EP4193227A1 - Dispositif et procédé de détection d'anomalies dans une installation industrielle pour l'exécution d'un processus de production - Google Patents
Dispositif et procédé de détection d'anomalies dans une installation industrielle pour l'exécution d'un processus de productionInfo
- Publication number
- EP4193227A1 EP4193227A1 EP21778101.2A EP21778101A EP4193227A1 EP 4193227 A1 EP4193227 A1 EP 4193227A1 EP 21778101 A EP21778101 A EP 21778101A EP 4193227 A1 EP4193227 A1 EP 4193227A1
- Authority
- EP
- European Patent Office
- Prior art keywords
- production process
- anomaly
- measurement data
- anomalies
- performance indicators
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 112
- 238000000034 method Methods 0.000 title claims abstract description 90
- 238000013473 artificial intelligence Methods 0.000 claims abstract description 65
- 230000008569 process Effects 0.000 claims abstract description 48
- 238000005259 measurement Methods 0.000 claims description 79
- 238000001514 detection method Methods 0.000 claims description 27
- 238000012549 training Methods 0.000 claims description 25
- 238000004590 computer program Methods 0.000 claims description 7
- 239000002994 raw material Substances 0.000 claims description 6
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 6
- 241000196324 Embryophyta Species 0.000 description 36
- 239000000047 product Substances 0.000 description 28
- 238000012805 post-processing Methods 0.000 description 14
- 238000004422 calculation algorithm Methods 0.000 description 12
- 238000013528 artificial neural network Methods 0.000 description 11
- 238000004088 simulation Methods 0.000 description 8
- 238000007405 data analysis Methods 0.000 description 7
- 238000012545 processing Methods 0.000 description 7
- 238000004891 communication Methods 0.000 description 6
- 239000000203 mixture Substances 0.000 description 5
- 230000009467 reduction Effects 0.000 description 5
- 230000005856 abnormality Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 4
- 230000006399 behavior Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000009434 installation Methods 0.000 description 3
- 238000007726 management method Methods 0.000 description 3
- 238000004886 process control Methods 0.000 description 3
- 239000000126 substance Substances 0.000 description 3
- 238000012731 temporal analysis Methods 0.000 description 3
- 238000000700 time series analysis Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000002955 isolation Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000003754 machining Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000013341 scale-up Methods 0.000 description 2
- 230000002123 temporal effect Effects 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 239000007795 chemical reaction product Substances 0.000 description 1
- 230000001276 controlling effect Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013434 data augmentation Methods 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000007717 exclusion Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 239000007789 gas Substances 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 229910052751 metal Inorganic materials 0.000 description 1
- 238000010327 methods by industry Methods 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 239000003921 oil Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000000123 paper Substances 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000012913 prioritisation Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0243—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults model based detection method, e.g. first-principles knowledge model
- G05B23/0254—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults model based detection method, e.g. first-principles knowledge model based on a quantitative model, e.g. mathematical relationships between inputs and outputs; functions: observer, Kalman filter, residual calculation, Neural Networks
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0224—Process history based detection method, e.g. whereby history implies the availability of large amounts of data
- G05B23/024—Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
- G05B19/41875—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM] characterised by quality surveillance of production
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0221—Preprocessing measurements, e.g. data collection rate adjustment; Standardization of measurements; Time series or signal analysis, e.g. frequency analysis or wavelets; Trustworthiness of measurements; Indexes therefor; Measurements using easily measured parameters to estimate parameters difficult to measure; Virtual sensor creation; De-noising; Sensor fusion; Unconventional preprocessing inherently present in specific fault detection methods like PCA-based methods
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/20—Pc systems
- G05B2219/24—Pc safety
- G05B2219/24015—Monitoring
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Definitions
- the invention relates to a device and a method for detecting anomalies in an industrial plant for carrying out a production process, the plant comprising a large number of sensors for measuring process variables of the production process.
- WO 2019/141593 A1 discloses a device for detecting anomalies in an actuator system (for example a system of motors, pumps or conveyor belts).
- the device includes an anomaly detector with a trained artificial in- Intelligence that is designed and trained to detect anomalies in the actuator system based on a large number of output data from the actuator system.
- the anomaly detector outputs an anomaly signal.
- the output data are generated, for example, with the aid of sensors that are arranged in or on the actuators.
- defects or impending failures of the actuators can be detected early and countermeasures (e .g . repair, maintenance, replacement) can be initiated in good time.
- US 2015/0324329 A1 discloses a process modeling technique that uses a single statistical model that was developed from historical data for a typical process and uses this model to perform quality prediction or error detection for various process states of a process. In each case, separate models are used for quality prediction and error detection. The modeling technique does this by determining mean values (and possibly standard deviations) of process parameters for each of a set of product grades, throughputs, etc. , and compares them with online process parameter measurements.
- Process analysis programs can be stored and executable in different devices of a process control system.
- An overall quality parameter for the entire production process or for a product resulting from the entire production process can also be determined by means of a central Process Monitoring and Quality Prediction System (PMS).
- PMS Process Monitoring and Quality Prediction System
- CN 107797537 A discloses a fault prediction system for bearings for rotating components in an automated production line. Real-time monitoring of the bearings is carried out on the basis of measurement data from sensors with the help of a deep learning neural network model. Failure prediction is thus limited to just the bearings on the production line.
- US Pat. No. 5,877,954 A discloses a hybrid analysis device comprising a data-derived primary trained model (e.g. a linear model such as a partial least squares model) for predicting output variables of an industrial process and an error correction model (e.g . a non-linear model such as a neural network) for error correction of the outputs of the primary model .
- a data-derived primary trained model e.g. a linear model such as a partial least squares model
- an error correction model e.g a non-linear model such as a neural network
- this is intended to enable the entire plant to be operated optimally from different points of view and to support the plant's managerial staff in making the necessary operational decisions.
- the device serves to detect anomalies in an industrial plant for carrying out a production process for a product, the plant comprising a large number of sensors for measuring process variables of the production process.
- the device includes an anomaly detector with at least one trained artificial intelligence that is designed and trained to detect and/or predict anomalies in the production process based on a large number of measurement data from the sensors.
- the detector is designed to output anomaly information when an anomaly is detected and/or predicted.
- the anomaly information preferably also includes a period of time in which the anomaly occurred and/or a probability of the existence of the anomaly.
- the anomaly detector is designed to simultaneously detect and predict anomalies in several different performance indicators of the production process, with the performance indicators each relating to the entire production process of the product.
- the performance indicators relate to the entire production process of a resulting (end) product.
- the artificial intelligence is educated and trained to both detect and predict anomalies, i . H . for example, a common data model is used for both detecting and predicting anomalies.
- the invention is based on the knowledge that the operating or business goals of the plant operator can change during operation of the plant, for example depending on a market situation. This also affects the operation of the system. At a given point in time, it is therefore important to detect and/or predict those anomalies in the production process that have the greatest impact on the currently prevailing operational or business objective.
- Such operational or business goals usually relate to the entire production process. Accordingly, performance indicators obtained from the sensor data must also relate to the entire production process.
- performance indicator is understood to mean a key figure that is used to measure the degree of fulfillment with regard to a specific objective (e.g. an operational or business objective) with regard to the production process can be.
- a performance indicator is preferably a physically measurable variable, such as the purity or dimensional accuracy of a product, a number of items, a volume, a quantity, etc..
- the performance indicator is therefore preferably one of the measured process variables or a variable derived from them (e.g an average of a process variable) is used.
- production process is understood here not only as a manufacturing or manufacturing process, but also as a machining, processing or conversion process (e.g. also an energy generation process).
- the measurement data can be data on process variables that are based directly on measurement values. However, it can also be data on process variables that were derived from measured values of other process variables.
- the artificial intelligence can be trained by an algorithm from the field of machine learning or with the help of rule-based or mathematical (e.g. statistical) methods. Examples of such algorithms or methods are neural networks, autoencoders, Gaussian Mixture Model, Boosted Gaussian Mixture Ensemble and Isolation Forest, as well as combinations of all these methods.
- the invention thus makes it possible to define various different performance indicators that represent, for example, different operational or business goals, and then to monitor selectively, depending on the goal, for anomalies that have the greatest impact on the achievement of this goal.
- the invention can also be applied to systems in which several different production processes are carried out at the same time. Anomalies in performance indicators of the different production processes can then also be detected and predicted with the invention.
- the invention can run as a kind of "anomaly assistant" accompanying the operation of the system. Measures can be proposed to the system operator depending on the type and duration of an anomaly. Management personnel of the system can thus be supported in necessary operational decisions. The proposed measures can automatically also flow directly into the control and/or regulation of the system or the respective production process plant are transferred. This brings the plant or the respective production process into line with the operating or business objectives of the plant.
- the anomaly detector can each have a separate anomaly detector unit for each of the performance indicators.
- Each of the anomaly detector units can each include a separate trained artificial intelligence that is designed and trained to detect and predict the anomalies in the production process for the respective performance indicator.
- the anomaly detector is designed to determine, for a detected and/or predicted anomaly, a relevance of the anomaly in relation to at least one, in particular several, different operational or business goal(s) of the production process.
- the anomalies or Countermeasures for this can then be prioritized very easily and quickly by the operating personnel of the production process based on the relevance.
- the relevance can be determined particularly easily by weighting the performance indicators (e.g. throughput, quality) depending on the at least one, in particular several, different operating or business goals.
- performance indicators e.g. throughput, quality
- the system operator can then visualize the relevance particularly easily by classifying the anomalies into different message classes (e.g. simple message, warning, alarm). For example, if the quality of the end product is more important than throughput for a particular operational or business objective, then an anomaly in throughput would only result in a "warning" and an anomaly in quality would result in an "alert” .
- the anomaly detector preferably includes means for acquiring selection information about at least one current operating or business goal of the production process.
- the device includes a configuration device that is designed to configure the anomaly detector, in particular the artificial intelligence, depending on at least one operating or business goal of the production process.
- the anomaly detector can then be better adapted to the respective task. For example, a specially adapted selection of measurement data or a special combination of data analysis algorithms can be set in this way, which ensures a particularly good quality of the anomaly detection for the various performance indicators.
- the anomaly detector preferably has a common, trained artificial intelligence for all of the multiple performance indicators.
- a common trained artificial intelligence serves for the simultaneous detection and prediction of anomalies in the multiple performance indicators.
- the anomaly detector is advantageously designed in such a way that it uses a number of algorithms that interact with one another at the same time. As a rule, better results can be achieved in anomaly detection as a result.
- the anomaly detector can use several algorithms of the same type that work together at the same time (e.g. an ensemble of neural networks or a boosted ensemble). However, it can also use several different algorithms (e.g. multistage systems with autoencoders, random forests and convolutional neural networks) that work together at the same time.
- the anomaly detector is advantageously designed as a multi-stage system with a number of data analysis and processing stages arranged one behind the other.
- the artificial intelligence is designed and trained to take into account a chronological sequence of the measurement data and/or chronological relationships between the measurement data when detecting and/or predicting the anomalies.
- a time series analysis is carried out by the artificial intelligence. Changes in the time series of measurement data can thus be recognized and future time series of performance indicators can be predicted.
- a high degree of robustness and high accuracy of the anomaly detection can be achieved in this way (ie fewer “false positives”).
- the measurement data can be current and historical measurement data.
- An advantage is the artificial intelligence on a normal state of the production process or. of performance indicators trained . For this purpose, "bad states" are removed from the training data. The artificial intelligence thus searches for abnormalities in the performance indicators that differ from the trained normal state. This enables a very broad application of anomaly detection even with only a few historical anomaly data.
- the artificial intelligence is at least partially trained with simulated measurement data from sensors.
- the simulated measurement data can, for example, be generated by a simulator (or a digital twin) for the production process.
- the simulator digital twin
- the simulator can be based on physical/chemical models of the production process.
- the anomaly detector is preferably designed in such a way that it performs a validation of the anomaly detection and/or prediction based on deviations between the (real) measurement data of the sensors and simulated measurement data of the sensors.
- the anomaly detector can thus use deviations between the results of the simulation (or the digital twin) and the real process data (sensor measurement data) to improve the accuracy of the anomaly detection and prediction. These deviations can also be supplied to the anomaly detector as time series data or be generated by it.
- the anomaly detector can act directly with the plant operator and him in real time Assist in decisions regarding suitable countermeasures for detected anomalies .
- the anomaly detector does not have to focus on every single anomaly. Rather, it can act as a filter and only forward those anomalies that have a significant impact on operational or business goals of the plant operator have . This enables the plant operator to take immediate countermeasures to reduce the impact of the anomaly on the production process.
- the performance indicators are at least two different ones from the group of:
- the at least two different performance indicators preferably include at least the quality of the product produced and the quantity of a product produced per unit of time (throughput).
- a method for detecting anomalies in an industrial plant for executing a production process for a product comprises the following steps: a) Receiving a large number of measurement data from the sensors, b) detecting and/or predicting anomalies in the production process based on the plurality of measurement data using at least one trained artificial intelligence, c) outputting anomaly information upon detection and/or prediction of an anomaly, wherein in step b ) Anomalies can be detected and predicted simultaneously in several different performance indicators of the production process, whereby the performance performance indicators relate to the entire production process for the production of the product.
- a relevance of the anomaly in relation to at least one, in particular several different, operating or business goal(s) of the production process is determined.
- the method in particular the artificial intelligence, is configured as a function of at least one operating or business goal of the production process.
- a separate trained artificial intelligence can be used for each of the performance indicators.
- a common trained artificial intelligence is preferably used for all of the performance indicators.
- the artificial intelligence is preferably designed and trained to take into account a chronological sequence of the measurement data and/or chronological relationships between these measurement data when detecting and/or predicting the anomalies.
- the artificial intelligence is advantageously trained using normal state data from the production process.
- the artificial intelligence is advantageously at least partially trained with simulated measurement data from sensors.
- a validation of the anomaly detection and/or prediction based on deviations between the Measurement data from the sensors and simulated measurement data from the sensors take place.
- the performance indicators are at least two different ones from the group of:
- a method according to the invention for providing trained artificial intelligence for detecting anomalies in an industrial plant for carrying out a production process for a product, the plant comprising a large number of sensors for measuring process variables of the production process comprises the following steps:
- the initial training data including assignments to at least one of several different performance indicators of the production process, the performance indicators each relating to the entire production process for the production of the product,
- a first computer program (or computer program product) according to the invention comprises instructions which, when the program is run on a computer, cause the computer to carry out the method for detecting anomalies described above.
- a second computer program (or computer program product) according to the invention comprises instructions which, when the program is run on a computer, cause the computer to carry out the method described above for providing a trained artificial intelligence.
- FIG. 1 shows an exemplary basic structure of an industrial plant with a local arrangement of a device according to the invention
- FIG. 3 shows a second exemplary embodiment of an anomaly detector
- 4 shows a method sequence according to the invention for the detection of anomalies
- 5 shows an exemplary basic structure of an industrial plant with a cloud-based arrangement of a device according to the invention
- FIG. 6 shows an example of an output of anomalies to a performance indicator on a graphical user interface
- FIG. 8 shows an exemplary process sequence for providing a trained artificial intelligence for detecting anomalies
- FIG. 9 shows a first exemplary embodiment of a data pipeline in an anomaly detector according to the invention.
- FIG. 10 shows a second exemplary embodiment of a data pipeline in an anomaly detector according to the invention
- 11 shows an example of an output of anomalies for a number of performance indicators with relevance information in relation to different operating or business goals on a graphical user interface.
- production process is understood here not only as a manufacturing or manufacturing process, but also as a machining, processing or conversion process (e.g. also an energy generation process).
- Such systems 1 are used in a wide variety of industrial sectors, for example in the process industry (e.g. paper, chemicals, pharmaceuticals, metal, oil and gas), the discrete manufacturing industry and in power generation.
- the automatic Automation system 2 includes, for example, a number of industrial controllers 4 , an automation server 5 and an engineering server 8 .
- Each of the controllers 4 controls the operation of a respective sub-area of the process 3 depending on its operating states.
- the process 3 includes actuators 6 that can be controlled by the controllers. This can involve individual actuators (e.g. a motor, a pump, a valve, a switch), groups of such actuators or entire sections of a system.
- the process also includes sensors 7 that make measured values of process variables (eg temperatures, pressures, fill levels, flow rates) available to the controllers 4 .
- a communication network of the system 1 comprises, at a higher level, a system network 11 via which the servers 5 , 8 are in communication with a human-machine interface (HMI) 10 , and a control network 12 via which the controls 4 are in communication with each other and with the servers 5 , 8 .
- the controls 4 can be connected to the actuators 6 and sensors 7 via discrete signal lines 13 or via a fieldbus.
- the human-machine interface (HMI) 10 is usually designed as an operating and monitoring station and is arranged in a control room of the system 1 .
- the automation server 5 can be, for example, a so-called “operator system server” or “application server” (application server) in which one or more system-specific application programs are stored and during operation of the system 1 to be carried out. These are used, for example, to configure the controls 4 in the system 1, to record and execute operator activities on the human-machine interface (HMI) 10 (e.g. to set or change setpoints of process variables) or to send messages for to generate plant personnel and to be displayed on the human-machine interface (HMI) 10 .
- HMI human-machine interface
- HMI human-machine interface
- the automation system 2 without the field devices is often also referred to as a "process control system".
- the sensors 7 therefore provide a large number of measurement data M of process variables of the process 3 .
- This measurement data M is stored on a process data archive server 14 together with messages from the automation server 5 and with additional information (e.g. batch data, status information of intelligent field devices).
- a device 15 for detecting anomalies in the system 1 .
- This includes an anomaly detector 16 with a trained artificial intelligence 18 which is designed and trained to detect and predict anomalies in the production process 3 based on a large number of measurement data M. If an anomaly is detected and/or predicted, anomaly information A is output on a (preferably graphical) user interface 17 .
- the anomaly information A preferably also includes information about a probability of the presence of an anomaly and information about a period of time in which the anomaly occurred.
- the anomaly detector 16 is designed to detect and predict anomalies in several different performance indicators of the production process 3 , the performance indicators each relating to the entire production process 3 .
- performance indicators are: - quality of a manufactured product, - Amount of a product produced per unit of time (throughput),
- the device 15 is connected to the system network 11 and uses it to access current and historical measurement data M (and possibly additional data from the automation system 2 ), which are provided by the process data archive server 14 and the server 5 .
- the automation system also includes a simulator (or digital twin) 9 for the production process 3 .
- the simulator 9 simulates the production process 3 for example on the basis of physical and/or chemical models.
- the simulator 9 is also connected to the plant network 11 .
- the device 15 can use this to access simulation data S of the simulator 9 or received from this .
- the anomaly detector 16 can each have a separate anomaly detector unit 16a-16d for each different performance indicator La-Ld.
- Each of the anomaly detector units 16a-16d each includes a separate trained artificial intelligence 18a-18d, which is designed and trained to detect and predict the anomalies in the production process 3 for the respective performance indicator La-Ld.
- the device 15 is designed to offer the system personnel, in particular their management level, the different performance indicators La - Ld for selection via the user interface 17 and a subsequent input of selection information W via a selection of one of the performance indicators katoren La - Ld, here for example the performance indicator La to capture.
- the device 15 then activates the anomaly detector unit 16a-16d assigned to the selected performance indicator La-Ld, here the anomaly detector unit 16a, depending on the detected selection information W, and brings the performance indicator L and the anomaly information A on the user interface 17 for display.
- the anomaly detector units 16a-16d are also designed to independently select and process the measurement data M required for the respective operation from the large number of historical and current measurement data M.
- the anomaly detector 16 or its trained artificial intelligence 18 can also--as shown in FIG. 3--be designed to simultaneously detect and predict anomalies in the several different performance indicators La-Ld. Depending on the selection information W, however, only the respectively selected performance indicator, here for example the performance indicator La, with the associated anomaly information A is displayed on the user interface 17 .
- the device 15 is also designed to offer the plant personnel, in particular their management level, different operating or business goals Z for selection via the user interface 17 and subsequent input of selection information Y about a selection of one or more of the operating or business goals Z to capture .
- the device 15 includes a configuration device 19 which is designed to use the anomaly detector 16, in particular the artificial intelligence 18 or 18a - 18d, depending on at least one operating or business goal of the production process to configure.
- the anomaly detector 16 can then be better adapted to the respective task. be matched .
- a specially adapted selection of measurement data or a special combination of data analysis algorithms can be set in this way, which ensures a particularly good quality of anomaly detection for the various performance indicators with high relevance for a selected operating or business goal.
- FIG. 4 shows a simplified representation of a method sequence 20 according to the invention in the anomaly detector 16 from FIG.
- a first step 21 this receives a large number of current and historical measurement data M from the sensors 7.
- anomalies in the production process 3 based on the multitude of measurement data M are detected and predicted using the artificial intelligence 18.
- the anomalies can be detected and predicted simultaneously for several different performance indicators La-Ld of the production process 3, with the performance indicators La-Ld each relating to the entire production process 3 for a product, d. H . a (end) product resulting from the production process.
- anomaly information A is output on the user interface 17 when an anomaly is detected and/or predicted.
- the current and historical measurement data M are preferably available as time series of measurement data, and when the anomalies are detected and predicted by the artificial intelligence 18, a chronological sequence of the measurement data (current and historical measurement data) or temporal relationships between these measurement data are taken into account. In other words, a time series analysis is carried out by the artificial intelligence. Changes in the time series of measurement data can thus be recognized and future time series of performance indicators can be predicted.
- the device 15 for detecting anomalies is installed directly on site in the system 1 ("on-premise").
- the Device 15 for detecting anomalies can also be installed remotely from the system in a distributed computer system (“cloud”).
- FIG. 5 shows the plant 1 already shown in FIG. 1 with the automation system 2 and the production process 3 .
- the device 15 for anomaly detection is installed in a cloud 30 .
- the device receives the measurement data M and, if necessary. Simulation data S of the simulator 9 via a connection server 31 of the automation system 2 and a public communication network 36, such as. B. the Internet .
- the performance indicators and the anomaly information generated are in turn output to personnel of the facility 1 via a user interface 37 which is located in the facility 1 .
- the information to be output on the user interface 37 is also received by the device 15 via the communication network 36 .
- Firewalls 32 , 33 can also be arranged between the connection server 31 and the device 15 and between the device 15 and the user interface 37 .
- the simulator 9 can also be installed in the cloud 30 instead of in the system 1 .
- FIG. 6 shows an example of a performance indicator and associated anomaly information, such as can be displayed on the graphical user interfaces 17, 37.
- Time ranges in which there is only a relatively low probability of an anomaly are denoted by 42 .
- 43 designates a time range in which, with a higher probability, an anomaly is present and therefore a warning message was generated.
- 44 designates a time range in which there is a comparatively high probability of an anomaly. An alarm message was therefore generated.
- measurement data from some sensors are also output in area 41, such as pressures (Pressure 1, Pressure 2, Pressure 3), electrical currents (Current 1, Current 2) and temperatures (Temperature 1, Temperature 2).
- the latest alarm messages (“Alarm (level 66)”) and warning messages (“Warning (level 66)”) of all monitored production processes or a selected production process (here process A2) are output in an area 45 .
- the monitored performance indicator here the quality or "Quality”
- the effect of the anomaly here a quality reduction or “Quality Reduction”
- a cause here "Quality reduction due to change in immediate product ”
- the production process with the most recently detected or predicted anomalies is displayed at the top of a list of production processes (here process A2, process Al, process Bl) together with information about the last anomaly, the number of anomalies in the last 24 h and the total number of anomalies.
- FIG. 7 shows, by way of example, an output of detailed information on the time profile of the probability AL for an anomaly in the selected performance indicator.
- time curves of measurement data are output in an area 51 over time, for which a legend is output in an area 52 .
- a classification (“rating”) of the alarm, a classification or effect in relation to the performance indicator (“classification”) and a cause analysis (“note”) are output in an area 53 .
- FIG. 8 shows a simplified representation of a method sequence 60 according to the invention for providing a trained artificial intelligence for detecting anomalies in the system 1 of FIG. 1 or FIG. 5.
- a first step 61 input training data representing historical measurement data M from the sensors 7 is received. These are preferably time series of measurement data.
- initial training data which represent anomalies in the historical measurement data M, is received, the initial training data including assignments to at least one of a number of different performance indicators La-Ld of the production process 3.
- the performance indicators La-Ld each relate to the entire production process 3 .
- the artificial intelligence is trained on the basis of the input training data and the output training data in such a way that these anomalies are detected and predicted simultaneously in the case of several different performance indicators in the production process 3 .
- the artificial intelligence can be trained by an algorithm from the field of machine learning or with the help of rule-based or mathematical (e.g. statistical) processes.
- Particularly advantageous algorithms or methods are neural networks, autoencoders, Gaussian mixture models, boosted Gaussian mixture ensembles and isolation forests, and combinations of all of these methods.
- Backpropagation with gradient-based optimization can be used.
- the artificial intelligence is preferably designed and trained to use a chronological sequence of the measurement data (current and historical measurement data) or temporal relationships between the measurement data must be taken into account.
- a time series analysis is carried out by the artificial intelligence. Changes in the time series of measurement data can thus be recognized and future time series of performance indicators can be predicted. For example, correlations between all current and historical measured values of all sensors are determined.
- the artificial intelligence is preferably based on a normal state of the production process or of performance indicators trained .
- “poor states” are removed from the training data.
- the artificial intelligence thus searches for abnormalities in the performance indicators that differ from the trained normal state.
- a fourth step 64 the trained artificial intelligence is provided.
- FIG. 9 shows an exemplary data pipeline PI in a simplified representation for the anomaly detector 16 from FIG. H .
- various data analysis and/or processing elements arranged one behind the other, with a subsequent data analysis and/or processing element processing the results of at least one previous data analysis and/or processing element.
- Optional components and data flows are shown with dashed lines.
- the anomaly detector 16 or the data pipeline PI has a multi-stage structure and comprises three stages connected in series: detection (D), classification (K) and post-processing (Postprocessing) N, each receiving measurement data M of the process variables (current and historical) in the form of time series.
- normal values E estimated from the received measurement data M i.e. "normal” values without the presence of an anomaly
- performance indicators e.g. quality, throughput
- times T of deviations of the performance indicators from normal values are determined and output with the aid of artificial intelligence
- a search is made for abnormalities in the measurement data or in their correlations that differ from the trained normal state.
- the artificial intelligence preferably works with the help of unsupervised learning (but supervised and semi-supervised learning is also possible).
- an assignment (classification) to known, trained anomaly types is carried out with the aid of artificial intelligence from the received measurement data M and the received estimated normal values E of the performance indicators (e.g. quality, throughput) and durations T of deviations from the normal values.
- Anomaly information A is generated, which includes the affected performance indicator L (eg quality, throughput), the type AT of the anomaly (eg quality reduction, throughput reduction), a possible reason C for the anomaly and a probability AL for the presence of the anomaly.
- the artificial intelligence at level K can work with supervised or rule-based methods. Neural networks are preferably used.
- an anomaly relevance AR i.e. a relevance of a determined anomaly
- an anomaly relevance AR is then determined in relation to one or more operating or business goals Z of the installation 1 .
- the post-processing stage N receives the normal values E of the performance indicators L determined by stage D (eg quality, throughput) and times T of deviations of the performance indicators L from the normal values as input variables. Furthermore, stage N receives the anomaly information A generated by stage K as an input variable.
- level N there is information about current operating or business goals Z of the system 1 in level N.
- This information can already be permanently stored in level N, or—as explained in connection with FIGS. 1-3—can be recorded via the user interface 17 or received externally in some other way.
- the information can, for example, be in the form of weightings for the different performance indicators L in relation to different operating or business goals Z of the system.
- corresponding point values are determined by level N for the importance of identified anomalies in relation to the achievement of the operational or business goals Z and as the relevance AR of the respective anomaly for the or the operational or Business goals Z issued .
- the operator of the system can thus see what relevance a determined anomaly has in relation to each of his operating or business goals Z, and he can initiate suitable measures according to a prioritization of his operating or business goals Z. For example, he can Immediately initiate measures to eliminate anomalies relevant to high-priority operational or business goals Z and delay countermeasures to eliminate anomalies relevant to low-priority operational or business goals Z for the time being.
- Stage N particularly advantageously implements post-processing in order to improve the quality of the anomaly detection.
- post-processing for example, anomalies that were determined using other methods (e.g. characteristic map methods or rule-based methods) or originate from a classic model (hybrid) can be used. This means that anomalies are validated with anomalies determined in other ways. There is also the option of using the measurement data to increase the quality of the anomaly detection. So e.g. B. a specific behavior of the measured values can be used to validate the anomalies.
- the post-processing can be done by any form of filtering, if necessary also with an AI, but also e.g. B. be done by rules regarding the measurands .
- the classification stage K is first followed by an evaluation stage B and then a post-processing stage N'.
- an anomaly relevance AR ie a relevance of an identified anomaly is determined in relation to one or more operating or business goals Z of the installation 1 .
- the evaluation level B receives as input variables the normal values E of the performance indicators L output by the level D as well as times T of deviations of the performance indicators L from the normal values. Stage B also receives the anomaly information A generated by stage K as an input variable. Level B also contains information about the current operating or business goals Z of the system 1 . This information can already be permanently stored in level B, or it can be received from level B via the user interface 17, for example. This information can be present, for example, in the form of weightings for the different performance indicators in relation to different operating or business goals Z of the plant.
- corresponding point values for the importance of identified anomalies in relation to the achievement of the operating or business goals Z are determined by level B and output as the relevance AR of the anomaly for the operating or business goals Z.
- the post-processing stage N' serves to filter the information determined in the previous stages. For example, irrelevant warning messages or only slightly probable anomaly information, e.g. B. taking into account the history of the measurement data or the performance indicators. This is also done advantageously depending on the operating or business goals Z.
- anomaly information e.g. relevance, time range T, probable cause
- warning messages or alarms on a display, for example user interface 17 in FIGS. 1-3 or user interface 37 in FIG Anomalies with sufficiently high relevance for his ( e ) operational or business objective ( s ) are displayed.
- the information about current operational or business goals Z of Annex 1 is also available in stages D and K.
- this information can be used to configure the data pipeline P2 or of the individual levels D, K, B, N 'take place, z. B. for a selection and combination of particularly suitable methods or models for data analysis or for a selection (or exclusion) of data for further analysis. For example, if it is known that a bad condition in the training data does not impact the KPI for an operational or business objective, it can remain in the training data and does not need to be removed.
- stage N' implements post-processing in order to improve the quality of the anomaly detection.
- anomalies that were determined using other methods (e.g. characteristic map methods or rule-based methods) or originate from a classic model (hybrid) can be used. This means that anomalies are validated with anomalies determined in other ways. A specific behavior of the measured values can also be used to validate the anomalies.
- simulated sensor data S generated by the simulator (or digital twin) 9 of the production process 3 can also be fed to the stages D, K, B and N′.
- the artificial intelligence in the data pipeline P2 can then be continuously improved, in particular in the case of a scale-up of a production from a laboratory scale to a production plant that is significantly larger in scale.
- a digital triplet (“digital triplet") focusing on anomalies is thus created from anomaly detector 16, simulator (digital twin) 9 and process data (sensor measurement data M), but not only from a technical perspective, but also from a perspective of the operational or business objectives of Appendix 1 .
- the anomaly detector 16 can use discrepancies between the results of the simulator (or digital twin) 9 and the real process data (sensor measurement data M) to validate anomalies and thus to improve the accuracy of the anomaly detection and prediction. These deviations can also be supplied to the anomaly detector as time series data or be generated by it.
- the anomaly detector 16 acts directly with the operator of the system and assists him in real time - Decisions regarding suitable countermeasures for detected anomalies.
- the anomaly detector 16 does not focus on every single abnormality. Rather, it acts as a filter and forwards only those anomalies that have a significant impact on the plant operator's operational or business goals. This enables the plant operator to take immediate countermeasures to reduce the impact of the anomaly on the production process.
- the artificial intelligence e.g. B. a neural network
- Some data sources may contain detailed meta information, others less detailed background knowledge but more realistic correlations, e.g. B. from the real plant .
- a neural network could be used that is pre-trained in the first step with data from a Digital Twin, which allows learning some basic categories of anomalies, and that is then re-trained in a second step on the basis of historical plant data.
- the artificial intelligence can also be trained in a first step with real plant data with regard to correlations (unsupervised, semi-supervised or mainly supervised) and later improved in a second step in order to associate discovered patterns with relevant anomalies.
- the benefits of this approach do not have to be achieved by training a single algorithm, but can be split across multiple learning or rule-based systems. For example, a division into a neural network that has been trained on historical data and a rule-based system or a supervised tree learning system could take place. Which measurement data are used in which stages can result from the rules of a rule-based part of the post-processing, from a trained neural network or as the result of a simulation model.
- FIG. 11 shows an example of an output of anomalies for a number of performance indicators with relevance information in relation to different operating or business goals on a graphical user interface.
- An anomaly level AL i . H . a probability of the presence of an anomaly plotted against time t.
- the anomaly level AL takes into account anomalies in several different performance indicators. In the areas 71 and 73 there are in each case anomalies in the area of quality. In contrast, in area 72 there is an anomaly in the area of the throughput. Since the current operational or business objective places more importance on quality than throughput, an alarm 74 was generated in relation to the quality anomaly, and only a warning 75 in relation to the throughput anomaly.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Manufacturing & Machinery (AREA)
- Quality & Reliability (AREA)
- General Factory Administration (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
L'invention concerne un dispositif de détection d'anomalies dans une installation industrielle (1) pour l'exécution d'un processus de production (3) d'un produit, l'installation comprenant une pluralité de capteurs (7) destinés à la mesure de grandeurs de processus du processus de production (3), lequel dispositif comprend un détecteur d'anomalie (16) doté d'au moins une intelligence artificielle entraînée (18). Celle-ci est conçue et entraînée pour détecter et/ou prédire des anomalies dans le processus de production (3) sur la base d'une pluralité de données de mesure (M) des capteurs (7). Lors de la détection et/ou de la prédiction d'une anomalie, le détecteur d'anomalie (16) émet une information d'anomalie (A). Il est ainsi possible de détecter et prédire des anomalies pour plusieurs indicateurs de puissance différents (La-Ld) du processus de production (3) simultanément, les indicateurs de puissance (La-Ld) se rapportant respectivement à l'ensemble du processus de production (3) pour la production du produit.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
EP20199293 | 2020-09-30 | ||
PCT/EP2021/075610 WO2022069258A1 (fr) | 2020-09-30 | 2021-09-17 | Dispositif et procédé de détection d'anomalies dans une installation industrielle pour l'exécution d'un processus de production |
Publications (1)
Publication Number | Publication Date |
---|---|
EP4193227A1 true EP4193227A1 (fr) | 2023-06-14 |
Family
ID=72709122
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
EP21778101.2A Pending EP4193227A1 (fr) | 2020-09-30 | 2021-09-17 | Dispositif et procédé de détection d'anomalies dans une installation industrielle pour l'exécution d'un processus de production |
Country Status (4)
Country | Link |
---|---|
US (1) | US20230376024A1 (fr) |
EP (1) | EP4193227A1 (fr) |
CN (1) | CN116235121A (fr) |
WO (1) | WO2022069258A1 (fr) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2023169060A (ja) * | 2022-05-16 | 2023-11-29 | 横河電機株式会社 | 装置、方法およびプログラム |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5877954A (en) | 1996-05-03 | 1999-03-02 | Aspen Technology, Inc. | Hybrid linear-neural network process control |
US9110452B2 (en) | 2011-09-19 | 2015-08-18 | Fisher-Rosemount Systems, Inc. | Inferential process modeling, quality prediction and fault detection using multi-stage data segregation |
EP3098681B1 (fr) * | 2015-05-27 | 2020-08-26 | Tata Consultancy Services Limited | Gestion de la santé d'intelligence artificielle de système hôte |
CN107797537A (zh) | 2017-11-10 | 2018-03-13 | 上海第二工业大学 | 一种应用于自动化生产线的故障预测与健康管理方法 |
EP3514555B1 (fr) | 2018-01-22 | 2020-07-22 | Siemens Aktiengesellschaft | Appareil de surveillance d'un système d'actionneur, procédé permettant de fournir un appareil de surveillance d'un système d'actionneur et procédé de surveillance d'un système d'actionneur |
US11714397B2 (en) * | 2019-02-05 | 2023-08-01 | Samsung Display Co., Ltd. | System and method for generating machine learning model with trace data |
-
2021
- 2021-09-17 EP EP21778101.2A patent/EP4193227A1/fr active Pending
- 2021-09-17 CN CN202180067036.XA patent/CN116235121A/zh active Pending
- 2021-09-17 US US18/029,277 patent/US20230376024A1/en active Pending
- 2021-09-17 WO PCT/EP2021/075610 patent/WO2022069258A1/fr active Search and Examination
Also Published As
Publication number | Publication date |
---|---|
WO2022069258A1 (fr) | 2022-04-07 |
US20230376024A1 (en) | 2023-11-23 |
CN116235121A (zh) | 2023-06-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
DE102017000536B4 (de) | Zellsteuereinheit zum Feststellen einer Ursache einer Anomalie bei einer Fertigungsmaschine | |
DE60212121T2 (de) | Erzeugung von prozessverwandten daten | |
EP0893746B1 (fr) | Système et méthode de diagnostic d'un processus | |
EP3279756B1 (fr) | Dispositif de diagnostic et procédé de surveillance du fonctionnement d'une installation technique | |
EP3077878B1 (fr) | Procédé implémenté par ordinateur et système de surveillance automatique et de détermination de l'état de l'intégralité des étapes de processus dans une unité de traitement | |
DE112011101738T5 (de) | Mehrstufiges Prozessmodellierungsverfahren | |
EP3232282B1 (fr) | Dispositif de diagnostic et procede de surveillance du fonctionnement d'une installation technique | |
EP0768584B1 (fr) | Méthode de surveillance d'une machine ou installation | |
EP1543394A1 (fr) | Procede et dispositif pour surveiller une installation technique comprenant plusieurs systemes, notamment une centrale electrique | |
EP1250632A1 (fr) | Systeme et procede pour determiner l'efficacite d'unites de production, des evenements d'erreurs et la cause desdites erreurs | |
EP0789864B1 (fr) | Systeme de controle pour installation industrielle | |
DE19743600B4 (de) | Verfahren zur Überwachung eines zyklischen Produktionsprozesses | |
EP3282399B1 (fr) | Procede de reconnaissance ameliore d'anomalies de processus d'une installation technique et systeme de diagnostic correspondant | |
DE102021124253A1 (de) | Maschinelles Lernverfahren zur Anomaliedetektion in einem elektrischen System | |
EP2971768B2 (fr) | Développement d'un modèle supérieur pour contrôler et/ou surveiller un système de compresseurs | |
EP4193227A1 (fr) | Dispositif et procédé de détection d'anomalies dans une installation industrielle pour l'exécution d'un processus de production | |
DE102008042969A1 (de) | Verfahren und Vorrichtung zur Detektion von Prozesszuständen in alternierenden Produktionsprozessen | |
EP3474091B1 (fr) | Procédé et dispositif de réglage d'un processus dans un système, a savoir d'un processus de broyage dans un dispositif de broyage | |
DE102008032885A1 (de) | Verfahren und Vorrichtung zur Überprüfung und Feststellung von Zuständen eines Sensors | |
WO2022263458A1 (fr) | Surveillance de la production de panneaux de matériau, en particulier de panneaux en matériau dérivé du bois, en particulier à l'aide d'une carte auto-organisatrice | |
WO2022195050A1 (fr) | Procédé et système de prédiction du fonctionnement d'une installation technique | |
EP4060441A1 (fr) | Procédé et système de détection des anomalies dans le fonctionnement d'une installation technique | |
EP4127848A1 (fr) | Procédé et système de diagnostic de messages | |
AT502241B1 (de) | Verfahren und anordnung zur feststellung der abweichung von ermittelten werten | |
WO2005029206A1 (fr) | Procede de deduction automatique de recommandations de maintenance |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: UNKNOWN |
|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE |
|
PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE |
|
17P | Request for examination filed |
Effective date: 20230310 |
|
AK | Designated contracting states |
Kind code of ref document: A1 Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR |
|
DAV | Request for validation of the european patent (deleted) | ||
DAX | Request for extension of the european patent (deleted) |