Procedure and system for evaluating a technical plant

Abstract

A method for evaluating a technical system comprising at least one electric motor is proposed. During an operational phase, an operational data set is recorded, comprising a plurality of measured variables, each of which comprises a plurality of measured values. The measured values ​​of each variable are recorded sequentially. Several temporally successive segments are extracted from the operational data set. Each segment of the operational data set is assigned to a class by a classifier. Each segment of the operational data set is then examined using at least one previously trained anomaly model. A warning message is issued if an anomaly is detected in any of the segments of the operational data set during the examination. A system according to the invention for evaluating a technical system comprises at least one anomaly model and one classifier.The system is set up to carry out the method according to the invention.

Description

[0001] The invention relates to a method for evaluating a technical system comprising at least one electric motor. The invention also relates to a system for evaluating a technical system, which is configured to carry out the method according to the invention.

[0002] Technical systems, such as drive systems, comprise a multitude of technical components, including an electric motor, a converter to generate three-phase alternating current for the electric motor, and a gearbox to reduce the motor's speed. Examples of such technical systems include rotary tables, conveyor belts, and stacker cranes. After extended operation, the components of these systems can malfunction due to wear and tear.

[0003] It is known to monitor such technical systems by recording and evaluating specific measurements at set times. If the recorded measurements deviate too significantly from predefined target values, a fault in the system is assumed, and a corresponding message is sent to the operator. For example, a current generated by the inverter that exceeds a defined limit could indicate stiffness in the gearbox due to wear.

[0004] From DE 11 2022 006 638 T5, a method for diagnosing anomalies in a plant is known. In this method, data representing a state variable of an evaluation element are obtained from the plant, the data are classified for each operating state of the plant, and a multitude of parameter values ​​are calculated to configure a trained model.

[0005] From DE 10 2023 207 829 A1, a method for detecting anomalies in a set of signals is known, wherein the signal represents a quantity of a technical system to be checked. A model is provided for mapping signals from a past time window to an output target, where the output target maps the signals in a future time step as future signals.

[0006] From EP 4 160 229 A1, a method for monitoring a machine is known. In a training phase, training data with state variables of several operating points of the machine are provided; operating points of the training data are grouped into operating point clusters by means of clustering; a classifier, which assigns operating points to the recognized operating point clusters, is trained; and an anomaly detection model for anomaly detection is trained.

[0007] The invention is based on the objective of further developing a method and a system for evaluating a technical plant, and in particular of detecting anomalies in the operation of the technical plant.

[0008] The problem is solved by a method for evaluating a technical system with the features specified in claim 1. Advantageous embodiments and further developments are the subject of the dependent claims. The problem is also solved by a system for evaluating a technical system with the features specified in claim 11.

[0009] A method for evaluating a technical system comprising at least one electric motor is proposed. During an operational phase, an operational data set is recorded, containing multiple measured variables. Each measured variable comprises multiple measured values, and these values ​​are recorded sequentially. Several temporally sequential segments are extracted from the operational data set. Each segment is assigned a class by a classifier. Each segment is then analyzed using at least one previously trained anomaly model. A warning message is issued if an anomaly is detected in any of the segments during the analysis.

[0010] The technical system comprises an application such as a storage and retrieval machine, a shuttle with or without a satellite, a roller conveyor, a belt conveyor, a toothed belt conveyor, a chain conveyor, a link belt conveyor, a scissor roller conveyor, a corner transfer unit, a rotary table, a telescopic belt conveyor, a swivel wheel conveyor, a lifting station, a pallet conveyor shuttle, an electric monorail, a rail-guided vehicle, a gantry, a scissor lift table, a pump, a blower, a fan, a transverse transfer carriage, a push platform, a 3-axis gantry, an automated guided vehicle (AGV), a hoist, an electric floor conveyor, or a spiral conveyor. The application includes, for example, a drive system comprising an electric motor, an inverter for generating three-phase alternating current for the electric motor, and a gearbox for reducing the speed of the electric motor.

[0011] Each class is assigned a driving segment of an application cycle. Possible application cycles include, for example, forward driving, reverse driving, engine operation, generator operation, transporting a light load, and transporting a heavy load. Possible driving segments of the application cycles include, for example, driving at a constant speed, accelerating, and braking.

[0012] If a large number of different application cycles occur during the operation of the application, with each individual application cycle occurring relatively infrequently, the recorded operational data records do not provide a sufficient data basis. Evaluating complete application cycles is not productive in such a case. The extracted segments can originate from various application cycles and therefore occur more frequently. The segments of the operational data records provide a sufficient data basis. The method according to the invention thus allows for the evaluation of the technical system whose applications exhibit a large number of different application cycles.

[0013] According to the invention, the measured values ​​of the operating data set, which lie in a time interval in which one of the measured quantities has an at least approximately constant course or an at least approximately linear course, are recognized as belonging to a segment.

[0014] A constant curve corresponds, for example, to driving at a constant speed. A linearly increasing curve corresponds, for example, to an acceleration process. A linearly decreasing curve corresponds, for example, to a braking process.

[0015] According to the invention, a second derivative of one of the measured quantities with respect to time is calculated for the operational data set. In this second derivative, the time points of local extrema are determined. The measured values ​​of the operational data set that were recorded between two adjacent local extrema are recognized as belonging to the same segment.

[0016] The second derivative of the measured quantity with respect to time exhibits a significant extremum when there is a change between a constant and a linear trend. An extremum can be either a maximum or a minimum.

[0017] According to an advantageous embodiment of the invention, the measured quantity, of which the second derivative with respect to time is calculated, is a rotational speed of the electric motor.

[0018] The rotational speed of the electric motor, which is measured indirectly via the inverter by measuring the frequency of an output current to the electric motor, is particularly suitable for extracting segments with at least an approximately constant or at least an approximately linear profile.

[0019] According to an advantageous embodiment of the invention, a further operating data record is recorded each time a defined trigger condition is met and when a defined period of time has elapsed since the recording of the previous operating data record.

[0020] One such trigger condition is, for example, the start of an application cycle of a technical system application when, for instance, the speed of an electric motor in the system exceeds a predefined minimum speed. This ensures that the recorded operating data record originates from an application cycle.

[0021] According to an advantageous embodiment of the invention, after extracting the segments from the operational data set, several temporally successive segments are combined to form a segment sequence. A classifier assigns the segment sequence of the operational data set to a class. The segment sequence of the operational data set is then examined using at least one previously trained anomaly model. A warning message is issued if an anomaly is detected in the segment sequence during the examination of the operational data set.

[0022] This also allows for the additional investigation of segment sequences, especially repeatedly occurring segment sequences.

[0023] According to an advantageous embodiment of the invention, several reference datasets are recorded during a preliminary reference phase. Each reference dataset comprises a plurality of measured variables, with each measured variable comprising a plurality of measured values, and the measured values ​​of each measured variable are recorded sequentially. Several temporally successive segments are extracted from each of the reference datasets. Segments of the reference datasets that exhibit a relatively high degree of similarity to one another are identified as belonging to a common cluster. The segments belonging to a common cluster are assigned to a common class. The classifier is trained with the segments of the reference datasets of the previously defined classes in order to assign segments of future recorded operational datasets to the previously defined classes.The at least one anomaly model is trained with the segments of the reference datasets in order to examine segments of future recorded operational datasets.

[0024] Each class is assigned a driving segment of an application cycle. Possible application cycles include, for example, forward driving, reverse driving, engine operation, generator operation, transporting a light load, and transporting a heavy load. Possible driving segments of the application cycles include, for example, driving at a constant speed, accelerating, and braking.

[0025] According to an advantageous embodiment of the invention, the measured values ​​of a reference data set which lie in a time interval in which one of the measured quantities has an at least approximately constant course or an at least approximately linear course are recognized as belonging to a segment.

[0026] A constant curve corresponds, for example, to driving at a constant speed. A linearly increasing curve corresponds, for example, to an acceleration process. A linearly decreasing curve corresponds, for example, to a braking process.

[0027] According to an advantageous embodiment of the invention, a second derivative of one of the measured quantities with respect to time is calculated for each reference data set. In the second derivative of said measured quantity, the time points of local extrema are determined. The measured values ​​of the reference data set that were recorded between two adjacent local extrema are recognized as belonging to the same segment.

[0028] The second derivative of the measured quantity with respect to time exhibits a significant extremum when there is a change between a constant and a linear trend. An extremum can be either a maximum or a minimum.

[0029] According to an advantageous embodiment of the invention, the measured quantity, of which the second derivative with respect to time is calculated, is a rotational speed of the electric motor.

[0030] According to an advantageous embodiment of the invention, segments of reference data sets which do not exhibit a sufficiently high degree of similarity to any of the other segments of reference data sets are discarded.

[0031] Segments of reference datasets that show no similarity to segments of other reference datasets originate from unusual application cycles or malfunctions.

[0032] According to an advantageous embodiment of the invention, after extracting the segments from the reference datasets, several temporally successive segments are combined to form a segment sequence. Segment sequences from reference datasets that exhibit a relatively high degree of similarity to one another are identified as belonging to a common cluster. The segment sequences belonging to a common cluster are assigned to a common class. The classifier is trained with the segment sequences of the reference datasets of the previously defined classes in order to assign segment sequences from future recorded operational datasets to the previously defined classes. The at least one anomaly model is trained with the segment sequences of the reference datasets in order to analyze segment sequences from future recorded operational datasets.

[0033] This also allows for the additional investigation of segment sequences, especially repeatedly occurring segment sequences.

[0034] A system according to the invention for evaluating a technical plant comprises at least one anomaly model and a classifier. The system is configured to carry out the method according to the invention.

[0035] The system according to the invention thus allows an evaluation of the technical system whose applications have a large number of different application cycles.

[0036] The system according to the invention comprises, for example, a single anomaly model for investigating all different segments. The system according to the invention also comprises, for example, several anomaly models, each of which is provided for investigating a specific segment or several different segments.

[0037] According to an advantageous embodiment of the invention, the classifier is designed as a neural network.

[0038] Examples of such neural networks include a Convolutional Neural Network or a Recurrent Neural Network.

[0039] Alternatively, the classifier can be designed as a Support Vector Machine, K-Nearest Neighbor, Random Forest Classifier, Gradient Boosting (XGBoost or LightGBM), Naive Bayes, Decision Tree, Logistic Regression, or Ensemble methods (such as Bagging or Stacking). It is also conceivable that the classifier is rule-based.

[0040] According to an advantageous embodiment of the invention, the at least one anomaly model is designed as a neural network.

[0041] One such neural network is, for example, an autoencoder or a variational autoencoder, which directly provide an anomaly assessment.

[0042] Alternatively, anomaly monitoring can be performed using One-Class SVM, Isolation Forest, or statistics-based distance methods, such as Mahanalobis distance. The features on which these models are trained can optionally be applied using rule-based feature reduction, feature reduction via PCA, or to the features of the latent space of an autoencoder. A combination of feature reduction methods is possible. In the simplest case, anomaly detection can also be performed as threshold monitoring.

[0043] The invention is not limited to the combination of features stated in the claims. For a person skilled in the art, further meaningful combinations of claims and / or individual claim features and / or features of the description and / or the figures will become apparent, in particular from the problem statement and / or the problem arising from a comparison with the prior art.

[0044] The invention will now be explained in more detail with reference to the illustrations. The invention is not limited to the embodiments shown in the illustrations. The illustrations only depict the subject matter of the invention schematically. They show: Fig. 1: A flowchart of the recording of a reference dataset during a reference phase, Fig. 2: A flowchart of the processing of recorded reference datasets during the reference phase, Fig. 3: a flowchart of the processing of recorded operational data records during an operational phase and Fig. 4: A diagram of a measured quantity from exemplary data sets.

[0045] Fig. Figure 1 shows a flowchart of the recording of a reference data set 10 in a technical plant during a reference phase. The reference phase is carried out as a preparatory step.

[0046] The technical system in question is a drive system. The drive system comprises an electric motor, a converter for generating a three-phase alternating voltage for the electric motor, and a gearbox for reducing the speed of the electric motor.

[0047] During the reference phase, a reference data set 10 is recorded in step 101. The recorded reference data set 10 comprises a number of measured variables. These measured variables include, in particular, the rotational speed and torque of the electric motor. These measured variables, rotational speed and torque, are, for example, measured indirectly via the inverter by measuring the frequency and current of the output current to the electric motor.

[0048] Each measurand comprises a plurality of measured values, for example, 2048 measured values. The measured values ​​of each measurand are recorded sequentially and synchronously. For example, the measured values ​​of each measurand are recorded at equidistant time intervals of 5 ms each. The recorded reference dataset 10 thus comprises, for example, two different measurands, each with 2048 measured values.

[0049] The recorded reference data set 10 is stored in a data storage device 20 in step 102.

[0050] During the reference phase, step 101 is repeated and several more reference data records 10 are recorded. The second step 102 is also repeated and the recorded reference data records 10 are stored in the data storage 20.

[0051] Another reference data set 10 is recorded when a defined trigger condition is met. Such a trigger condition is, for example, the start of an application cycle of the technical system, such as when the rotational speed of an electric motor in the system exceeds a predefined minimum speed. Another trigger condition is when the recorded measured values ​​of the parameters in reference data set 10 remain at least approximately constant for a period of, for example, 10 seconds.

[0052] Fig. Figure 2 shows a flowchart of the processing of recorded reference data sets 10 during the reference phase. The processing of the recorded reference data sets 10 according to Fig. Step 2 occurs only once. In step 103, the previously recorded and stored reference data records 10 are loaded from the data storage 20.

[0053] In step 104, several temporally sequential segments S1, S2, S3, S4 are extracted from each of the reference data sets 10. For this purpose, the trend of a measured quantity is examined. The measured quantity being examined is, for example, the rotational speed of the electric motor.

[0054] The measured values ​​of a reference data set 10, which lie in a time interval in which one of the measured quantities, for example the rotational speed of the electric motor, has an at least approximately constant or at least approximately linear course, are recognized as belonging to a segment S1, S2, S3, S4.

[0055] For example, for each reference data set 10, a second derivative of one of the measured quantities with respect to time is calculated. The measured quantity from which the second derivative with respect to time is calculated is, for example, the rotational speed of the electric motor.

[0056] In the second derivative of the measured quantity, for example, the rotational speed of the electric motor, the times of local extrema are determined. The measured values ​​of the reference data set 10, which were recorded between two adjacent local extrema, are recognized as belonging to a segment S1, S2, S3, S4.

[0057] Segments S1, S2, S3, and S4 of reference datasets 10, which exhibit a relatively high degree of similarity to each other, belong to a common cluster and are recognized as belonging to a common cluster. In step 104, reference datasets 10 that belong to a common cluster and have been recognized as belonging to a common cluster are assigned to a common class.

[0058] The clusters are automatically detected using a clustering method, such as DBSCAN. Alternatively, the clusters are detected by manual assignment. Segments S1, S2, S3, and S4 of reference datasets 10 that do not exhibit a sufficiently high degree of similarity to any of the other segments S1, S2, S3, and S4 of reference datasets 10 are discarded.

[0059] In step 105, a classifier is trained using segments S1, S2, S3, and S4 from the reference datasets 10 of the previously defined classes. The classifier is trained using a machine learning method to assign segments S1, S2, S3, and S4 from future recorded operational datasets 12 to the previously defined classes. The goal is an error-free assignment of segments S1, S2, S3, and S4 from recorded operational datasets 12 to the previously defined classes.

[0060] In step 106, the classifier and the classes assigned to segments S1, S2, S3, S4 of the reference data records 10 are stored in the data store 20.

[0061] In step 107, an anomaly model is trained using segments S1, S2, S3, S4 of the reference datasets 10. The anomaly model is then trained to examine segments S1, S2, S3, S4 of future operational datasets 12.

[0062] Alternatively, in step 107, several segment-specific anomaly models are trained using segments S1, S2, S3, and S4 of the reference datasets 10. For example, each anomaly model is trained using exactly one of the segments S1, S2, S3, and S4 of the reference datasets 10. Thus, each segment S1, S2, S3, and S4 is assigned exactly one segment-specific anomaly model. The anomaly models are trained to examine segments S1, S2, S3, and S4 of future operational datasets 12.

[0063] In step 108, the anomaly model is also stored in data storage 20. Alternatively, in step 108, multiple anomaly models are also stored in data storage 20.

[0064] Fig. Figure 3 shows a flowchart of the processing of recorded operational data records 12 during an operational phase. The evaluation of the technical system is carried out during the operational phase.

[0065] During the operational phase, an operational data record 12 is recorded. The recorded operational data record 12 is stored in the data storage 20 in a first step 121.

[0066] The recorded operating data set 12 comprises a number of measured variables. These measured variables include, in particular, the rotational speed and torque of the electric motor. These measured variables, rotational speed and torque, are, for example, measured indirectly via the inverter by measuring the frequency and current of the output current to the electric motor.

[0067] Each measured quantity comprises a plurality of measured values, for example, 2048 measured values. The measured values ​​of each measured quantity are recorded sequentially and synchronously. For example, the measured values ​​of each measured quantity are recorded at equidistant time intervals of 5 ms each. The recorded operational data set 12 thus comprises, for example, two different measured quantities, each with 2048 measured values.

[0068] In step 122, several temporally successive segments S1, S2, S3, S4 are extracted from the operating data set 12. For this purpose, the course of a measured variable is examined. The measured variable being examined is, for example, the rotational speed of the electric motor.

[0069] The measured values ​​of the operating data set 10, which lie in a time interval in which one of the measured variables, for example the speed of the electric motor, has an at least approximately constant or at least approximately linear course, are recognized as belonging to a segment S1, S2, S3, S4.

[0070] For example, a second derivative of one of the measured variables with respect to time is calculated for operating data set 10. The measured variable in question, from which the second derivative with respect to time is calculated, is, for example, the rotational speed of the electric motor.

[0071] In the second derivative of the measured quantity, for example, the rotational speed of the electric motor, the times of local extrema are determined. The measured values ​​of the operating data set 12, which were recorded between two adjacent local extrema, are recognized as belonging to a segment S1, S2, S3, S4.

[0072] In step 123, each segment S1, S2, S3, S4 of the operational data record 12 is fed to the classifier. The classifier then assigns each segment S1, S2, S3, S4 of the operational data record 12 to one of the previously defined classes. Each segment S1, S2, S3, S4 of the operational data record 12 is assigned to the class to which it has the greatest similarity.

[0073] In step 124, each segment S1, S2, S3, S4 of the operational dataset 12 is examined using the previously trained anomaly model. Alternatively, in step 124, each segment S1, S2, S3, S4 of the operational dataset 12 is examined using one of the previously trained segment-specific anomaly models.

[0074] In step 125, a warning message is issued if an anomaly is detected in segment S1, S2, S3, S4 during the examination of one of the segments S1, S2, S3, S4 of the operational data set 12.

[0075] Steps 121 to 125 are executed repeatedly. Specifically, these steps are repeated when a defined trigger condition is met and when a defined period of time has elapsed since the previous operational data record was recorded.

[0076] One such trigger condition is, for example, the start of an application cycle of the technical system when, for example, the speed of an electric motor of the technical system exceeds a predetermined minimum speed.

[0077] The time period is defined, for example, as six hours. Therefore, operational data records 12 are only recorded and processed four times a day.

[0078] Fig. Figure 4 shows a diagram of a measured quantity from exemplary data sets, namely an exemplary reference data set 10 and an exemplary operating data set 12. The measured quantity, which is plotted on the ordinate of each diagram, is the rotational speed of an electric motor. Time is plotted on the abscissa of each diagram.

[0079] The reference data set 10 comprises, in chronological order, a first segment S1, a second segment S2, a third segment S3, a fourth segment S4, another first segment S1, another second segment S2, another third segment S3, another fourth segment S4 and another first segment S1.

[0080] In the first segment S1, the speed of the electric motor is zero; the electric motor does not rotate. Therefore, in the first segment S1, the speed of the electric motor has a constant profile.

[0081] In the second segment S2, the speed of the electric motor increases uniformly; the electric motor accelerates uniformly. Therefore, in the second segment S2, the speed of the electric motor has a linear profile.

[0082] In the third segment S3, the speed of the electric motor is constant; the electric motor rotates at a constant speed. Therefore, in the third segment S3, the speed of the electric motor has a constant profile.

[0083] In the fourth segment S4, the speed of the electric motor decreases uniformly; the electric motor is braked uniformly. Therefore, in the fourth segment S4, the speed of the electric motor has a linear profile.

[0084] The operational data set 12 comprises, in chronological order, a first segment S1, a second segment S2, a third segment S3, a fourth segment S4 and another first segment S1. Reference symbol list 10 Reference data set 12 Operational data set 20 data storage devices 101-108 steps during the reference phase 121-125 steps during the operational phase S1..S4 segments

Claims

[1] Method for evaluating a technical installation comprising at least one electric motor, wherein During an operational phase, an operational data record (12) is recorded, which includes a plurality of measured quantities, and wherein Each measurement quantity comprises a plurality of measured values, and wherein the measured values ​​of each quantity are recorded sequentially over time; and wherein several temporally successive segments (S1, S2, S3, S4) are extracted from the operational data set (12); and wherein each segment (S1, S2, S3, S4) of the operational data set (12) is assigned to a class by a classifier; and wherein each segment (S1, S2, S3, S4) of the operational data set (12) is examined using at least one previously trained anomaly model; and wherein a warning message is issued if an anomaly is detected in the segment (S1, S2, S3, S4) of the operational data set (12) during the examination of one of the segments (S1, S2, S3, S4), characterized by , that the measured values ​​of the operational data set (12), which lie within a time interval, in which one of the measured variables exhibits an at least approximately constant or at least approximately linear course, is recognized as belonging to a segment (S1, S2, S3, S4), and that for the operational data set (12) a second derivative of one of the measured quantities with respect to time is calculated; and The second derivative of the aforementioned measured quantity determines the times of local extrema; and the measured values ​​of the operational data set (12), which were recorded between two adjacent local extrema, are recognized as belonging to one segment (S1, S2, S3, S4). [2] Method according to claim 1, characterized by that the measured quantity, from which the second derivative with respect to time is calculated, is a rotational speed of the electric motor. [3] Method according to any of the preceding claims, characterized by , that each additional operational data record (12) is recorded, when a defined trigger condition is met, and when a defined period of time has passed since the recording of the previous operational data record (12). [4] Method according to any of the preceding claims, characterized by , that after extracting the segments (S1, S2, S3, S4) from the operational data set (12), several temporally successive segments (S1, S2, S3, S4) are combined to form a segment sequence, and that the segment sequence of the operational data set (12) is assigned to a class by a classifier; and that the segment sequence of the operational data set (12) is examined using at least one previously trained anomaly model; and that A warning message is issued if an anomaly in the segment sequence is detected during the examination of the segment sequence of the operational data set (12). [5] Method according to any of the preceding claims, characterized by that in preparation during a reference phase several reference data sets (10) are included, which each comprise a plurality of measured quantities, and that Each measurement quantity comprises a plurality of measured values, and that The measured values ​​of each quantity are recorded sequentially over time, and that Several temporally successive segments (S1, S2, S3, S4) are extracted from each of the reference data sets (10), and that Segments (S1, S2, S3, S4) of the reference datasets (10) which exhibit a relatively high similarity to each other are recognized as belonging to a common cluster, and that the segments (S1, S2, S3, S4) belonging to a common cluster are assigned to a common class, and that the classifier is trained with the segments (S1, S2, S3, S4) of the reference data sets (10) of the previously defined classes in order to assign segments (S1, S2, S3, S4) of future recorded operational data sets (12) to the previously defined classes, and that that at least one anomaly model is trained with the segments (S1, S2, S3, S4) of the reference data sets (10) to examine segments (S1, S2, S3, S4) of future recorded operational data sets (12). [6] Method according to claim 5, characterized by , that the measured values ​​of a reference data set (10) which lie within a time interval, in which one of the measured variables exhibits at least an approximately constant or at least an approximately linear course, can be recognized as belonging to a segment (S1, S2, S3, S4). [7] Method according to any one of claims 5 to 6, characterized by , that for each reference data set (10) a second derivative of one of the measured quantities with respect to time is calculated; and The second derivative of the aforementioned measured quantity determines the times of local extrema; and the measured values ​​of the reference data set (10), which were recorded between two adjacent local extrema, are recognized as belonging to one segment (S1, S2, S3, S4). [8] Method according to claim 7, characterized by that the measured quantity, from which the second derivative with respect to time is calculated, is a rotational speed of the electric motor. [9] Method according to any one of claims 5 to 8, characterized by, that segments (S1, S2, S3, S4) of reference data sets (10) which do not have a sufficiently high similarity to any of the other segments (S1, S2, S3, S4) of reference data sets (10) are discarded. [10] Method according to any one of claims 5 to 9, characterized by , that after extracting the segments (S1, S2, S3, S4) from the reference data sets (10) several temporally successive segments (S1, S2, S3, S4) are combined to form a segment sequence, and that Segment sequences of reference data sets (10) which exhibit a relatively high similarity to each other are recognized as belonging to a common cluster, and that the segment sequences belonging to a common cluster are assigned to a common class, and that the classifier is trained with the segment sequences of the reference data sets (10) of the previously defined classes in order to assign segment sequences of future recorded operational data sets (12) to the previously defined classes, and that that at least one anomaly model is trained with the segment sequences of the reference data sets (10) to investigate segment sequences of future recorded operational data sets (12). [11] System for the evaluation of a technical installation, comprising at least one anomaly model and a classifier, whereby the system is set up to carry out the method according to one of the preceding claims. [12] System according to claim 11, characterized by that the classifier is designed as a neural network. [13] System according to one of claims 11 to 12, characterized by that at least one anomaly model is designed as a neural network.

Images