preceding claims, comprising a configuration interface (Config-Ul) as a frontend for configuring and / or training a model. [5] Fault detection module (FM) according to at least one of the preceding claims, wherein the automation system (AA) comprises a pneumatic system with a pneumatic drive, in which several drives and / or actuators are connected to a valve and several valves are arranged on a valve manifold and several valve manifolds are connected to a supply unit. [6] Method for detecting and evaluating anomalies in a pneumatic automation system (AA), comprising the following process steps: - Reading (S1) digital signals from the automation system (AA) via a reading interface (I1), wherein the signals originate from at least two different digital sensors (S) and a switching command and represent times of two limit switches on a cylinder of the pneumatic system and a valve switching time; - Executing a detection algorithm (S2) to calculate an anomaly score for the automation system based on the number of signals read in; - if the calculated anomaly score indicates an anomaly: triggering a machine localization procedure (S34) to locate the fault, wherein the machine localization procedure (S34) has been trained in a training phase to calculate (S3, S4) probabilities for possible fault causes in relation to individual components (K) of the automation system (AA) based on a captured circuit diagram of the automation system (AA) and to provide them as a result (S5), - where four time intervals are calculated from the three digital signals: - Reaction time when extending the cylinder; - Travel time when extending the cylinder; - Reaction time when retracting the cylinder; - Travel time when retracting the cylinder. [7] Method according to the preceding method claim, wherein the detection algorithm (S2) for calculating the anomaly score is a pattern recognition algorithm or is carried out by accessing a memory in which a trained detection model is stored. [8] Method according to any of the preceding method claims, wherein the machine localization method (S34) calculates probabilities for possible causes of failure with respect to individual sub-components of a component (K) (S4). [9] Method according to any of the preceding method claims, wherein the signals from two limit switches are read and comprise a valve switching time signal and / or a pressure signal and / or a flow signal. [10] Method according to any of the preceding method claims, wherein the detection algorithm (S2) performs the following processing steps after calculating the reaction time and travel time when extending and retracting the cylinder: - Feature Extraction; - Z-score normalization; - Principal Component Analysis; - Classification, especially using K-means; - Logistical function; and / or - Smoothing. [11] Method according to any of the preceding method claims, wherein the detection algorithm (S2) comprises an anomaly score and a sensor relevance value as a result. [12] Method according to any of the preceding method claims, wherein the machine localization method (S34) comprises a decision tree method, wherein the decision tree is calculated on the basis of the captured circuit diagram or comprises a Bayesian network method. [13] Method according to one of the preceding method claims, wherein the machine localization method (S34) extracts data relations between data sets from the captured circuit diagram based on the input signals. [14] Method according to any of the preceding method claims, wherein the result of the machine localization method (S34) includes an error probability value for all components (K) and / or all sub-components of the components and wherein the method further performs the following processing steps: - Aggregating all error probability values; - Access to a memory containing a set of rules for locating the fault with regard to individual components (K) and / or sub-components of the automation system (AA). [15] Fault detection system for the detection and evaluation of anomalies in automation systems (AA), in particular in a pneumatic system, comprising: - A fault detection module (FM) according to one of the claims directed to the fault detection module; - A gateway (GW) and - A cloud-based server (SV) that is connected to the fault detection module (FM) via a web interface (WSS). [16] Computer program with computer program code for carrying out all process steps of a process according to any of the preceding process claims, when the computer program is executed on a computer.

Images