DE102022213123A1 - Method and device for determining the condition of a component to be monitored in a machine with a rotating machine element - Google Patents

Abstract

The invention relates to a method for monitoring the condition of a component (2) in a machine (1) with at least one rotating machine element (3), with the following steps:- detecting (S1) a temporal progression of a measured variable of a structure-borne sound sensor (4) as a measurement signal;- converting (S2) the temporal progression of the measurement signal into an excitation signal using a predetermined structure-borne sound transfer function, wherein the structure-borne sound transfer function describes a transfer function for the structure-borne sound from the location of the component (2) to be monitored to the location of the monitoring structure-borne sound sensor (4),- evaluating (S4) the excitation signal and in particular at least one operating variable using a data-based condition monitoring model, wherein the condition monitoring model is trained to assign a state variable depending on the excitation signal and in particular the at least one operating variable, wherein the state variable indicates the functionality of the component to be monitored;- signaling (S5) the state variable.

Description

Technical area

The invention relates to methods for monitoring the condition of components on rotating machine elements, such as rolling bearings and the like. In particular, the invention relates to the condition monitoring of components that are connected to rotating machine elements using structure-borne sound sensors with limited placement options.

Technical background

In machines with rotating machine elements, components between the rotating machine elements and fixed parts of the machines, such as rolling bearings and the like, are often subject to wear, which limits the service life and/or functionality of the machine. These components must therefore be regularly maintained and/or replaced. Due to a lack of condition monitoring, this maintenance is usually carried out too early for safety reasons, i.e. the component in question is fully functional, or too late, i.e. the component has already caused a failure or impaired functionality of the machine.

By monitoring the condition of the component in question and using the resulting predictions of the remaining service life, the components in question can be maintained in a condition-based manner, which can avoid unnecessary downtime and thus save considerable costs. Condition-based maintenance makes it possible to minimize downtime due to premature failure of a component. The more precisely the remaining service life and/or the degree of functionality can be predicted, the more cost-effectively the maintenance can be planned and carried out.

With the help of structure-borne sound sensors, changes in the vibration behavior of the component during operation of the machine can be detected due to the rotation of a rotating machine element. By evaluating the measurement signals from the structure-borne sound sensors, it is thus possible to draw conclusions about the condition of the component to be monitored. However, structure-borne sound sensors cannot always be arranged directly on the component to be monitored, as the installation space may not be sufficient or the functionality of the component to be monitored would be impaired.

In order to assign the corresponding states to the measured structure-borne sound signals, data-driven models can be used using machine learning. An algorithm learns the relationships between the signals and the state using training data. The data acquisition and state recording required for training is complex.

The transferability of such trained machine learning models to other machine types has so far generally not been possible, as the measured structure-borne noise depends on a large number of influencing factors. One of these influences is the structural dynamic behavior of the machine. For example, if stiffening measures are carried out as part of a new product generation or the positioning of the structure-borne noise sensor is changed, this leads to measurement signals with different properties, even if the force excitation by the component to be monitored is exactly the same. These changes in the signal lead to incorrect predictions by the already trained machine learning model.

It is the object of the present invention to provide a condition monitoring of a component in a machine with rotating machine elements, which can be carried out using measurement signals from a structure-borne sound sensor that is not arranged directly on the component to be monitored. It is also the object of the present invention to enable the transferability of trained data-based condition monitoring models to other machine types from a structural dynamics perspective.

Disclosure of the invention

According to the invention, a method for monitoring the condition of a component in a machine with rotating machine elements according to claim 1 as well as a corresponding device and a training method for a data-based monitoring model according to the independent claims are provided.

Further embodiments are specified in the dependent claims.

• - Erfassen eines zeitlichen Verlaufs einer Messgröße eines Körperschallsensors als Messsignal;
• - Umwandeln des zeitlichen Verlaufs des Messsignals mithilfe einer vorgegebenen Körperschallübertragungsfunktion in ein Anregungssignal, wobei die Körperschallübertragungsfunktion eine Übertragungsfunktion für den Körperschall von dem Ort der zu überwachenden Komponente zu dem Ort des überwachenden Körperschallsensors beschreibt,
• - Auswerten des Anregungssignals sowie insbesondere mindestens eine Betriebsgröße mithilfe eines datenbasierten Zustandsüberwachungsmodells, wobei das Zustandsüberwachungsmodell trainiert ist, um abhängig von dem Anregungssignal und insbesondere der mindestens eine Betriebsgröße eine Zustandsgröße zuzuordnen, wobei die Zustandsgröße die Funktionsfähigkeit der zu überwachenden Komponente angibt;
• - Signalisieren der Zustandsgröße.

According to a first aspect, a method for monitoring the condition of a component in a machine with at least one rotating machine element is provided, comprising the following steps:

• - Recording a temporal progression of a measured value of a structure-borne sound sensor as a measurement signal;
• - converting the temporal progression of the measurement signal into an excitation signal using a predetermined structure-borne sound transfer function, whereby the structure-borne sound transfer function describes a transfer function for the structure-borne sound from the location of the component to be monitored to the location of the monitoring structure-borne sound sensor,
• - Evaluating the excitation signal and in particular at least one operating variable using a data-based condition monitoring model, wherein the condition monitoring model is trained to assign a state variable depending on the excitation signal and in particular the at least one operating variable, wherein the state variable indicates the functionality of the component to be monitored;
• - Signaling the state variable.

• - Durchführen einer Merkmalextraktion aus dem zeitlichen Verlauf des Anregungssignals, um ein oder mehrere Anregungsmerkmale zu erhalten;
• - Auswerten des einen oder der mehreren Anregungsmerkmale sowie insbesondere mindestens eine Betriebsgröße als Eingangsgrößenvektor mithilfe des datenbasierten Zustandsüberwachungsmodells, wobei das Zustandsüberwachungsmodell trainiert ist, um die Anregungsmerkmale und die mindestens eine Betriebsgröße einer Zustandsgröße zuzuordnen, wobei die Zustandsgröße die Funktionsfähigkeit der zu überwachenden Komponente angibt;

Furthermore, the following further steps may be planned:

• - performing feature extraction from the time course of the excitation signal to obtain one or more excitation features;
• - evaluating the one or more excitation characteristics and in particular at least one operating variable as an input variable vector using the data-based condition monitoring model, wherein the condition monitoring model is trained to assign the excitation characteristics and the at least one operating variable to a state variable, wherein the state variable indicates the functionality of the component to be monitored;

To monitor the condition of a component of a machine with rotating machine elements, it is necessary to record a suitable measurement variable as a measurement signal that contains information about the condition of the component to be monitored. For components in the vicinity of rotating machine elements, such as rolling bearings, the measurement of structure-borne noise in the form of vibration accelerations or force vibrations (force signal) has proven to be useful for this purpose. Structure-borne noise sensors can be, for example, acceleration sensors in the form of a piezo sensor or a micromechanically implemented acceleration sensor. The recording of the temporal course of the measurement signal is preferably carried out in a constant operating state, in particular a constant speed of the at least one rotating machine element by which the component to be monitored is mechanically excited.

As a rule, progressive wear leads to a change in the force excitation within the component to be monitored, which leads to a change in structure-borne noise that spreads through the machine system. The acquisition of the measurement signals requires a good structural-dynamic connection of the structure-borne noise sensor to the component to be monitored.

Since the structure-borne sound sensor cannot usually be installed directly on the component to be monitored, e.g. due to space constraints and the like, it is therefore arranged at a distance from the component to be monitored. The structure-borne sound sensor and the component to be monitored are then separated from each other by a structure-borne sound transfer path due to the structure of the machine in between, the transmission behavior of which depends on the structure of the entire machine. The measurement signal recorded at the location of the structure-borne sound sensor therefore does not exactly represent the structure-borne sound caused by the component to be monitored. The change caused between the structure-borne sound caused by the component to be monitored and the measurement signal recorded by the structure-borne sound sensor can then be mapped by a structure-borne sound transfer function. The structure-borne sound transfer function can include a frequency response function and/or a phase response function. The structural dynamic system behavior therefore has a strong influence on the measurement signals and thus also on the state detection derived from the measurement signals.

The above method provides for the derivation of the sound excitation from the measurement signals or the measured variable curves of the structure-borne sound sensor and the generation of an excitation signal using a structure-borne sound transfer function between the location of the component to be monitored and the location of the structure-borne sound sensor. The excitation signal corresponds to a virtual signal that a sensor would record at the location of the structure-borne sound excitation. The sound excitation can then be evaluated in order to determine a state of the component to be monitored. The evaluation of the sound excitation can be carried out directly from the virtual excitation signals or by previously deriving excitation characteristics from the temporal course of the sound excitation and evaluating the excitation characteristics using a data-based state determination model. The data-based state determination model can be trained based on the excitation signals or the excitation characteristics of the sound excitation. to determine a state variable. The state variable can be specified, for example, in the form of a remaining service life, a degree of remaining functionality of the component to be monitored and the like.

The determination of the structure-borne sound transfer function between the component to be monitored and the structure-borne sound sensor can be carried out experimentally (test using a modal hammer or shaker) or by simulation, e.g. based on a known finite element simulation or as a combination of simulation and experiment. The structure-borne sound sensors required for the experiment to transmit the structure-borne sound transfer function in the area of the component to be monitored often cannot be attached directly to the component to be monitored due to installation space restrictions, so that the transfer path between the component to be monitored and the location of the structure-borne sound sensor used for monitoring can be determined directly experimentally.

However, another, preferably similar, sound transmission path can be measured, i.e. measurement signals are recorded at a selected point on the machine as the location of the excitation and measurement signals at the location of the structure-borne sound sensor, from which a transfer function can be determined. For this other sound transmission path, a transfer function can be modeled for the same machine in the simulation. Using an iterative process, parameters of the simulation (parameters for sound transmission behavior of connections, material damping behavior and the like) can be adjusted so that the transfer function obtained from the measurement and the simulated transfer function match as closely as possible, so that the simulation is adapted to the properties of the structure of the specific machine. After validation has taken place in this way, parameters of the simulation can be modified if necessary so that the target transfer path between the component to be monitored (position of the virtual sensor) and the location of the monitoring structure-borne sound sensor is determined by simulation.

The structure-borne sound transmission function is determined with reference to the location of the structure-borne sound sensor to be monitored. A suitable choice of this location depends on the component to be monitored. The force excitation that triggers the structure-borne sound and that is caused by mechanical damage can occur at various points within the component. If a rolling bearing is the component to be monitored, damage can occur on the inner ring, the outer ring, the rolling elements or the cage. If, for example, material breaks off from one of the components mentioned, this can lead to shock excitation when rolling over. Since it is not possible to detect from the outside where the damage occurred, a uniform location of the force excitation can be selected to determine the structure-borne sound transmission function. The contact surface between the bearing outer ring and the machine is suitable for this. All effective contact stresses can be summarized on this surface in the form of resulting force excitations.

The structure-borne sound transfer function can be applied to the measurement signals or measured value curves measured by the monitoring structure-borne sound sensor, so that the structure-borne sound actually measured at the location of the structure-borne sound sensor is transformed into a curve of an excitation signal at the location of the component to be monitored. The virtual excitation signal thus determined can now be used for condition monitoring.

The excitation signal can be evaluated for evaluation in the data-based condition monitoring model using feature extraction methods to calculate excitation characteristics. For example, the excitation signals can be evaluated in the time and frequency domain using envelope analysis, empirical mode decomposition (EMD), Hilbert-Huang transformation (HHT), wavelet transformation and Fourier transformation (in particular Fast Fourier Transform, FFT). The frequency spectrum can also be divided into frequency bands, for example of equal width or into octave or third-octave bands. Characteristics in the time domain, e.g. root mean square, standard deviation, skewness, kurtosis, peak-to-peak and the like, can be determined as statistical characteristics, e.g. as mean values or statistical moments of a higher order.

The data-based condition monitoring model can be trained, for example, by errors or failures that have occurred in the component to be monitored. For this purpose, input variable vectors from one or more of the excitation characteristics determined from the excitation signals and optionally at least one operating variable, such as a speed of the rotating machine component connected to the component to be monitored, a temperature, a medium in the machine or the like, are assigned to a state variable as training data sets. The state variable can, for example, be assigned to a remaining service life or a degree of functionality of the component to be monitored, for example 100% fully functional, 0% defective.

Signaling the state variables may include issuing an instruction to a user of the machine and/or adapting an operating mode of the machine according to the state variable.

The data-based condition monitoring model can be provided, for example, as a neural network, Gaussian process model or the like as a classification and/or regression model.

If necessary, the condition monitoring model can also be used to monitor the status of several components to be monitored based on an evaluation of a structure-borne sound sensor. In this case, a classification model is suitable as a condition monitoring model.

By preprocessing the measurement signal of the structure-borne sound sensor into an excitation signal using the structure-borne sound transfer function, the quality of the condition monitoring can be significantly improved, since the condition monitoring model can be trained directly on the properties of the excitation signal of the component to be monitored. Furthermore, such a condition monitoring model can be transferred in an improved manner to structurally dynamically different systems/machines with components of the same type to be monitored. The virtual excitation signal can also be calculated if the machine system is changed or if the monitoring structure-borne sound sensor is located at a different location using only a newly determined structure-borne sound transfer function, so that the condition monitoring model can continue to be used unchanged and is only adapted to the design and type of component to be monitored. Retraining the data-based condition monitoring model is therefore not absolutely necessary if the mechanical structure of the machine changes, which means that test bench measurements and the associated costs can be saved.

The use of the above method with the prior determination of the virtual excitation signal has the advantage that a trained condition monitoring model can be used to monitor the component in question, regardless of the structure of the machine in which the component is used. It is only necessary to carry out a new determination of the structure-borne sound transfer function of this modified machine for a new machine.

According to a further aspect, a device, in particular a control device, is provided which is designed to carry out the above method.

Short description of the drawings

• 1 eine schematische Darstellung einer Maschine mit einer zu überwachenden Komponente in Form eines Wälzlagers und eines Körperschallsensors, der entfernt zu der überwachenden Komponente angeordnet ist; und
• 2 ein Flussdiagramm zur Veranschaulichung eines Verfahrens zum Ermitteln eines Zustands eines Wälzlagers in der Maschine der 1.

Embodiments are explained in more detail below with reference to the attached drawings. They show:

• 1 a schematic representation of a machine with a component to be monitored in the form of a rolling bearing and a structure-borne sound sensor which is arranged remotely from the component to be monitored; and
• 2 a flow chart illustrating a method for determining a condition of a rolling bearing in the machine of 1 .

Description of embodiments

1 shows a schematic representation of a machine 1 with a rolling bearing as the component 2 to be monitored, which is in contact with a rotating machine element 3, in particular a drive shaft of a drive motor 6 mounted in the rolling bearing. The machine 1 also has a structure-borne sound sensor 4, which is arranged on a further component 5 of the machine 1. The further component 5 can be a frame that holds the drive motor 6 and the rolling bearing.

During operation, the rolling bearing 2 generates structure-borne noise, which is transmitted via a transfer path to the other component 5 and can be measured there by the structure-borne noise sensor 4.

The structure-borne sound sensor 4 can be designed in the form of an acceleration sensor, such as a piezo sensor or a micromechanical acceleration sensor or the like. The structure-borne sound sensor 4 supplies acceleration signals as measurement signals, which can optionally be converted into force signals for the corresponding force excitations. The measurement signals correspond to temporal progressions of the corresponding measurement variable.

For condition monitoring, a method is carried out in a control unit 10, which can also control functions of the machine 1, which is implemented there in the form of software and/or hardware.

The procedure is explained below using the flow chart of the 2 described in more detail.

In step S1, a temporal progression of the measured variable is first recorded as a measurement signal by the body personic sensor 4 for a predetermined period of time. The recording of the temporal course of the measuring signal preferably takes place at a constant speed, by which the component to be monitored is excited.

Using a structure-borne sound transfer function, the temporal progression of the measurement signal is converted into a virtual excitation signal in step S2. The structure-borne sound transfer function characterizes the transfer function or the transformation of the structure-borne sound from the location of the component 3 to be monitored to the location of the monitoring structure-borne sound sensor 4.

In step S3, feature extraction is carried out from the temporal profile of the excitation signal (transformed measurement signal). The excitation features obtained in this way can be evaluated in the time and frequency domain based on an envelope analysis, an empirical mode decomposition (EMD), a Hilbert-Huang transformation (HHT), a wavelet transformation and a Fourier transformation (in particular Fast Fourier Transform, FFT). Statistical characteristics can be determined as features in the time domain, e.g. root mean square, standard deviation, skewness, kurtosis, peak-to-peak and the like, as well as statistical characteristics in the frequency domain, e.g. as mean values or statistical moments of higher order.

The excitation characteristics derived from the excitation signal as well as one or more operating variable profiles, such as the speed and/or the temperature, are fed as an input variable vector to a data-based condition monitoring model in step S4. The condition monitoring model is trained to assign the excitation characteristics and the speed and/or temperature to a state variable. The state variable indicates the functionality of the rolling bearing as the component 2 to be monitored. For example, the state variable can be defined in the form of a remaining service life, a degree of functionality (in %) or the like.

The evaluation of the data-based condition monitoring model leads to the determination of the state variable, which in step S5 can form the basis of an instruction to a user of the machine. In particular, the state variable is signaled to the user of the machine so that he or she can carry out maintenance or replace the component to be monitored. Alternatively, the operating mode of the machine can be set according to the state variable. For example, in the case of a state variable that indicates incipient wear of component 2, the maximum speed of the drive shaft as a rotating machine element 3 can be limited.

A simulation or an experimental analysis of the transmission behavior can be used to determine the structure-borne sound transmission function. A combination of simulation and experimental methods is also conceivable. The latter is particularly useful if the position of the virtual sensor intended for operation is not experimentally accessible (e.g. due to installation space restrictions). In this case, a partial validation of the simulation can be carried out by measuring the transmission function at an accessible sensor position close to the component to be monitored.

To train the data-based condition monitoring model, in particular in the form of an artificial deep neural network, Gaussian process model or the like, training data sets are provided based on real monitoring of machines with identical components to be monitored, for which the condition monitoring model is to be determined. In particular, curves of measurement signals from machines in which the component to be monitored is already heavily worn or has failed are determined in order to create corresponding training data sets. The training data sets map one or more of the excitation characteristics listed above and, if applicable, other operating variables, such as a speed, a temperature and the like, onto a label in the form of a state variable. The state variable can, for example, represent a percentage measure of functionality or a remaining service life. For example, a defective component to be monitored can receive a label of 0, while a properly functioning or as good as new component can receive a label of 1.

To create a training data set with respect to a failed component to be monitored, the temporal course of the measurement signal for a predetermined period of time immediately before the failure of the component to be monitored can be transformed into a corresponding excitation signal and the corresponding features for the input variable vector of the training data set can be formed from this.

The training of the data-based condition monitoring model can be done in a conventional manner using gradient-based methods such as backpropagation and the like.

Claims (10)

Method, in particular an at least partially computer-implemented method, for monitoring the condition of a component (2) in a machine (1) with at least one rotating machine element (3), with the following steps: - detecting (S1) a temporal progression of a measured variable of a structure-borne sound sensor (4) as a measurement signal; - converting (S2) the temporal progression of the measurement signal into an excitation signal using a predetermined structure-borne sound transfer function, wherein the structure-borne sound transfer function describes a transfer function for the structure-borne sound from the location of the component (2) to be monitored to the location of the monitoring structure-borne sound sensor (4), - evaluating (S4) the excitation signal and in particular at least one operating variable using a data-based condition monitoring model, wherein the condition monitoring model is trained to assign a state variable depending on the excitation signal and in particular the at least one operating variable, wherein the state variable indicates the functionality of the component to be monitored; - signaling (S5) the state variable. Procedure according to Claim 1 , with the further steps: - carrying out a feature extraction from the temporal course of the excitation signal in order to obtain one or more excitation features; - evaluating the one or more excitation features and in particular at least one operating variable as an input variable vector using the data-based condition monitoring model, wherein the condition monitoring model is trained to assign the excitation features and the at least one operating variable to a state variable. Procedure according to Claim 2 , where the excitation features are evaluated based on the excitation signal using an envelope analysis, an empirical mode decomposition (EMD), a Hilbert-Huang transform (HHT), a wavelet transform and a Fourier transform (in particular Fast Fourier Transform, FFT) in the time and frequency domain. Method according to one of the Claims 1 until 3 , wherein the signaling of the state variables comprises issuing an instruction to a user of the machine (1) and/or adapting an operating mode of the machine according to the state variable. Method according to one of the Claims 1 until 4 , wherein the detection of the temporal course of the measuring signal is preferably carried out at a constant operating state, in particular a constant speed of the at least one rotating machine element (3) by which the component (2) to be monitored is excited. Method according to one of the Claims 1 until 5 , where the data-based condition monitoring model comprises an artificial deep neural network or a Gaussian process model. Method according to one of the Claims 1 until 6 , wherein the predetermined structure-borne sound transmission function is determined by a simulation of a sound transmission path between the location of the component to be monitored (2) and the structure-borne sound sensor (4). Device, in particular control device (10), which is designed to carry out a method according to one of the Claims 1 until 7 to execute. Computer program product comprising instructions which, when the program is executed by at least one data processing device, cause the device to carry out the steps of the method according to one of the Claims 1 until 7 to execute. Machine-readable storage medium comprising instructions which, when executed by at least one data processing device, cause the device to carry out the steps of the method according to one of the Claims 1 until 7 to execute.

Images