Method for monitoring the state of a drive train

Abstract

The invention relates to a method for monitoring the state of a drive train (D) which has an electric motor (M) operated on an electric power network (N), said method comprising the following steps: - detecting measurement values of a voltage (Um) present at the electric motor (M), said voltage being provided by the electric power network, and measurement values of a current (Im) which flows through the electric motor (M) as a result of the voltage present at the electric motor (M); - analysing the measurement values (Um, Im); - ascertaining a voltage anomaly (Uanomal) if the detected voltage (Um) is outside a defined quality range (Uqual); - ascertaining a current anomaly (Ianomal) if the detected current (Im) does not show a specified current response (Iresp) to the detected voltage (Um); - generating a fault message for a fault (ED) in the drive train (D) if only a current anomaly (Ianomal), but not a voltage anomaly (Uanomal) was ascertained; - generating a fault message for a fault (EN) in the power network (N) if a voltage anomaly (Uanomal) was ascertained.

Description

[0001]202218759 1 Description Method for monitoring the condition of a drive train The invention relates to a method, preferably a computer-implemented method, and a device for monitoring the condition of a drive train. The invention also relates to a computer-implemented method for providing a trained function for anomaly detection. Due to their cost-effective and robust design, asynchronous motors are widely used in industry: they can be found in many applications such as fans, saws, and pumps, from relatively small systems in the watt range to systems in the MW range. Electric motors are therefore among the largest consumers of the electrical energy generated worldwide. The widespread use of asynchronous motors offers enormous potential in terms of availability and reliability, and the associated savings of time and money.In this context, the terms "condition monitoring" (CM) and "predictive maintenance" (PM) are used. A well-known method for condition monitoring of electric motors is the so-called motor current signature analysis (MCSA), see, for example, EP3961230A1 (Siemens AG) March 2, 2022. The idea behind this is to measure the currents in the supply cables of an electric motor or the electrical voltage at a connection point. These measured values ​​are then analyzed. For example, the fast Fourier transformation (FFT) is used to detect or quantify faults and operating states of the motor in the frequency domain. The classic MCSA is applied in the quasi-stationary case, i.e., at a nominally constant speed.An extension of MCSA is the analysis with short-time FFTs or wavelets in order to analyze processes with time resolution. 202218759 2 Asynchronous motors operate according to the principle of electromagnetic induction, i.e. interactions between currents or magnetic fields in the stator and rotor generate a torque that drives the rotor shaft. Motor faults such as eccentricity, in particular air gap eccentricity (when the rotation axes of the stator and rotor of the motor do not exactly coincide), broken rotor bars, defective bearings or misalignment change the amplitudes at frequencies characteristic of the faults at constant slip. Based on the change in the amplitudes of these characteristic frequencies in the frequency spectrum, motor faults can be detected using MCSA. The fault frequencies, in turn, are slip-dependent, i.e. a change in slip changes the frequency position of the fault frequencies.However, if significant distortion occurs in the motor's supply voltage, i.e., the input voltage to the motor contains significant components of higher harmonics of the supply frequency or, more generally, the input voltage to the motor exhibits spurious frequencies, significant spurious frequencies result in the current. These spurious frequencies in the current impair the MCSA and can lead to incorrect conclusions, which are described below under cases a and b. It is assumed that a detection algorithm with an AI function is used to detect current anomalies within the framework of an MCSA. This detection algorithm is initially trained using training data in a training phase before it can independently detect anomalies in a subsequent detection phase (AI = Artificial Intelligence).Case a: If the detection algorithm was trained during its training phase using data that does not contain any voltage interference frequency events, but only current interference frequency events caused by a motor fault, then in the detection phase, if voltage interference frequencies and resulting current interference frequencies are present, the motor current spectrum will be detected as "faulty" ("anomaly"), and a motor fault will be diagnosed. However, the cause of this detected anomaly is not a motor fault, but merely a deviation in the quality of the supply voltage, which, due to I = U / Z, results in a deviation in the motor current spectrum (the current I flowing through the electric motor corresponds to the quotient of the voltage U applied to the electric motor and the impedance Z of the electric motor: I = U / Z). This results in high “false positive” rates in the detection algorithm.Case b: If, however, the detection algorithm was trained during its training phase using training data that also includes events from voltage interference frequencies, it is possible that, during the detection phase, current frequency lines resulting from the voltage interference frequencies mask existing fault state spectral lines that indicate an actual motor fault. Consequently, high "false negative" rates occur in the detection algorithm. WO00 / 29902A2 (General Electric Company) May 25, 2000 describes an MCSA method to prevent a fault in the power grid from being incorrectly interpreted as a fault or defect in the monitored electric motor. Current interference signals caused by a disturbance in the power grid (voltage anomalies, disturbances in the voltage frequency) are removed from the current measurement values ​​before they are fed to the MCSA of the electric motor.For this purpose, an “active” motor model of the electric motor is used which can simulate the behavior of the - fault-free running - electric motor. WO00 / 29902A2 describes the method on page 4, lines 5-11 as follows: “The method includes acquiring simultaneous measurements of voltage and current at the motor, applying the measured voltage to the motor model and determining an equivalent current produced in the motor model by the applied voltage, subtracting the equivalent current from the measured current to produce a corrected motor current and then processing the corrected motor current through conventional motor current signature analysis.” The disadvantage of this method is that such non-linear motor models generally require a high computational effort and are subject to considerable uncertainties in their creation and parameterization. One object of the present invention is improved condition monitoring of a drive train.This object is achieved according to the invention by a method having the features specified in claim 1. The method according to the invention serves for monitoring the condition of a drive train. The drive train has an electric motor that is operated on an electrical power grid. The electric motor can be an asynchronous motor, but is not limited to this type; any other type of electric motor, e.g. a permanent magnet synchronous machine or a single-phase series-wound motor, is also suitable for the method according to the invention. Alternatively, it is also possible to carry out the method according to the invention with a generator instead of an electric motor. The electrical power grid can be a single- or multi-phase power grid, designed as an island grid or as part of an interconnected grid. In addition to the electric motor in question, the power grid can supply one or more other electrical loads with electrical energy.The voltage in the power grid can be provided by one or more generators and / or by one or more batteries. One step of the method is the recording of measured values ​​of a voltage applied to the electric motor, which is provided by the electrical power grid. The voltage provided by the electrical power grid can fluctuate in terms of amplitude, frequency and waveform, e.g. due to other loads that draw electrical energy from the power grid and due to feedback effects on the power grid. This recording of voltage measured values ​​can be carried out by voltage sensors, e.g. shunt resistors, which measure the voltage on supply lines that electrically connect the electrical terminals of the electric motor to the power grid.A further step in the process is the recording of measured values ​​of an electric current flowing through the electric motor due to the voltage applied to it. The current I flowing through the electric motor corresponds to the quotient of the voltage U applied to the electric motor and the impedance Z of the electric motor: I = U / Z. This recording of current measured values ​​can be carried out by current sensors, e.g., a shunt, Hall sensors, GMR sensors, which measure the current flowing through the electrical connections of the electric motor. A further step in the process is the analysis of the recorded measured values; the analysis is carried out by a dual anomaly detector. If the recorded voltage lies outside a defined voltage quality range, a voltage anomaly is detected. If the recorded current does not show an expected, defined current response to the recorded voltage, for example, I = U / Z, a current anomaly is detected.Once it has been determined whether a voltage anomaly, a current anomaly, or both exist, the state of the drive train is derived from this: If only a current anomaly but no voltage anomaly was detected, a fault in the drive train is reported. If a voltage anomaly was detected, a fault in the power grid is reported. The invention therefore proposes a condition monitoring system for a drive train, wherein the condition monitoring analyzes both the voltage curve of the mains voltage 202218759 6 applied to the electric motor and the current curve of the current I flowing through the electric motor, which ultimately results from the voltage applied to the electric motor and the impedance Z of the electric motor according to I=U / Z. Since both the voltage and the current are examined for anomalies in the anomaly detection method according to the invention, the detector is referred to as a "dual" anomaly detector.Voltage artifacts can be learned during a training phase. If interference frequencies occur in the voltage, the resulting current spectrum is also learned. The invention enables an MCSA, which is an analysis of the motor current of an electric motor in the drive train to be monitored, to be expanded to include monitoring of the quality of the voltage applied to the electric motor, i.e., the voltage quality in the power grid. This reduces false alarms related to the machine status and avoids "false negative" alarms. In addition, the "voltage anomaly" information is available in a dedicated manner and, if necessary, with an indication of the possible cause. The invention does not require an active machine model to calculate current characteristics that result from specific voltage anomalies.These types of nonlinear models typically require a high computational effort and involve considerable uncertainties during creation and parameterization in the field, which means they cannot reliably predict the current contributions during voltage disturbances. It is also known that fault types such as misaligned shafts in drive trains (= motor + driven machine(s)) manifest themselves through deviations in theoretical MCSA spectral lines, primarily in voltage-based features (MCSA-U) and only to a limited extent in current-based features (MCSA-I). Reliable detection of such drive train condition faults can therefore only be achieved by considering both current and voltage features. The Dual Anomaly Detector avoids the disadvantages of previous approaches and optimally combines the current and voltage feature data sets.Advantageous embodiments and further developments of the invention are specified in the dependent claims. According to a preferred embodiment of the invention, the analysis and detection of a voltage anomaly and / or a current anomaly is carried out by at least one trained AI function of the dual anomaly detector. The method therefore uses machine methods to detect an anomaly. The voltage and current variations that occur in reality can be extremely diverse because there are practically infinitely many different possibilities for the structure and elements of a power grid and a drive train. Therefore, it can be difficult to define all the criteria for determining when an anomaly exists in advance in such detail that it can be clearly determined whether or not an anomaly exists for every possible voltage and current variation that may occur.An AI function is advantageous in that it can continuously learn and evolve. This means that the AI ​​function can apply learned criteria to voltage and current variations that occur in reality and that an operator had not previously considered, allowing it to classify whether an anomaly is present or not. By applying the trained artificial intelligence function, or AI function for short, to the current characteristics and the voltage characteristics, function output data is generated that contains information about current and voltage anomalies. The trained anomaly detection function is able to extract information regarding the quality of the power supply from the measured voltage and current values ​​and information about whether the motor current is consistent with the currently measured power supply quality.202218759 8 The trained function is a function that is trained by an ML algorithm (ML = Machine Learning). Training is generally understood to be the optimization of a mapping of input parameters of a parameterized system model, e.g., a neural network, to one or more target parameters. This mapping is optimized during a training phase according to predefined, learned and / or to-be-learned criteria. In pattern recognition models, for example, a percentage of successfully recognized patterns can be used as criteria. A training structure can, for example, include a network structure of neurons in a neural network and / or weights of connections between the neurons, which are developed through training so that the criteria are met as well as possible.In the present exemplary embodiment, a neural network, abbreviated to NN, is trained using predetermined temporal voltage and current curves or voltage and current frequency spectra to detect a voltage anomaly in the supply voltage of an electric motor in a drive train and / or a current anomaly in the motor current of an electric motor in a drive train. In order to distinguish a specific voltage or current pattern from other voltage or current patterns, properties of the voltage or current pattern must be described and mathematically mapped. The more precisely the pattern can be described and the more analyzable information is available, the more reliably the pattern recognition works. To measure the voltage / grid quality, voltage feature data sets are created, e.g.Voltage spectrum or autoregression features, and used to train a voltage anomaly detector. The trained function is part of a pattern recognition device. For configuration, the pattern recognition device has one or more trainable functions, i.e. algorithms that can be executed in computing modules of the pattern recognition device and that implement machine learning methods in order to optimize pattern recognition through training. These trainable functions can be trained using known standard machine learning methods to recognize specified temporal voltage and current curves or voltage and current frequency spectra in input files as accurately as possible. For this purpose, for example,a respective pattern recognized by the function in input files is compared with a pattern actually present in the input files and the function is trained to minimize any deviation. Such a learning method is often also referred to as supervised learning. As a rule, training is more successful the more training data, i.e. files that can be used for training, are available. In particular, the training data should representatively cover as wide a range as possible of possible configurations of an electric motor-driven drive train. As already mentioned above, according to the present invention, the aim of training the neural network NN is for it to recognize a voltage and / or current anomaly. For this purpose, the function output data output by the neural network NN is compared with the anomalies actually contained in the input data.During the comparison, a deviation or difference D is formed between the function output data and the corresponding, actually present anomaly(s). The difference D represents a detection error of the neural network NN. The difference D is fed back to the neural network NN, which is trained to minimize the difference D, i.e., to recognize the type of anomaly as accurately as possible. 202218759 10 A variety of standard training methods for neural networks, in particular supervised learning, can be used to train the neural network NN. The deviation D to be minimized can be represented by a suitable cost function. In particular, a gradient descent method can be used to minimize the deviation D. In general, a trained function imitates cognitive abilities that people associate with the minds of other people.In particular, a trained function is able to adapt to new circumstances and discover and extrapolate patterns through training on training data. In general, parameters of a trained function can be adjusted through training, i.e., learning. In particular, supervised learning, semi-supervised learning, unsupervised learning, reinforcement learning, and / or active learning can be used for this purpose. In addition, representation learning, also called “feature learning,” can be used. In particular, the parameters of the trained functions can be adjusted iteratively through multiple training steps. In particular, a trained function can comprise a neural network, a support vector machine, a decision tree, and / or a Bayesian network, and / or the trained function can be based on k-means clustering, Qlearning, genetic algorithms, and / or association rules.In particular, a neural network can be a deep neural network, a convolutional neural network, or a convolutional deep neural network. Furthermore, a neural network can be an adversarial network, a deep adversarial network, and / or a generative adversarial network. According to a preferred embodiment of the invention, a feature vector for state monitoring is used with the aid of a self-organizing map (SOM). The idea of ​​the SOM is to generate a grid of X * Y neurons N. Each neuron N contains a weight vector of length L. F, i.e., equal to the length of a feature vector. Determining the number of neurons (X * Y) is a compromise between the amount of data / computational effort on the one hand, and the desired reliability of an anomaly detector on the other. In tests already conducted, 20 x 20 = 400 neurons have led to very good results. Self-organizing maps, like most artificial neural networks, work in two modes: training and mapping. In training mode, an input data set ("input space") is used to generate a map ("map space") as a lower-dimensional representation of the input data. In mapping mode, measurement data is classified using the generated map, e.g., as normal or anomalous. ^ Learning the SOM, i.e., training, is an iterative process. In the initial state, the weight vectors contain random values.The learning process now proceeds by assigning similar feature vectors from the training dataset to the same neuron. "Similar" is to be understood in the sense of a vectorial distance, i.e., depending on the application, the Euclidean distance, the cosine distance, or the Manhattan distance can be used. The goal of training is to represent a p-dimensional "input space," i.e., with p variables, as a 2-dimensional "map space." A "map space" consists of components called nodes or neurons, arranged on a hexagonal or rectangular grid with two dimensions. The number of nodes and their position on the grid are determined in advance based on the goal pursued by analyzing and investigating the "input space." Each node in the "map space" is assigned a weight vector that defines the node's position in the "input space."While the nodes remain anchored in their position in the map space, the weight vectors are approximated to the input data, i.e., the distance, e.g., the Euclidean distance, between the weight vector and the input data is minimized without destroying the topology generated by the map space. During the training phase, feature vectors that are "similar" to each other are grouped. This similarity is defined as the Euclidean distance, i.e., the smaller the distance between the feature vectors, the more similar the feature vectors are. For each "group" of similar feature vectors, one or more representatives are selected, as is the maximum distance Dmax for each group. After the training phase is completed, the SOM generated in this way can be used to create additional input space vectors, i.e.,Observations in the "input space", here: time series of current values, are to be classified by identifying the node whose weight vector is closest to the "input space" vector, i.e., has the smallest distance metric, e.g., the Euclidean distance. A feature vector generated according to the invention with p vector elements, which were obtained using an AR method from a time series measured on an electric motor, is assigned to a neuron of the said SOM and a weight vector assigned to the neuron. In order to use the trained SOM as an anomaly detector, a function is required that assigns a degree to which the feature vector Fν represents an anomaly 202218759 13. For this purpose, the "quantization error" Eν is calculated for each feature vector Fν: Eν = | Fν – Gν| for all ν, i.e., for all feature vectors. Where |...| the previously defined vectorial distance and Gν the respective weight vector of the "optimal" neuron, i.e., the neuron whose weight vector G has the minimum distance to the feature vector. The quantization error E describes how well the respective feature vector can be mapped to the distribution of the data with which the SOM was trained. A high error indicates that the feature vector is very different from the training data and therefore describes anomalous motor behavior. In practice, an anomaly in the electric motor is detected if the quantization error Ex exceeds a fixed threshold value E. Threshold exceeds: Ex > E ThresholdAfter training, anomaly detection takes place as follows: A feature vector is generated from the input values, as described above. For this vector, the Euclidean distance to the learned "groups" is determined. The minimum value of all distances is determined. If this value is greater than Dmax, an anomaly exists. According to a preferred embodiment of the invention, MCSA methods are used to analyze the measured values ​​of the motor current of the electric motor and to decide whether a current anomaly exists. In MCSA, a frequency spectrum is generated from a time series of measured values ​​of the motor current, e.g. using a Fast Fourier Transformation (= FFT). If the monitored electric motor has a motor fault, the frequency spectrum of the motor current changes compared to that of a "healthy", i.e., a fault-free running electric motor.The methods used in MCSA have long been known to those skilled in the art and have been described in many publications. According to a preferred embodiment of the invention, MCSA methods are used analogously to analyze the measured values ​​of the electric motor's supply voltage and to decide whether a voltage anomaly exists. For this purpose - analogous to the typical MCSA application to a time series of current values ​​of the electric motor - an MCSA method for determining FFT amplitudes can be applied to a time series of voltage values ​​of the electric motor in order to determine the frequency spectrum amplitudes at the frequencies specified below, see Eqs. (1) to (6) below. As described in the theory of MCSA, the frequency spectrum of an asynchronous motor (position of the frequency lines) is highly dependent on the load condition, i.e., on the slip s.Furthermore, it is known from classical MCSA that frequency lines also vary with the frequency of the supply voltage applied to the ASM (“supply frequency f. supply The MCSA primarily distinguishes between slip-dependent motor-specific spectral lines, e.g., the Principal Slot Harmonics (PSHs), various slip-dependent error lines, and the motor-specific Winding Harmonics (WHs), which are multiples of the supply frequency. The following apply to the PSH: f PSH = [k•(R / p)•(1-s) ± ν] • f supply Equation (1) and for typical motor faults: i) Broken rotor bars (Broken Bars BB): f BB1 = (1 ± 2k•s) • f supply Eq. (2) ii) Static / Dynamic rotor eccentricity (see ISO 20958:2013) f ecc1 = [1 ± (k / p)•(1-s)] • f supply Eq. (3) f ecc2 = [(R ± n d )•(k / p)•(1-s) ± ν] • f supply Eq. (4) iii) Bearing damage to the rolling bearing f o = [(N b / 2)•(1 – (D b / D p )•cos(β)] • f rotEq. (5) f i = [(N b / 2)•(1 + (D b / D p )•cos(β) / D p )] • f rot Eq. (6) 202218759 15 where f BB1 Frequency at rotor bar breakage f supply Frequency of the supply voltage, e.g., 50 Hz or 60 Hz f ecc1 Frequency of eccentricity (1st family) f ecc2 Frequency of eccentricity (2nd family) f PSH Frequency of the Principal Slot Harmonics (= PSH) f o Frequency of the bearing outer ring (o = outer) f i Frequency of the bearing inner ring (i = inner) f rot Rotation frequency of the rolling bearing k natural number p number of pole pairs = number of pole pairs R number of rotor bars = number of rotor bars s slip ν odd numbers 1, 3, 5, 7, ... = order of the harmonics of f supply n d Index depending on the eccentricity type (static: n d = 0; dynamic: n d = 1, 2, 3, 4, ...) N b Number of rolling elements D b / D pCharacteristic diameters on the rolling bearing ß contact angle According to a preferred embodiment of the invention, feature data is determined from the measured current and / or voltage values, which are used to evaluate whether the detected voltage lies within a defined quality range and / or whether the detected current shows a defined current response to the detected voltage and / or which are used to train the dual anomaly detector. Determining feature data, e.g. a mean value, minimum values ​​and / or maximum values, average values, expected value and variance of a distribution, from measured current and / or voltage measurement series is generally accompanied by a reduction in the amount of data, which considerably simplifies and accelerates data processing: instead of the extensive measured current and / or voltage measurement series, the dual anomaly detector then only has to calculate with the feature data.202218759 16 According to a preferred embodiment of the invention, the feature data are determined on the basis of measured values ​​of one or more of the following variables or parameters: voltage frequency, frequency of voltage dips and increases, voltage asymmetry, a voltage spectrum, autoregression features, the electrical active power P, the electrical reactive power Q, the electrical apparent power S, symmetry of the voltage in outer conductors, distribution of the network frequency harmonics in at least one of the voltage frequency spectra.For this purpose, conventional measuring devices can be used to record the basic electrical quantities in the low-voltage power distribution, as shown in the following extract from the Siemens system manual SENTRON PAC4200 Measuring device for recording the basic electrical quantities in the low-voltage power distribution (SENTRON Multifunction Monitoring Device SENTRON PAC4200 System Manual, Edition 05 / 2019, A5E02316180A-05, Siemens AG, Smart Infrastructure, Low Voltage & Products, PO Box 1009 53, 93009 Regensburg, Germany, 3ZX1012-0kM42-3AB0, (C) Siemens AG 2019, pages 11-12): Chapter 2 - Description - 2.1 Performance features - Measurement ● Measurement in 2-, 3- and 4-wire networks. Suitable for TN, TT and IT networks ● Measurement of all relevant electrical quantities of an alternating current system ● Recording of minimum and maximum values ​​of all measured quantities ● Determination of true effective value for voltage and current up to the 63rdHarmonics ● 4 quadrant measurement (supply and export) ● Averaging of all measured values ​​directly in the device in two independent and freely configurable stages (aggregation) ● Measurement of harmonics 1st to 64th (even and odd) ● Calculation of average values ​​across all phases for voltage and current ● Zero blind measurement 202218759 17 ● High measurement accuracy: e.g. accuracy class 0.2 according to IEC 61557-12 for active energy. This means: An accuracy of 0.2% based on the measured value under reference conditions ● Detection of voltage dips, overvoltage and voltage interruptions with user-defined threshold values ​​● Detection of the N-conductor current ● Detection of the differential and PE-conductor current via external summation current transformer ● Detection of physical quantities (e.g.Temperature, pressure, humidity) with external 0 / 4 mA to 20 mA measuring transducer According to a preferred embodiment of the invention, the supply harmonic amplitudes are determined in relation to the fundamental wave amplitude or time-dependent fluctuations in amplitude (RMS = Root Means Square) and fundamental wave frequency. The aforementioned quantities are selected in accordance with DIN EN 61000-4-7 "Methods and apparatus for measuring harmonics and interharmonics in power supply systems and connected equipment" and DIN EN 61000-4-30 "Testing and measurement techniques - Methods for measuring power quality", which also form the basis for the definition of industry / supply networks, with the aim of ensuring the supply voltage within defined limits so that electrical machines can function perfectly.The supply grids will generally meet this requirement, and the quality is often also measured in corresponding devices. However, the current measurements taken at any given time for the condition analysis of electrical machines are directly influenced by the three-phase voltage signals prevailing at that time. By determining the essential parameters of the grid quality, the current-based condition diagnosis can be improved. According to a preferred embodiment of the invention, the dual anomaly detector is a self-organizing map or an autoencoder, which detects a deviation of voltage measurement values ​​202218759 18 from a predetermined grid quality. This anomaly detector, designed as an SOM or an autoencoder, can, for example, detect deviations from undisturbed sinusoidal voltage waveforms.Deviations from undisturbed sinusoidal voltage waveforms are contained in the harmonic components, and the distribution of the harmonics in the supply can be used to correct the resulting harmonic components of the current in order to evaluate the harmonic components resulting from the motor. A further aspect of the invention is a device for monitoring the condition of a drive train having an electric motor operated by an electrical power grid. The device has sensors for detecting measured values ​​of a voltage applied to the electric motor and a current resulting from the applied voltage flowing through the electric motor.The device has a dual anomaly detector for analyzing the acquired measured values, wherein the dual anomaly detector has been trained to detect a voltage anomaly if the acquired voltage does not meet a predetermined power quality requirement, and to detect a current anomaly if the acquired current does not show an expected current response to the acquired voltage. The dual anomaly detector reports a) a fault in the powertrain if it detects only a current anomaly but no voltage anomaly, or b) a fault in the power grid if it detects a voltage anomaly. A further aspect of the invention is a computer program comprising instructions that, when the program is executed on a computer, cause the device for monitoring the condition of a powertrain to execute the method for monitoring the condition of a powertrain.A further aspect of the invention is a method for providing a trained dual anomaly detector for detecting anomalies in a drive train. The drive train has an electric motor operated on an electrical power grid, as well as sensors for measuring voltage and current values. One step of the method is receiving input training data, which represents measurement data obtained from the sensors, i.e., measured values. A further step of the method is receiving output training data, which represents anomalies in the measurement data. A further step of the method is training an AI function of the dual anomaly detector based on the input training data and the output training data such that the artificial intelligence detects voltage and / or current anomalies.During the training phase, the dual anomaly detector can also be trained using voltage feature data sets and corresponding current feature data sets. A further step of the method is providing the trained dual anomaly detector. The trained dual anomaly detector is adapted to detect current anomalies and voltage anomalies in current and voltage measurement series. A further aspect of the invention is a computer program comprising instructions which, when executed on a computer, cause the computer to execute the method according to the invention for providing a trained dual anomaly detector. The invention is explained below with the aid of the accompanying drawings. These show, schematically and not to scale, Fig. 1, an embodiment of an inventive device for condition monitoring of a drive train; Fig.2 shows a flow diagram of an embodiment of the method according to the invention; Fig. 3 shows an embodiment of an artificial neural network; 202218759 20 Fig. 4 shows a flow diagram of a computer program according to an embodiment of the present invention. Fig. 1 shows an electric motor M which is electrically connected to a power grid N by means of a branch 12. The branch can, for example, be a three-phase electrical line which connects the three-phase network N to the connection terminals of the electric motor M. The electric motor M is designed as a drive machine of a drive train D which also has a mechanical transmission G and a working machine W. A torque provided by the electric motor M is transmitted to the working machine W via the transmission G. The working machine can, for example, be a conveyor belt, a roller or a cylinder.In addition to the electric motor M of the drive train D, further electrical loads 20 are connected to the power grid N. A first sensor S1, which is arranged, for example, at the branch 12, detects voltage values ​​U. m an electrical supply voltage of the electric motor M, e.g. a voltage between two phases of the branch 12 or between a phase of the branch 12 and a reference potential, and sends the detected voltage values ​​U m via a transmission means 10, e.g. a data cable, to a device 15 for monitoring the condition of the drive train D. A second sensor S2, which is arranged, for example, on the electric motor M, detects current values ​​I m of an electric current flowing through the electric motor M and sends the current values ​​I mvia a transmission means 11, e.g., a data cable, to the device 15 for condition monitoring of the drive train D. The device 15 for condition monitoring of the drive train D has a dual anomaly detector 16, a user interface 17, and a data memory 19. The dual anomaly detector 16 has a trained AI function 18. The device 15 receives the detected voltage values ​​U 202218759 21 m and the recorded current values ​​I m and evaluates them using the trained AI function 18. The AI ​​function 18 analyzes the recorded measured values ​​U m , I m . The AI ​​function 18 represents a voltage anomaly U anomal fixed if the detected voltage U m outside a defined quality range U qual which is defined by range limits stored in the data memory 19. The AI ​​function 18 represents a current anomaly I anomalfixed if the detected current I m no predetermined current response I resp to the detected voltage U m shows; predetermined current responses are stored in the data memory 19. The device 15 reports a fault in the drive train D if only one current anomaly I anomal , but no voltage anomaly U anomal was detected. The device 15 reports a fault in the power grid N if a voltage anomaly U anomal was detected. The messages are sent via the user interface 17, e.g., a display. Fig. 2 shows a flowchart of an embodiment of the method according to the invention. In a first step 201, voltage values ​​U m an electrical supply voltage of the electric motor M and current values ​​I mof an electric current flowing through the electric motor M. In a second step 202, these measured values ​​are analyzed: it is checked a) whether the recorded voltage values ​​U m within a defined quality range U qual and b) whether the recorded current values ​​I m a given current response I resp to the detected voltage U m In a next step 203, depending on the results of the analysis, it is determined whether a voltage anomaly U anomal and / or a current anomaly I anomal is present; one of the four possible states shown in Table 204 must be present, where 0 = false and 1 = true represents the truth value of the statements “voltage anomaly U anomal is present?” or “Stromal anomaly I anomal is present?“ show: 202218759 22 - a first state {U anomal = 0; I anomal = 0} is determined if the recorded voltage values ​​Um within a defined quality range U qual and the recorded current values ​​I m a given current response I resp on the detected voltage U m show; - a second state {U anomal = 0; I anomal = 1} is determined if the recorded voltage values ​​U m within a defined quality range U qual and the recorded current values ​​I m no predetermined current response I resp on the detected voltage U m show; - a third state {U anomal = 1; I anomal = 0} is determined if the recorded voltage values ​​U m outside a defined quality range U qual and the recorded current values ​​I m a given current response I resp on the detected voltage U m show; - a fourth state {U anomal = 1; I anomal= 1} is determined if the recorded voltage values ​​U m outside a defined quality range U qual and the recorded current values ​​I m no predetermined current response I resp on the detected voltage U m show. If the first state {U anomal = 0; I anomal = 0} is detected, no error is reported 205. If the second state {U anomal = 0; I anomal = 1} is detected, an error E D reported in the drive train D 206. If the third state {U anomal = 1; I anomal = 0} or the fourth state {U anomal = 1; I anomal = 1} is detected, an error E Nreported in the power grid N 207. Fig. 3 shows a configuration of an artificial neural network 100 as used in the trained AI function 18. Alternative terms for "artificial neural network" are "neural network," "artificial neural network," or "neural network." The artificial neural network 100 has nodes 120, ..., 132 and edges 140, 141, 142, where each edge 140, 141, 142 is a directed connection of a first node 120, ..., 132 to a second node 120, ..., 132. While in general the first node 120, ..., 132 and the second node 120, ..., 132 are different nodes 120, ..., 132, it is also possible that the first node 120, ..., 132 and the second node 120, ..., 132 are identical. For example, in Fig.6 the edge 140 is a directed connection from the node 120 to the node 123, and the edge 142 is a directed connection from the node 130 to the node 132. An edge 140, 141, 142 from a first node 120, ..., 132 to a second node 120, ..., 132 is also referred to as an "ingoing edge" with respect to the second node 120, ..., 132 and as an "outgoing edge" with respect to the first node 120, ..., 132: "incoming edges" of a node are directed edges whose destination point is the node, "outgoing edges" of a node are directed edges whose origin is the node. In this embodiment, the nodes 120, ..., 132 of the artificial neural network 100 are arranged in layers 110, ..., 113, wherein the layers may have an intrinsic order introduced by the edges 140, 141, 142 between the nodes 120, ..., 132.In particular, edges 140, 141, 142 can only exist between adjacent layers of nodes. In the illustrated embodiment, there is an input layer 110 containing only nodes 120, 121, 122 without an incoming edge, an output layer 113 containing only nodes 131, 132 without an outgoing edge, and hidden layers 111, 112 between input layer 110 and output layer 113. In general, the number of hidden layers 111, 112 can be chosen arbitrarily. The number of nodes 120, 121, 122 in the input layer 110 is typically related to the number of input values ​​of the neural network, and the number of nodes 131, 132 in the output layer 113 is typically related to the number of output values ​​of the neural network. 202218759 24 In particular, each node 120, ..., 132 of the neural network 100 can be assigned a (real) number as a value. Here, x denotes (n)i denotes the value of the i-th node 120, ..., 132 of the n-th layer 110, ..., 113. The values ​​of nodes 120, 121, 122 of the input layer 110 are equivalent to the input values ​​of the neural network 100, and the values ​​of nodes 131 and 132 of the output layer 113 are equivalent to the output values ​​of the neural network 100. In addition, each edge 140, 141, 142 can have a real number as a weight, in particular a real number in the interval [-1, 1] or the interval [0, 1]. Here, w denotes (m,n) i ,j the weight of the edge between the i-th node 120, ..., 132 of the m-th layer 110, ..., 113 and the j-th node 120, ..., 132 of the n-th layer 110, ..., 113. In addition, the abbreviation w (n) i ,j defined as w (n,n+1) i ,j. In particular, to calculate the output values ​​of the neural network, the input values ​​are propagated through the neural network. Specifically, the values ​​of nodes 120, ..., 132 of the (n+1)-th layer 110, ..., 113 can be calculated based on the values ​​of nodes 120, ..., 132 of the n-th layer 110, ..., 113 as follows: x (n+1) j = f( Σ i x (n) i ∙ w (n) i ,j). Here, the function f is a transfer function (also referred to as an “activation function”). Well-known transfer functions are step functions, sigmoid functions (e.g., the logistic function, the generalized logistic function, the hyperbolic tangent, the arctangent function, the error function, the smoothstep function), or rectifier functions. The transfer function is mainly used for normalization purposes. In particular, the values ​​are propagated layer by layer through the neural network, where values ​​of the input layer 110 are given by the input to the neural network 100, values ​​of the first hidden layer 111 can be calculated based on the 202218759 25 values ​​of the input layer 110 of the neural network 100, values ​​of the second hidden layer 112 can be calculated based on the values ​​of the first hidden layer 111, etc. To determine the weights w (m,n) i ,jfor the edges, the neural network 100 must be trained using training data. In particular, training data includes training input data and training output data (denoted by t i ). For a training step, the neural network 100 is applied to the training input data to generate computed output data. In particular, the training data and the computed output data have a number of values ​​equal to the number of nodes of the output layer. In particular, a comparison between the computed output data and the training data is used to recursively adjust the weights in the neural network 100 (backpropagation algorithm). In particular, the weights are changed according to w' (n) i ,j = w (n) i ,j – γ ∙ δ (n) j ∙ x (n) i where γ is the learning rate, and the numbers δ (n) j can be calculated recursively as follows: δ(n) j = (Σ k δ (n+1) k ∙ w (n+1) j ,k ) ∙ f'(Σ i x (n) i ∙ w (n) i ,j ) based on δ (n+1) j, if the (n+1)-th layer is not the output layer, and δ (n) j = (x (n+1) k - t (n+1) j) ∙ f'(Σ i x (n) i ∙ w (n) i ,j ) if the (n+1)-th layer is the output layer 113, where f' is the first derivative of the activation function and t (n+1)j is the comparison training value for the j-th node of the output layer 113. 202218759 26 Fig. 4 shows a flow diagram of a computer program according to an embodiment of the present invention. In a first step 41, measured values ​​of a voltage applied to the electric motor, which is provided by the electrical power grid, and measured values ​​of a current flowing through the electric motor due to the voltage applied to the electric motor are recorded. In a second step 42 following the first step 41, the measured values ​​are analyzed. In a third step 43 following the second step 42, a voltage anomaly is determined if the detected voltage lies outside a defined quality range. In a fourth step 44 following the third step 43, a current anomaly is determined if the detected current does not show a specified current response to the detected voltage.In a fifth step 45 following the fourth step 44, an error message indicating a fault in the powertrain is generated if only a current anomaly but no voltage anomaly was detected. In a sixth step 46 following the fifth step 45, an error message indicating a fault in the power grid is generated if a voltage anomaly was detected.

Claims

202218759 27 claims 1. Method for monitoring the condition of a drive train (D) which has an electric motor (M) operated on an electrical power grid (N), comprising the following steps: - recording measured values ​​of a voltage (U m ), which is provided by the electrical power grid, and measured values ​​of a current (I m ), which flows through the electric motor (M) due to the voltage applied to the electric motor (M); - Analyzing the measured values ​​(U m , I m ); - Detection of a voltage anomaly (U anomal ), if the detected voltage (U m ) outside a defined quality range (U qual ) is located; - Detection of a current anomaly (I anomal ) if the measured current (I m ) no predetermined current response (I resp ) to the detected voltage (U m ) shows; - Generating an error message of an error (E D) in the drive train (D), if only a current anomaly (I anomal ), but no voltage anomaly (U anomal ) was detected; - Generating an error message of an error (E N ) in the power grid (N) if a voltage anomaly (U anomal ) was detected.

2. The method according to claim 1, wherein the analysis and detection of a voltage anomaly and / or a current anomaly is carried out by a dual anomaly detector (16) with at least one trained AI function (18).

3. The method according to one of the preceding claims, wherein feature data is determined from the measured current and / or voltage values, which feature data are used to evaluate whether the detected voltage (U m ) outside a defined quality range (U qual ) and / or whether the measured current (I m ) does not have a defined current response (I resp ) to the detected voltage (U m) and / or which are used to train the dual anomaly detector (16). 202218759 28 4. The method according to claim 3, wherein the feature data is determined based on one or more of the following variables: voltage frequency, frequency of voltage dips and swells, voltage asymmetry, a voltage spectrum, autoregression features, the electrical active power P, the electrical reactive power Q, the electrical apparent power S, symmetry of the voltage in outer conductors, distribution of the grid frequency harmonics in at least one of the voltage frequency spectra.

5. The method according to claim 3, wherein the supply harmonic amplitudes are determined in relation to the fundamental wave amplitude or time-dependent fluctuations in the amplitude and fundamental wave frequency.

6. The method according to one of the preceding claims, wherein the dual anomaly detector (16) for the voltage signals is a self-organizing map or an autoencoder that detects a deviation from a predetermined grid quality.Device for monitoring the condition of a drive train (D) having an electric motor (M) operated on an electrical power grid (N), comprising: - sensors (S1, S2) for detecting measured values ​​of a voltage applied to the electric motor (M) and measured values ​​of a current resulting from the applied voltage and flowing through the electric motor (M); - a dual anomaly detector (16) for analyzing the detected measured values, wherein the dual anomaly detector has been trained to detect a voltage anomaly if the detected voltage does not meet a predetermined grid quality, and to detect a current anomaly if the detected current does not show an expected current response to the detected voltage, wherein the dual anomaly detector a) reports a fault in the drive train if it only detects a current anomaly but no voltage anomaly, or. 202218759 29 b) reports a fault in the power grid if it detects a voltage anomaly.

8. A computer program comprising instructions which, when executed on a computer, cause the device of claim 7 to perform the method of any one of claims 1 to 6. 9.A method for providing a trained dual anomaly detector (16) for detecting anomalies in a drive train (D) having an electric motor operated on an electrical power grid, the drive train comprising sensors (7) for measuring voltage and current values, comprising the following steps: - receiving input training data representing measurement data (M) from the sensors (7), - receiving output training data representing anomalies in the measurement data (M), - training an AI function (18) of the dual anomaly detector based on the input training data and the output training data such that the AI ​​function detects voltage and / or current anomalies, - providing the trained dual anomaly detector.

10. A computer program comprising instructions which, when the program is executed on a computer, cause the computer to carry out the method according to claim 9.

Images