WO2020216452A1 - State analysis of a system - Google Patents

Abstract

The invention relates to an automated method for analyzing the state of a system (6a, 6b, 6c), wherein a sequence of measurement data of at least one measurement variable of the system (6a, 6b, 6c) is detected by means of an interface (2a, 2b, 2c), and at least two statistical characteristics are determined on the basis of the sequence by means of a computing unit (3) according to a specified set of parameters. The statistical characteristics are weighted according to the set of parameters, and a classification analysis is carried out on the basis of the weighted statistical characteristics by means of the computing unit (3) according to the set of parameters. Furthermore, the state of the system (6a, 6b, 6c) is determined by means of the computing unit (3) on the basis of the result of the classification analysis.

Description

description

Condition analysis of a plant

The invention relates to a method for analyzing the status of a plant, wherein a sequence of measurement data of at least one measured variable, which depends on a status of the plant, is recorded by means of an interface of a computer system. The invention also relates to a method for providing a set of parameters, a status analysis system, a training system and associated computer programs.

Hardware is usually required for an evaluation of industrial systems, for example an energy efficiency evaluation

and / or software adjustments to the system or the system control required in order to determine a state, for example an energetic state, of the system. However, this goes hand in hand with downtimes, as well as the risk of loss of warranty rights and high implementation costs. Intervention in the system control can also be impossible or undesirable by a system operator.

Against this background, it is an object of the present invention to provide an improved concept for the status analysis of a system which allows an automated status analysis in which access to the system control can be dispensed with.

According to the invention, this object is achieved by a method for analyzing the status of a system, a method for providing a parameter set, a status analysis system, a training system and computer programs according to the independent claims. Advantageous further developments and further embodiments are the subject of the dependent claims.

According to a first independent aspect of the improved concept, a method, in particular an at least partially wise computer-implemented method for analyzing the condition of a system, wherein a sequence of measurement data of at least one measured variable, which depends on a condition of the system, is recorded by means of an interface of a computer system, in particular according to a predetermined set of parameters. At least two statistical parameters are determined based on the sequence of measurement data by means of a computing unit of the computer system according to the specified parameter set. The statistical parameters are weighted by means of the computing unit according to the parameter set. A classification analysis is carried out by means of the arithmetic unit in accordance with the parameter set and based on the weighted statistical parameters. By means of the computing unit, the condition of the installation is determined based on a result of the classification analysis.

The system can in particular include an industrial system, a machine, an electrical machine, a pneumatic machine, a heat engine, an electrical consumer or an energy consumer.

The term “state” can be understood to mean, for example, a current operating state of the system. In particular, the term “state” is not to be understood as an aging state or as a state in relation to an expected service life of the system.

The at least one measured variable can in particular contain a measured variable of the system, in particular a state variable or an operating variable of the system. The at least one measured variable can in particular contain an energy-related variable, for example electrical power consumption, electrical voltage or electrical current intensity. The at least one measured variable can also contain a converted heat, a thermal output, an operating pressure, a pneumatic energy variable, an operating temperature, an ambient temperature, an insulating medium temperature or a temperature of a component of the system. The sequence of measurement data can be stored, for example, by means of the computing unit or another computing unit, for example on a storage medium of the computer system, in particular in order to carry out the acquisition of the following method steps.

The sequence of measurement data can, for example, correspond to a work section, in particular a time work section of the system. For example, the sequence of measurement data can include values of the at least one measured variable measured during the section.

The sequence can be understood to mean an ordered set of N-tuples of the at least one measured variable, where N corresponds to a number of measured variables which are comprised by the at least one measured variable, and N can be greater than or equal to one. Each N-tuple then corresponds to a measurement data item in the sequence.

The fact that the at least two statistical parameters can be determined according to the specified parameter set can be understood in such a way that the at least two statistical parameters are determined using one or more parameters of the parameter set. In particular, at least two initial statistical parameters based on the sequence of measurement data can be determined by means of the computing unit independently of the parameter set. The initial statistical parameters can then, for example, be scaled or normalized using one or more scaling parameters or scaling factors that are included in the parameter set in order to obtain the two statistical parameters.

The classification analysis includes, in particular, the application of a classification method to the at least two weighted statistical parameters. The classification analysis or the classification process can assign the sequence, in particular the section, to one of several, in particular predetermined, classes. For this purpose, the classification analysis can use one or more classification parameters included in the parameter set, for example. The assignment of the sequence to the class can in particular be unambiguous, that is to say the sequence is assigned to exactly one class from a predetermined set of classes according to the classification method.

Here and below, an interface can in particular contain a hardware interface and / or a software interface of the computer system. The interface can in particular contain a PCI bus, a USB or a Firewire interface.

According to the understanding on which this is based and in the following, a computing unit can have hardware elements, for example one or more microprocessors, and / or software elements.

According to the understanding assumed here, a storage medium can, in particular, be a non-permanent main memory, for example a random access memory, or a permanent mass storage medium, for example a hard drive, a flash memory, an SD card or a solid-state memory. State disk included.

Which parameters the specified parameter set contains and which values these parameters have in the parameter set depend on a specific application, in particular on a specific design of the system, and on a specific task for the status analysis.

According to a method for status analysis according to the improved concept, the status of the system can take place without intervening in the system or machine control. In particular, existing systems can also be analyzed. The condition of the plant determined by means of the method for condition analysis can be used, for example, to evaluate the plant. For example, a key performance indicator (KPI), in particular for the section concerned, an energy efficiency assessment or a potential assessment for the system can be carried out. Alternatively or additionally, the result of the status analysis can be used for process optimization of the system be used.

Advantageously, according to the improved concept, no manual data preprocessing or feature extractions, that is, determination of the statistical parameters, is required. In particular, specific knowledge, for example of a data analyst or a domain expert, is not necessary for the status analysis according to the improved concept.

According to at least one embodiment of the method for state analysis, the at least one measured variable contains a measured variable with regard to electrical or pneumatic energy, with regard to a pressure, an electrical current or a power, in particular an electrical power. The state is, for example, an energetic state, for example an energetic level, an energy level, a power level, a level of power consumption by the system or a state defined by characteristic signatures in the course of the at least one measured variable.

According to at least one embodiment, the at least two statistical parameters, in particular for each of the at least one measured variable, contain a maximum value of the at least one measured variable, a minimum value of the at least one measured variable, an average value of the at least one measured variable, a variance of the at least one measured variable , a standard deviation of the at least one measured variable, a sum of the values, and / or a sum of the absolute values of the few At least one measured variable or one of the named parameters scaled according to the specified set of parameters.

According to at least one embodiment, the at least two statistical parameters, in particular for each of the at least one measured variables, contain a sum of all absolute changes in the at least one measured variable, a sum of all positive changes in the at least one measured variable, and a sum of all negative changes in the at least one measurement size, a number of measured values above the mean value of the at least one measured variable, a maximum number of successive measuring points above the mean value of the at least one measured variable and / or a maximum number of consecutive measured values below the mean value of the at least one measured variable or one according to the specified Scaled parameter set of the parameters mentioned.

According to at least one embodiment, the at least one measured variable is measured by at least one sensor, in particular of the computer system, in order to generate the sequence of measured data.

The at least one sensor is in particular connected or can be connected to the interface and provides the measurement data so that they can be recorded by means of the interface.

According to at least one embodiment of the method for state analysis, the at least one sensor contains an energy measuring device, an energy meter, an ammeter, a voltage measuring device, a temperature sensor, a pressure sensor or a heat sensor.

According to at least one embodiment of the method for status analysis, the specified parameter set contains a length of the sequence and / or a set of weighting factors for weighting the statistical parameters and / or a classification parameter for the classification analysis. The length of the sequence is, for example, a number of measuring points or N-tuples within the sequence, so to speak the duration of a recording time period for recording the measuring points of the sequence. The recording period can be referred to as a window or a data window, for example.

In embodiments in which the acquisition of the sequence of measurement data takes place according to the parameter set, the acquisition takes place in particular according to the length of the sequence. This means, for example, that the at least one sensor from the

Interface, the computing unit or a control unit is controlled according to the length of the sequence in order to record a corresponding number of measuring points or to carry out the measurement over a corresponding period of time. Alternatively, the sensor can measure the at least one measured variable continuously and / or permanently or continuously, in particular over several sections, and the detection itself by the interface takes place according to the length of the sequence.

In corresponding embodiments, the parameter set can also contain one or more scaling parameters or scaling factors for scaling the initial statistical parameters.

By using one or more of the named parameters of the parameter set, a reliable status analysis, in particular an energy-related status, of the system can be achieved.

According to at least one embodiment, two or more of the at least two statistical parameters are assigned to a given feature block and the weighting of the two or more statistical parameters assigned to the feature block is assigned a common weighting factor, in particular a common weighting factor for the feature block of the parameter set.

The fact that the two or more parameters are assigned to the feature block can be understood to mean that it is predefined that the two or more of the at least two statistical parameters belong to the specified feature block, so that the assignment itself is not necessarily part of the method.

By treating the two or more statistical parameters in the same way within the specified feature block in the weighting, an equal or similar influence of the different statistical parameters within the block can be taken into account. In addition, the complexity of the process and the computing effort are reduced. Depending on the question or the type of state that is to be analyzed, the feature block and / or further feature blocks can be weighted differently, for example.

According to at least one embodiment, each of the at least two statistical parameters is assigned to one, in particular exactly one, feature block of a predetermined set of two or more feature blocks. The weighting of the statistical parameters assigned to the respective feature blocks is then carried out with a respective common weighting factor of the respective feature block.

According to at least one embodiment, the classification analysis by means of the computing unit identifies a class, in particular a class for the sequence, and a probability for the class is determined.

The identified class together with the probability can be viewed, for example, as the result of the status analysis. The class that is identified corresponds, for example, to the state of the system and the probability corresponds to a probability with which the system is actually in the identified state or in the identified class during the section belonging to the sequence.

According to at least one embodiment, post-processing of the result of the classification analysis is carried out by means of the computing unit if the probability for the identified class is less than a predefined threshold value.

The state can then be determined based on the result of the post-processing.

According to a second independent aspect of the improved concept, a method for providing a parameter set, in particular for a method for analyzing the condition of a plant, in particular according to the improved concept, is given. According to the method for providing the parameter set, a large number, i.e. three or more, of training sequences of measurement data of at least one measured variable, which depends on a state of a training system, is recorded by means of a first interface of a training computer system. For each of the training sequences, a first, a second and a third step are carried out by means of an arithmetic unit of the training computer system according to a predetermined initial set of parameters. In the first step, at least two statistical parameters based on the measurement data of the respective training sequence are determined by the computing unit. In the second step, the statistical parameters, in particular the respective training sequence, are weighted by means of the computing unit, in particular based on the initial parameter set. In the third step, a classification analysis is carried out by means of the computing unit, in particular based on the initial set of parameters, based on the weighted statistical parameters, in particular the respective training sequence. Evaluation data of a result of the classification analyzes, in particular the classification analyzes for all of the training follow is recorded by means of a second interface of the training computer system. By means of the computing unit, an adapted parameter set is generated depending on the evaluation data, in particular by adapting the initial parameter set. By means of the computing unit, the first, the second and the third step are repeated for each of the training sequences according to the adapted parameter set, in particular if the adapted parameter set was generated as a function of the evaluation data.

The evaluation data of the results of the classification analyzes can in particular relate to the respective results of the respective classification analyzes of individual training sequences or an overall result of all classification analyzes.

The steps are carried out in accordance with the initial set of parameters, in particular, in that each of the steps is carried out in accordance with the set of parameters.

Repeating the first, the second, and the third

Steps for each of the training sequences in accordance with the adapted parameter set include, in particular, performing the steps in accordance with the adapted parameter set instead of the initial parameter set.

The acquisition of the evaluation data is also repeated, for example after the repetition of the steps in accordance with the adapted set of parameters. In particular, the acquisition of the evaluation data and the repetition of the first, second and third steps with the adjusted parameter set are repeated iteratively until the evaluation data prove that the result of the classification analyzes is sufficient.

If it is determined that the result of the classification analysis is sufficient, the last used adjusted parameter set can be output or made available, for example by means of a third interface of the computing unit. in particular to be used as a predetermined parameter set of a method for condition analysis according to the improved concept.

Whether the result of the classification analyzes is sufficient can be determined, for example, automatically and / or based on an assessment by a human user.

The training system can in particular be a system that is identical or similar or comparable to a system whose state is obtained by means of a state analysis according to the improved concept based on a given parameter set by a method for providing a parameter set according to the improved concept should be analyzed. In particular, the training system can also be the same system.

If the evaluation data are generated by means of the processing unit or another processing unit of the training computer system, the second interface can be, for example, an internal or software interface of the training computer system. If the evaluation data is generated by a human user, the second interface can be an external or hardware interface or a user interface or a

Trade human-machine interface parts. A combination of the possibilities mentioned is also possible, so that the second interface can contain a combination of an internal and an external interface or user interface.

The training sequences can, for example, be generated analogously to the generation of the sequence for a method for state analysis according to the improved concept. In particular, the at least one measured variable can be recorded over a certain period of time, for example by means of a sensor. The recorded measured variables can then be divided into sections each section corresponds to one of the training sequences.

According to at least one embodiment, the at least one measured variable is measured by at least one sensor, for example of the training computer system, in order to generate the plurality of training sequences.

According to at least one embodiment of the method for providing the parameter set, the initial parameter set contains a length of the training sequences, the lengths of all training sequences being the same, and / or a set of weighting factors for weighting the at least two statistical parameters and / or one or several classification parameters for classification analysis.

According to at least one embodiment of the method for providing the parameter set, two or more of the at least two statistical parameters are assigned to a predetermined feature block and the weighting of the two or more statistical parameters assigned to the feature block is carried out with a common weighting factor of the initial parameter set.

According to at least one embodiment of the method for providing the parameter set, a class for the training sequence is identified by the classification analysis of a respective training sequence by means of the computing unit and a probability for the respective identified class is determined.

The parameters which are contained in the initial parameter set or the values of the parameters in the initial parameter set can, for example, correspond to empirically determined or at least partially randomly selected starting values for the corresponding parameters. According to at least one embodiment, the values of the initial parameters of the initial parameter set can be determined by user inputs, for example relating to a system or machine type or a work cycle of the system or expected states of the system or a question for the status analysis.

In particular, the weighting factors of the initial parameter set can be determined from the question. For example, a value for the length of the training sequences for the initial set of parameters can be determined from the work cycle and / or the system or machine type. From the states to be expected and / or the type of plant or machine, for example one or more classification parameters can be determined for the classification analyzes, for example a number of expected states or a number of classes.

The type of system can be, for example, a machine tool, a press machine or a robot.

The user input for determining the initial parameter set can accelerate a convergence or an overall duration of the method for providing the parameter set until the result of the classification analyzes is sufficient. But it is not absolutely necessary.

According to at least one embodiment, to carry out the classification analysis, in particular for each of the training sequences, an unsupervised classification is carried out by means of the arithmetic unit according to the parameter set, in particular the initial parameter set and / or the adjusted parameter set, and based on the weighted statistical parameters. The use of an unsupervised classification analysis or an unsupervised classification method advantageously means that no labeled or special training data set is required for the parameter set according to the improved concept to provide. This is particularly advantageous since the provision or generation of such a training data set is time-consuming and resource-intensive and can also require specific expert knowledge.

The unsupervised classification can include, for example, a K-means method, in particular a fuzzy K-means classification method. In such a method, the initial set of parameters can contain a number of classes as classification parameters, for example.

According to at least one embodiment, a user input is recorded by means of the second interface in order to record the evaluation data. For example, in various embodiments, based on the result of the classification analyzes, a graphic representation of the result of the classification analyzes can be generated and output on a user interface.

A user can evaluate the graphical representation and, based on this, enter or provide the evaluation data at least partially via the second interface, so that they can be recorded by means of the second interface. The evaluation data provided by the user can in particular contain information on whether the results of the classification analyzes are considered sufficient or not. Optionally, it can also contain a qualified statement, for example about whether a larger number of classes should be used or whether an incorrect class assignment was made or whether the number of classes used is considered sufficient. On the basis of such qualified statements, the adjustment of the parameter set can take place in a more targeted manner, so that a faster convergence of the parameter set is possible.

Acquiring the evaluation data by means of the user input has the advantage that it can be carried out quickly and easily and that an experience of the user can be used to advantage. In addition, a visual recording of the graphic representation by a person is particularly effective and efficient.

According to at least one embodiment of the method for providing a parameter set, the result of the classification analyzes is evaluated by means of the arithmetic unit or by means of a further arithmetic unit of the training computer system using a machine learning method, for example using an artificial neural network, in order to generate the evaluation data.

The artificial neural network can in particular be a folding neural network (English:

"Convolutional neural network", CNN). Alternatively, the artificial neural network can be combined, for example, with an autoencoder to generate the evaluation data.

By using an artificial neural network to generate the evaluation data, it is possible to reduce the dependency on human input and thus possibly the susceptibility to errors.

Particularly advantageous embodiments contain both the user input and the evaluation of the result of the classification analyzes using the artificial neural network for recording the evaluation data. In this way, an efficient yet particularly robust assessment can be made.

According to at least one embodiment, a comparison of marked time data series is carried out by means of analysis methods in order to acquire the evaluation data.

With regard to further advantages of a method for providing a parameter set, reference is made to the advantages of the referred to different embodiments of a method for state analysis according to the improved concept.

Further embodiments of the method for providing the parameter set according to the improved concept result directly from the various refinements and embodiments of the method for analyzing the status of a system according to the improved concept and vice versa.

According to at least one embodiment of the method for analyzing the status of a system according to the improved concept, the specified parameter set was or is provided by means of a method for providing a parameter replacement according to the improved concept.

According to a further independent aspect of the improved concept, a condition analysis system with a computer system is specified. The computer system has an interface and a computing unit. The interface is set up to acquire a sequence of measured data of at least one measured variable, which depends on a state of a system. The computing unit is set up to determine at least two statistical parameters based on the sequence of measurement data in accordance with a predefined set of parameters. The computing unit is set up to weight the statistical parameters according to the parameter set and to carry out a classification analysis according to the parameter set and based on the weighted statistical parameters. The computing unit is also set up to determine the state of the installation based on a result of the classification analysis.

Further embodiments of the condition analysis system according to the improved concept result directly from the method for condition analysis of a plant according to the improved concept and the corresponding configurations and vice versa. In particular, the state analysis system is for performing a method for state analysis according to the improved Concept trained or programmed. In particular, the method for state analysis is carried out by a state analysis system according to the improved concept.

According to a further independent aspect of the improved concept, a first computer program is specified. The first computer program has commands which, when the first computer program is executed by a computer system, in particular by a computer system of a status analysis system according to the improved concept, cause the computer system to perform a method for status analysis of a system according to the improved concept.

According to a further independent aspect, a computer-readable storage medium is specified on which a first computer program according to the improved concept is stored.

According to a further independent aspect of the improved concept, a training system with a training computer system is specified. The training computer system has a first and a second interface and a computing unit. The first interface is set up to record a large number of training sequences of measurement data from at least one measurement variable, which depends on a state of a training system. The processing unit is set up to carry out a first, a second and a third step for each of the training sequences in accordance with a predetermined initial parameter set. In the first step, at least two statistical parameters are determined based on the measurement data of the respective training sequence. In the second step, the statistical parameters of the respective training sequence are weighted. In the third step, a classification analysis based on the weighted statistical parameters of the respective training sequence is carried out. The second interface is set up to record evaluation data of a result of the classification analyzes, in particular of all training sequences, and the re- The computer unit is set up to generate an adapted parameter set as a function of the evaluation data. The computing unit is also set up to repeat the first, the second and the third step for each of the training sequences according to the adapted parameter set, in particular if the parameter set has been adapted as a function of the evaluation data.

Further embodiments of the training system result directly from the various embodiments of the method for providing the parameter set according to the improved concept and vice versa.

In particular, the training system is designed or programmed to carry out a method for providing the parameter set according to the improved concept. In particular, the method for providing the parameter set is carried out by the training system.

According to a further independent aspect of the improved concept, a second computer program is specified. The second computer program has commands which, when the second computer program is executed by a computer system, in particular by a training computer system of a training system according to the improved concept, cause the computer system to carry out a method for providing a parameter set according to the improved concept.

According to a further independent aspect of the improved concept, a computer-readable storage medium is specified on which a second computer program according to the improved concept is stored.

The invention or the improved concept is explained in more detail below using specific exemplary embodiments and associated schematic drawings. Identical or functionally identical elements are provided with the same reference symbols in the figures. In the figures show: 1 shows an exemplary embodiment of a status analysis system according to the improved concept and various aspects of an exemplary embodiment of a method for status analysis according to the improved concept; and

FIG 2 shows an exemplary embodiment of a training system according to the improved concept and various aspects of an exemplary embodiment of a method for providing the parameter set according to the improved concept.

In Figure 1 is a schematic representation of a state analysis system is shown according to the improved concept, wel Ches a computer system 1 has.

The computer system 1 is connected to one or more systems 6a, 6b, 6c in each case via a sensor 7a, 7b, 7c, for example an energy meter. The computer system 1 has a computing unit 3, for example containing at least one microprocessor 4, and an interface 2a, 2b, 2c for connecting the computing unit 3 to the sensors 7a, 7b, 7c. The computer system 1 is coupled to an input and output device 5, for example.

FIG. 2 shows a schematic representation of a training system based on the improved concept, which has a training computer system 9.

The training computer system 9 is connected to a training system 6d via a sensor 7d, for example an energy meter. The training system 6d and the sensor 7d can, for example, be identical or structurally identical or comparable to one of the systems 6a, 6b, 6c and one of the sensors 7a, 7b, 7c. The training computer system 9 has a processing unit 10, for example containing at least one microprocessor, and for connection to the transmitter sor 7d has a first interface 11. The training computer system 9 is, for example, coupled via a second interface 12 to a user interface 13, which can serve as an input and output device.

During a training phase, in a step 100 the sensor 7d can record at least one measured variable, in particular an electrical power consumption and / or a current intensity, of the training system 6d and transmit it to the computing unit 10.

In an optional step 200, information such as a type, a work cycle or possible energetic states of the training system 6d can be recorded by the computing unit 10, for example through a user input or a database query.

In a likewise optional step 300, the computing unit 10 can detect a question to be analyzed, for example through a user input or a database query, for example whether the energy levels of the training system 6d should be identified or whether recurring signature sections should be identified in the measured variable special regardless of the energy levels.

According to at least one embodiment, the values of the initial parameters of the initial parameter set can be determined by user inputs, for example relating to a system or machine type or a work cycle of the system or expected states of the system or a question for the status analysis.

In a step 400, the computing unit 10 can divide the recorded values for the measured variable into sections to be analyzed, for example depending on the work cycle of the system or a sampling rate when the measured variable is recorded. For example, a section length can vary with the work cycle be equated. The values of the measured variable of the individual sections can be viewed as respective training sequences. The segment length, or the length of the training sequences, represents a given initial parameter.

In a step 500, the computing unit 10 can automatically perform a feature extraction, for example specifically for the respective question. In particular, the feature extraction can be carried out based on, for example, empirically determined feature blocks. These feature blocks, which each contain one or more optionally scaled statistical parameters, can extract different information from the recorded measured variables.

In a step 600, the arithmetic unit can weight the feature blocks or the statistical parameters of the feature blocks differently and thereby select how much the individual feature blocks should be taken into account in order to answer the question.

The training system begins, for example, with an initial set of weighting parameters, empirically determined for example for the question, and corrects these if necessary in the further course of the training.

The parameters, which, optionally scaled by means of one or more initial scaling parameters, can be assigned to the feature blocks, can contain for each section: a maximum value of the measured variable, a minimum value of the measured variable, a mean value of the measured variable, a variance of the measured variable, a standard deviation of the Measured variable, a sum of the values of the measured variable, a sum of all absolute changes in the measured variable ("absolute sum of changes"), a sum of all positive changes in the measured variable ("positive sum of changes"), a sum of all negative changes in the few at least one measured variable ("negative sum of changes"), a number of measured values above the mean value of the measured variable ("count above mean"), a maximum number of consecutive measuring points above the mean value of the measured variable ("longest strength above mean") and / or a maximum number of consecutive measured values below the mean value of the measured variable ("Longest strike below mean").

The scaled extracted features are placed in an unsupervised classifier in a step 700. One possibility can be a fuzzy K-means classifier. In the training phase, the classifier is trained, for example, with all recorded, preprocessed measurement data. With different classifiers, there can be different parameters which, if necessary, can be adapted differently from the parameters initially used. For example, in the case of a fuzzy K-means classifier, a number of classes can be specified which also serves as an initial parameter. In the course of the training, this can also be iteratively adapted, for example, if a user wants a more detailed analysis of the data with more classes.

The classes can, for example, correspond to the energetic states of the training facility 6d, which, for example, have one or more of the states "Off", "Ready"

("Standby"), "Ready for operation" ("Operational"), "Working", "Powering up" or "Powering Down".

If a system such as the training system 6d, for example, generally has at least three states, an initial number of classes equal to three can be specified, for example.

A result of the classification contains, for example, a class for each section and optionally a probability for the respective class.

In a step 800, the result can optionally be output to a user, for example in the form of a graphic see illustration 8, which in the example shown can contain an electrical power consumption P of the training system 6d as a function of time t.

In representation 8, the classes or states i, ii, iii, iv found in each case can be assigned to the sections, for example by means of color coding.

The result can, for example, be evaluated in a step 900 by means of a user interaction. For this purpose, the user can, for example, confirm the result, i.e. evaluate it as sufficient or reject the result. Optionally, if the result is rejected, the user can provide additional information, for example whether the number of classes found is correct and / or whether the correct classes have been identified and / or whether a more detailed analysis with more classes is required or desired.

As an alternative or in addition to the evaluation by means of user interaction, an automatic evaluation can take place in step 1000, for example using the computing unit 10 or a cloud, based on known successful analysis results of the same or similar systems. For this purpose, for example, a comparison of image files representing the respective results can be carried out using a folding neural network (CNN) or an auto-encoder with a subsequent neural network.

In step 1100 it can be checked whether a sufficiently good result is present. If this is not the case, the

Steps 400 to 800 and optionally 900 to 1100 are iteratively run through during the training phase, depending on the evaluation of the result, until a sufficiently good result is available. The initial parameters can be replaced by adapted parameters and adapted again in each iteration. If the result is sufficiently good, the last used parameter set can be used in a step 1200 as a specified parameter set for status analysis. analysis of one or more of the systems 6a, 6b, 6c of FIG. 1 can be used or specified.

The analysis of the condition of plant 6a is described below. The versions are, however, basically independent of the type of system and, in particular, can be transferred to systems 6b, 6c.

For example, during a running time of the system 6a, the sensor 7a can be used to record a repeated, in particular continuous, recording of a measured variable, in particular an electrical power consumption and / or a current intensity, of the system 6a and transmit it to the computing unit 3. The measurement data can be stored in a storage medium, for example the computing unit 3.

If data corresponding to a length of a section are stored in accordance with the predefined set of parameters, the computing unit 3 can be used to extract features.

For this purpose, at least two statistical parameters of a sequence corresponding to the stored measurement data, as described with regard to the training phase, are determined and scaled using scaling parameters of the specified parameter set. In addition, the statistical parameters, for example the feature blocks described above, are weighted with the weighting parameters according to the specified parameter set.

The weighted statistical parameters are subjected to a classification analysis by means of the computing unit, for example by a fuzzy K-means classifier, the classification parameters of the specified parameter set being used, for example the number of classes for the fuzzy K-means determined during the training

Classifier. As a result of the classification analysis, a class, for example with a corresponding class probability, is determined and optionally output or used or further processed for evaluating the system 6a.

Post-processing of the classification can optionally take place, for example if the analyzed section could only be assigned to a class with a probability that is less than a predetermined limit value.

The output class can represent the status of the analyzed section and can in particular be passed on to a system that can use the information about the status to calculate, for example, KPIs per status, an energy efficiency or a potential estimate for the system 6a, or to optimize the operation of the system 6a.

Using the improved concept, the condition of a system can be determined automatically without having to access its control.

Reference character list

1 computer system

2a, 2b, 2c interfaces

3 arithmetic unit

4 microprocessor

5 input and output device

6a, 6b, 6c systems

6d training facilities

7a, 7b, 7c, 7d sensors

8 Presentation of results

9 Training computer system

10 arithmetic unit

11 interface

12 interface

13 User interface 10 0 process step

20 0 method step 300 method step 400 method step 500 method step 600 method step 700 method step 800 method step 900 method step

10 0 0 process step

11 0 0 process step

12 0 0 procedural step

Claims

Claims 1. A method for analyzing the status of a system (6a, 6b, 6c), a sequence of measurement data of at least one measured variable, which depends on a status of the system, being recorded by means of an interface (2a, 2b, 2c) of a computer system (1), characterized in that at least two statistical parameters based on the sequence of measurement data by means of a computing unit (3) of the computer system (1) according to a predetermined set of parameters be determined; the statistical parameters are weighted by means of the computing unit (3) according to the parameter set; a classification analysis is carried out by means of the computing unit (3) in accordance with the parameter set and based on the weighted statistical parameters; and the state of the installation (6a, 6b, 6c) is determined by means of the computing unit (3) based on a result of the classification analysis. 2. The method according to claim 1, characterized in that the at least one measured variable is measured by at least one sensor (7a, 7b, 7c) in order to generate the sequence of measurement data. 3. The method according to any one of claims 1 or 2, characterized in that the parameter set a length of the sequence; and or a set of weighting factors for weighting the statistical parameters; and or a classification parameter for the classification analysis; and or one or more scaling parameters for scaling the at least one measured variable includes. 4. The method according to any one of claims 1 to 3, characterized in that two or more of the at least two statistical parameters are assigned to a predetermined feature block and the weighting of the two or more statistical parameters assigned to the feature block is carried out with a common weighting factor of the parameter set. 5. The method according to any one of claims 1 to 4, characterized in that a class is identified by the classification analysis by means of the computing unit (3); and a probability for the identified class is determined. 6. Procedure for providing a parameter set, characterized in that, a plurality of training sequences of measurement data of at least one measurement variable, which depends on a state of a training system (6d), is recorded by means of a first interface (11) of a training computer system (9); for each of the training sequences, a first, a second and a third step are carried out by means of a computing unit (10) of the training computer system (9) according to a predetermined initial set of parameters; in the first step, at least two statistical parameters are determined based on the measurement data of the respective training sequence; in the second step the statistical parameters are weighted; In the third step, a classification analysis is carried out based on the weighted statistical parameters; Evaluation data of a result of the classification analyzes by means of a second interface (12) of the Training computer system (9) are recorded; an adapted parameter set is generated by means of the computing unit (10) depending on the evaluation data and the first, the second and the third step are repeated for each of the training sequences according to the adjusted parameter set. 7. The method according to claim 6, characterized in that to carry out the classification analysis according to the initial or adapted parameter set, an unsupervised classification is carried out by means of the arithmetic unit (10) based on the weighted statistical parameters; and or a class is identified by the classification analysis by means of the arithmetic unit (10) and a probability is determined for the identified class. 8. The method according to any one of claims 6 or 7, characterized in that a user input is acquired by means of the second interface (12) to acquire the evaluation data. 9. The method according to any one of claims 6 to 8, characterized in that by means of the arithmetic unit (10) or another arithmetic unit of the training computer system (9), the result of the classification analyzes is evaluated using a machine learning method in order to generate the evaluation data. 10. The method according to any one of claims 6 to 9, characterized in that the initial parameter set has a length of the respective training sequence; and or a set of weighting factors for weighting the statistical parameters; and or a classification parameter for the classification analysis and / or one or more scaling parameters for scaling the at least one measured variable includes. 11. The method according to any one of claims 1 to 5, characterized in that the specified parameter set is provided by means of a method according to one of claims 6 to 10. 12. Condition analysis system with a computer system (1); the computer system (1) having an interface (2a, 2b, 2c) and a computing unit (3); wherein the interface (2a, 2b, 2c) is set up to record a sequence of measurement data of at least one measurement variable, which depends on a state of a system (6a, 6b, 6c); characterized in that the computing unit (3) is set up to determine at least two statistical parameters based on the sequence of measurement data in accordance with a predetermined set of parameters; the computing unit is set up to weight the statistical parameters according to the parameter set; the computing unit (3) is set up to carry out a classification analysis according to the parameter set and based on the weighted statistical parameters; and the computing unit (3) is set up to determine the state of the installation (6a, 6b, 6c) based on a result of the classification analysis. 13. Computer program with instructions which, when the computer program is executed by a computer system, in particular by a computer system (1) of a status analysis system according to claim 12, cause the computer system to carry out a method according to one of claims 1 to 5 or claim 11. 14. Training system with a training computer system (9); the training computer system (9) having a first and a second interface (11, 12) and a computing unit (10); characterized in that the first interface (11) is set up to acquire a plurality of training sequences of measurement data of at least one measurement variable which depends on a state of a training facility (6d); the computing unit (10) is set up to carry out a first, a second and a third step for each of the training sequences in accordance with a predetermined initial set of parameters; in the first step, at least two statistical parameters are determined based on the measurement data of the respective training sequence; in the second step the statistical parameters are weighted; In the third step, a classification analysis is carried out based on the weighted statistical parameters; the second interface (12) is set up to record evaluation data of a result of the classification analyzes; and the computing unit (10) is set up to generate an adapted parameter set as a function of the evaluation data and to repeat the first, second and third step for each of the training sequences in accordance with the adapted parameter set. 15. Computer program with commands which, when the computer program is executed by a computer system, in particular by a training computer system (9) of a training system according to claim 14, cause the computer system to carry out a method according to one of claims 6 to 10.