Method and sensor arrangement for monitoring a function of a component of a machine

Abstract

The invention relates firstly to a method for monitoring the function of at least one component of a machine that is either in or mounted. The operating state of the component is continuously acquired using a wireless sensor node (01), thereby obtaining state measurements. These state measurements are recorded in the wireless sensor node (01) during an initialization phase in order to derive a threshold value (ψ). schwell ) to determine which is suitable for identifying state measurements associated with an irregular state. During an observation phase, the current threshold (ψ) serves as the basis. schwell ) to detect such condition measurements, which results in a warning being transmitted wirelessly from the radio sensor node (01) to a central monitoring device. It is checked whether the warning was caused by a faulty function of the component, which then leads to a false alarm message and the determination of a new threshold value (ψ). schwell) leads to. In an application phase, the current threshold (ψ) serves as the basis. schwell ) for detecting state measurement values ​​associated with an irregular condition. The invention also relates to a sensor arrangement for monitoring a function of at least one component of a machine.

Description

[0001] The present invention relates to a method and a sensor arrangement for monitoring the function of at least one mounted or supported component of a machine. The machine comprises at least one radio sensor node for detecting the operating state of the component.

[0002] DE 10 2012 015 654 A1 describes a method for signal-based detection and monitoring of condition parameters of at least one assembly, in particular of at least one bearing or slewing ring. For example, by detecting oscillations, vibrations, or noises in a bearing, conclusions can be drawn about the current conditions inside the bearing.

[0003] DE 10 2009 038 011 A1 discloses a method for automatically detecting and recognizing faults in a balancing machine during operation. In this method, an unbalanced rotor is mounted in a vibrating bearing within the balancing machine and set in rotation by a drive. The rotational frequency of the rotating rotor and the vibrations excited by the rotor are measured. For evaluation, mathematical models of the dynamic properties of the balancing machine are used, which are determined using, among other things, singular value decomposition.

[0004] EP 0 895 197 A1 discloses a method for monitoring systems with mechanical components. Measurement values ​​are recorded at predefined intervals for a defined set of state variables. The measurement values ​​recorded for various operating points during a modeling phase are used to create a model for the operating behavior of the components. This model is used to determine at least one monitoring variable at predefined intervals, which is independent of the current operating point. The temporal progression of the monitoring variable is used to assess wear in the components. A singular value decomposition can be used to determine the model parameters.

[0005] US 11,624,731 B2 discloses a method for predicting the remaining service life of a tool on a numerically controlled machine. In this method, indirect measurement indicators of the tool are selected based on the monitoring and analysis of the current condition of the tool. From this, a prediction model for the remaining service life of the tool is created, for which noise reduction and feature extraction are performed.

[0006] US Pat. No. 10,569,005 B2 discloses a monitoring device designed to detect an interruption in a fluid connection between two fluids using a pressure sensor. The pressure sensor detects first pulses from a system containing the first fluid and second pulses from a system containing the second fluid. A monitoring device receives signals from the pressure sensor and calculates eigenvectors and eigenvalues ​​from them using a source separation algorithm.

[0007] US 2011 / 060703 A1 describes a method for detecting correlations in a data set. For this purpose, one or more parameters are determined that are assigned to one or more data sets. The parameters are determined at runtime to generate one or more test data sets from the data sets. A test data set comprises one or more objects and one or more indices. The objects are assigned to the indices. The method further comprises calculating one or more correlation coefficients that are assigned to the test data sets.

[0008] CN 115329868 A teaches a method with dynamic variables based on the local reservation of parallel projections. First, a low-rank vector autoregression model is constructed. A quality correlation analysis is performed to calculate a score vector matrix. A parallel projection is performed on a previously obtained vector matrix, with the input and output data spaces simultaneously projected onto five subspaces. For error detection, error statistics in the five subspaces and corresponding error confidence bounds are generated.

[0009] US 2011 / 0060703 A1 describes a method for detecting correlations in a data set. One or more parameters associated with one or more data sets are determined. The one or more parameters are determined at runtime to generate one or more test data sets from the one or more data sets. Test data of the one or more test data sets comprise one or more objects and one or more indices. The one or more objects are associated with the one or more indices. The method further comprises calculating one or more correlation coefficients associated with the one or more test data sets.

[0010] DE 10 2022 203 195 A1 discloses a method for diagnosing the condition of a machine with rotating and / or oscillating motion sequences. A temporal profile of a vibration signal with multiple superimposed vibration frequencies is recorded. Amplitude transitions of the vibration signal through a specified signal level are recorded for a plurality of measurement intervals of specified length. One or more parameters of the determined amplitude transitions are extracted for each of the specified measurement intervals. A condition diagnosis for the machine is derived from a temporal profile of the one or more extracted parameters.

[0011] DE 10 2012 015 654 A1 discloses a condition detection and monitoring system for the temporary, value-based, signal-based, or data-based recording and monitoring of condition parameters of a bearing component. A contact sensor is attached to or incorporated into this bearing component.

[0012] EP 3 282 399 A1 describes a method for detecting a process anomaly in a technical plant. A self-organizing map is trained using historical process data as the plant's good states. The good states are used to determine the temporal order or path of the affected nodes, as well as the tolerances of the neuron hits. Thresholds for the Euclidean distance for the good states are determined and stored. The current process data of the plant are evaluated in the form of a state vector using the trained self-organizing map. The Euclidean distance of the current state vector from the affected neuron is checked for exceedance of the threshold.

[0013] EP 1 892 597 A1 discloses a method for monitoring technical processes, in which influences due to wear and tear and external influences are to be differentiated. Measurement data is acquired via measurement channels. The measurement data from one measurement channel is processed into several time series of state values ​​with different time bases. In a learning phase, limit value ranges are determined. In a monitoring phase, monitoring-related steps are triggered when limit values ​​are exceeded. In the monitoring phase, a good value range of the state values ​​for a time series is calculated with an upper and a lower limit value, taking into account a limit value range predetermined in the learning phase and a state base value of the time series determined in the monitoring phase.

[0014] One problem with state-of-the-art sensors, which record the operating status of a machine component and transmit the resulting measured values ​​wirelessly to a central monitoring device, is their power supply. Since these sensors are generally not wired, they require their own power supply, for example, in the form of an electric battery. Wirelessly transmitting the measured values ​​to the central monitoring device shortens the service life and / or maintenance intervals for the sensor. If the sensors are equipped with complex circuit logic or even their own microprocessor, the energy requirement increases further, necessitating either a wired power supply or short maintenance intervals.

[0015] Based on the prior art, the object of the present invention is to significantly reduce the power consumption of sensors that detect the operating status of a machine component that is in or stored and transmit the resulting measured values ​​wirelessly to a central monitoring device, thereby extending the service life and maintenance intervals of the sensor. Nevertheless, all necessary measured values ​​should be able to be transmitted securely to a remote unit.

[0016] This object is achieved by a method according to the appended claim 1 and by a sensor arrangement according to the appended independent claim 8. Preferred embodiments are mentioned in the following description and in the subclaims.

[0017] The method according to the invention serves to monitor at least one function of at least one bearing or mounted component of a machine. The machine can be, for example, a machine tool, a vehicle or a production machine. The bearing component is in particular a rotary or linear bearing. The bearing is preferably formed by a rolling bearing or a plain bearing. The function of the bearing is to mount one machine part of the machine relative to another machine part of the machine. The mounted component is in particular a component that is directly or indirectly mounted by a bearing, such as a drive, a shaft, an axle, a wheel or a transmission element such as a gear. The respective function of the bearing or mounted componentThe bearing capacity of a supported component is defined by parameters such as a maximum permissible rotational speed, a maximum permissible transverse force, a maximum permissible acceleration, a maximum permissible vibration or a maximum permissible clearance fit.

[0018] The machine comprises at least one radio sensor node for detecting an operating state of the at least one component. The operating state is detected by the radio sensor node enabling the measurement of at least one physical quantity; for example, a speed, a rotational speed, an acceleration, a force, a torque, a sound pressure, or a temperature. For this purpose, the at least one radio sensor node preferably comprises a sensor component. The at least one radio sensor node preferably comprises a computing unit for processing measurement data from the sensor component. The computing unit is preferably formed by a microcontroller. The at least one radio sensor node preferably also comprises a radio component for transmitting measurement data wirelessly. The at least one radio sensor node is preferably designed for bidirectional data transmission wirelessly.The radio sensor node preferably comprises an integrated power supply; for example, a battery. The machine preferably comprises at least two of the radio sensor nodes for detecting multiple operating states of one component or for detecting operating states of multiple components of the machine.

[0019] The machine is also assigned a central monitoring device for communicating with the at least one radio sensor node. The at least one radio sensor node and the central monitoring device are configured for wireless data transmission between the at least one radio sensor node and the central monitoring device. The central monitoring device serves to centrally monitor the function of the at least one component of the machine. The central monitoring device preferably forms a component of the machine.

[0020] According to the method, the operating state of the at least one component is continuously recorded during operation of the machine using the at least one radio sensor node, thereby obtaining condition measurements. The condition measurements are initially available in the respective radio sensor node. The operating state is preferably recorded regularly during operation of the machine, with preferably at least one condition measurement per second and more preferably at least 1,000 to 100,000 condition measurements per second being obtained.

[0021] The process runs through several phases, beginning with an initialization phase, which is specifically designed to determine the proper condition of the component or machine. It is therefore assumed that the component is in proper condition at the beginning. The initialization phase takes place during regular operation of the machine. During this initialization phase, the recorded condition measurements are recorded in the respective wireless sensor node in order to derive a threshold value Ψ from the recorded condition measurements as a result of the initialization phase. schwell to determine which is suitable for identifying condition measurements associated with an irregular condition of the component. This threshold forms a current threshold Ψ schwellFor this threshold determination, it is assumed that the component is in a normal state during the initialization phase, operating as intended and not yet exhibiting any significant wear. These steps are performed in the respective wireless sensor node.

[0022] The initialization phase is followed by an observation phase, which typically lasts (significantly) longer than the initialization phase and in which irregularities are detected and verified. The observation phase takes place during regular machine operation. During the observation phase, the recorded condition measurements continue to be recorded in the respective wireless sensor node. The current threshold value Ψ schwellis used to detect state measurements associated with an irregular condition from the state measurements recorded during the observation phase, which is carried out in the respective wireless sensor node. In a case where such state measurements associated with an irregular condition are detected, a warning is transmitted wirelessly from the respective wireless sensor node to the central monitoring device. Since immediately after the initialization phase, the current threshold value Ψ schwell is based solely on an evaluation in the respective radio sensor node, the certainty with which the current threshold value Ψ schwellidentified condition measurements are actually caused by an irregular condition of the component is not yet very high. Therefore, a check is carried out to determine whether the warning transmitted to the central monitoring device was actually caused by a faulty function of the component. This check is carried out in the central monitoring device or by an operator of the central monitoring device. In particular, it may happen that the warning was not caused by a faulty function of the component. For example, external influences on the machine or other phenomena in the machine may have led to the condition measurements recorded at that time being evaluated by the respective radio sensor node using the current threshold value Ψ schwellwere associated with an irregular condition. The central monitoring device preferably has known the status measurements from other radio sensor nodes and / or measurements from additional internal or external sensors, which are used for the described check. Alternatively or additionally, this check is carried out by a machine operator. In a case in which the check has shown that the transmitted warning was not caused by a faulty function of the component, a false warning message is transmitted from the central monitoring device to the respective radio sensor node. This is preferably done via radio. The false warning message is therefore transmitted to signal to the respective radio sensor node that the warning previously transmitted by it was not due to a malfunction of the component monitored by this radio sensor node.

[0023] As long as the respective radio sensor node does not receive a false warning message, it is assumed that the current threshold Ψ schwell is sufficiently suitable to detect an irregular condition of the component, so that the current threshold value Ψ schwell is not updated and continues to be applied. However, in the case of a false warning message, a new threshold Ψ schwell determined from the last recorded condition measurements. The new threshold value Ψ schwell is in turn suitable for identifying state measurements associated with an irregular state and now forms the current threshold value Ψ schwell . The new threshold Ψ schwell replaces the current threshold value Ψ schwell and now forms the current threshold value Ψ schwell . Determining the new threshold value Ψ schwellis carried out during the observation phase in the respective radio sensor node. Since the new threshold Ψ schwell The result of the check on the central monitoring device is the certainty with which the new threshold value Ψ schwell identified condition measurements are actually caused by an irregular condition of the component. This can lead to multiple updates of the threshold value Ψ schwell during the observation phase. It should be noted that the threshold value can be composed of several individual values ​​or include several parameters from, for example, additional sensors in order to represent predetermined permissible or error-typical signal curves.

[0024] The observation phase is followed by an application phase, which covers most of the machine's operation time and can last several years. During the application phase, the wireless sensor node operates largely autonomously. During the application phase, the recorded condition measurements continue to be recorded in the respective wireless sensor node. The current threshold value Ψ schwellis used to identify condition measurements associated with an irregular condition from the condition measurements recorded during the application phase. In the event that such condition measurements associated with an irregular condition are detected, a warning is preferably transmitted wirelessly from the respective wireless sensor node to the central monitoring device. The condition measurements identified in this way indicate with a high probability a faulty function of the component or a faulty function of the component that is imminent due to wear.

[0025] A particular advantage of the method is that mainly only warnings need to be transmitted wirelessly from the wireless sensor node to the central monitoring device, so that the transmission of status measurement data is rarely or hardly necessary. This significantly reduces wireless data transmission, saving electrical energy for wireless data transmission and, for example, a battery in the respective wireless sensor node lasts significantly longer and the wireless sensor node can be operated for a significantly longer time before it needs to be replaced or serviced. In addition, only a small amount of data needs to be stored in the wireless sensor node, so that the memory in the wireless sensor node can be kept small. Since warnings from the wireless sensor node are checked at the central monitoring device and, in the event of false warnings, the threshold value Ψ schwellThis allows for a high degree of confidence that future warnings from the wireless sensor node are actually caused by a component malfunction or by a component malfunction that is becoming apparent due to wear and tear. Data sequences that do not exceed the threshold over a predetermined period of time can be deleted from the wireless sensor node, keeping the amount of data to be stored small.

[0026] The step of determining the threshold value Ψ to be carried out during the initialization phase in the respective radio sensor node schwellfrom the recorded state measurements comprises several sub-steps. In one sub-step, m time periods are selected from the initialization phase, whereby these m time periods are preferably of equal length and can follow one another directly or indirectly. In each of these m time periods, several sub-steps are carried out, so that m iterations take place. For an i-th time period of the m time periods, the following several sub-steps are carried out. First, an i-th state measurement value matrix is ​​formed from at least a selection of the state measurements recorded in the i-th time period. The state measurement value matrix comprises as elements at least the selection of the state measurements recorded in the i-th time period. Subsequently, a singular value decomposition of the i-th state measurement value matrix is ​​determined, whereby, among other things, singular values ​​of the i-th state measurement value matrix are obtained. From these singular values, the largest singular value σH of the i-th state measurement matrix, whereby an i-th largest singular value σ Hi After these sub-steps have been carried out for each of the m consecutive time periods, there are m largest singular values ​​σ H1 to σ Hm He will now have a standard deviation ψ m of these m largest singular values ​​σ H1 to σ Hm The determined standard deviation ψ m is evaluated with a predefined factor t, whereby the threshold Ψ schwell is obtained. The determined standard deviation ψ is preferred. m multiplied by the factor t, which is greater than one, whereby the threshold Ψ schwell is received.

[0027] The i-th state measurement matrix comprises as elements at least the selection of the state measurements recorded in the i-th time period, wherein the state measurements are divided into n parts, each forming a row of the state measurement matrix. The n parts of the state measurements can be recorded directly one after the other. Alternatively, n time sub-periods can be selected from each of the i-th time periods, which directly or indirectly follow one another. The n parts of the state measurements are each recorded in one of the n time sub-periods.

[0028] In preferred embodiments, only the last m largest singular values ​​σ Hi stored in the respective wireless sensor node, so that this memory can be kept small. After determining a new largest singular value σ Hm+1 becomes the oldest of the largest singular values ​​σ H1 overwritten in memory to obtain the new largest singular value σHm+1 to be stored in memory. Such new largest singular values ​​are determined repeatedly, particularly during the observation phase and the application phase.

[0029] In preferred embodiments, the step of applying the respective current threshold value Ψ to be carried out during the observation phase or during the application phase in the respective radio sensor node comprises schwellto detect state measurements associated with an irregular state from the state measurements recorded in the observation phase or in the application phase, several sub-steps are used. In one sub-step, a state measurement matrix is ​​formed from at least a selection of the state measurements recorded in a current time period of the observation phase or the application phase. The state measurement matrix comprises as elements at least the selection of the state measurements selected in this time period. Preferably, the state measurement matrix is ​​formed from all of the state measurements recorded in this current time period of the observation phase or the application phase. Subsequently, a singular value decomposition of the state measurement matrix is ​​determined, whereby, among other things, singular values ​​of the state measurement matrix are obtained. From these singular values, the largest singular value σ H of the state measurement matrix. The largest singular value σH is calculated with the current threshold value Ψ schwell , which determines the state measurements associated with an irregular state. The largest singular value σ H is assigned to a state measurement value associated with an irregular state if it is greater than the current threshold value Ψ schwell is.

[0030] In particularly preferred embodiments, the state measurement matrices formed in the respective radio sensor node are not only used to determine the threshold value Ψ schwellor the state measurements associated with an irregular condition, but also for a data-reduced transmission of the state measurements from the respective wireless sensor node to the central monitoring device. The transmission of one of the data-reduced state measurement matrices comprises the following substeps, which are carried out in the respective wireless sensor node during the observation phase or in the application phase. The k largest singular values ​​σ H of the data-reduced state measurement matrix to be transmitted, where k is a natural number greater than zero and less than or equal to three. Particularly preferably, k is equal to 2. Furthermore, the k largest singular values ​​σ HThe associated singular vectors u, v are determined. These are the left and right singular vectors, respectively. The singular vectors u, v are transformed into a frequency domain, resulting in transformed singular vectors. The transformation is preferably a Fourier transformation, in particular a fast Fourier transformation (FFT). These steps significantly reduce the data volume compared to the original state measurement matrix. Now, only the k largest singular values ​​and the associated transformed singular vectors are transmitted wirelessly from the respective wireless sensor node to the central monitoring device, from which the state measurements can be reconstructed in the central monitoring device. The reconstructed state measurements differ only insignificantly from the original state measurements.However, the amount of data to be transmitted for this purpose is significantly reduced, so that electrical energy for wireless data transmission is saved and, for example, a battery in the respective wireless sensor node lasts significantly longer and the wireless sensor node can be operated for a significantly longer time before it needs to be replaced or serviced.

[0031] Insofar as the k largest singular values ​​and the corresponding transformed singular vectors are transmitted by radio from the respective radio sensor node to the central monitoring device, the central monitoring device preferably reconstructs the state measurement values ​​from the respectively transmitted k largest singular values ​​and the respective transformed singular vectors.

[0032] In preferred embodiments, the reconstructed condition measurement values ​​are evaluated in the central monitoring device to determine whether they correspond to a proper condition of the machine. This preferably occurs when a warning has been transmitted from the respective radio sensor node to the central monitoring device, whereby it is checked whether the transmitted warning was caused by a faulty function of the component. First, the condition measurement values ​​are reconstructed from the k largest singular values ​​transmitted in each case and the respective transformed singular vectors. This occurs in the central monitoring device. As a result, reconstructed condition measurement values ​​are obtained. A check is then carried out to determine whether the reconstructed condition measurement values ​​correspond to a proper condition of the machine, so that the proper condition of the machine or the improper condition of the machine is determined.This check determines whether the reconstructed condition measurements are due to the machine's proper condition or are attributable to malfunctioning or increasing wear of the component. This check is performed in the central monitoring device or by an operator of the central monitoring device. In this respect, this check can be performed mechanically and / or manually.

[0033] If the proper condition of the machine has been determined, the proper condition of the machine is reported from the central monitoring device to the respective radio sensor node. This is preferably done in the form of a false warning message. If such a message has been reported to the respective radio sensor node, the respective condition measurements previously transmitted from this radio sensor node to the central monitoring device are stored in this radio sensor node. Otherwise, these condition measurements are preferably deleted to save memory in the radio sensor node. The stored condition measurements are used in this radio sensor node to determine the new threshold value Ψ schwell used. The current threshold value Ψ schwell allows the determination of condition measurements that are actually caused by an irregular condition of the component with a high degree of certainty, since the current threshold value Ψ schwellbased on a check of the proper condition of the machine.

[0034] Preferred embodiments allow predictive maintenance of the machine. For this purpose, the method comprises the following further steps, which are to be carried out in the respective radio sensor node. A correlation coefficient is determined between a newly detected condition measurement value associated with an irregular condition and the previously recorded condition measurements. This correlation coefficient is preferably determined by a canonical correlation analysis. If the correlation coefficient is smaller than a predefined correlation threshold, this indicates increasing wear of the part or other components of the machine, which may become critical for the operation of the machine, so that a warning is issued. In response to this warning, the machine can be repaired. The correlation threshold is preferably between 0.01 and 0.99; more preferably between 0.3 and 0.5; and most preferably equal to 0.4.

[0035] The determination of a correlation coefficient is the result of a newly detected measured value associated with an irregular condition. In preferred embodiments, a check is performed to determine whether this correlation coefficient is caused by a faulty function of the component. This check is performed in the central monitoring device or by an operator of the central monitoring device. For this purpose, the currently determined correlation coefficient is first transmitted from the respective radio sensor node to the central monitoring device, preferably via radio. The central monitoring device determines whether the transmitted correlation coefficient is caused by a faulty function of the component.In a case where it is determined that the transmitted correlation coefficient is not due to a faulty function of the component, a false warning message is transmitted from the central monitoring device to the respective wireless sensor node, preferably via radio. In the case of a false warning message, the condition measurements underlying the transmitted correlation coefficient are used to determine the new threshold value Ψ. schwell used, which takes place in the respective wireless sensor node. This new threshold Ψ schwell then forms the current threshold value Ψ schwell . Due to the false warning message, it can be assumed that the condition measurements underlying the transmitted correlation coefficient indicate a proper condition of the machine or component, so that they can be used to determine a better threshold value Ψ schwell are suitable.

[0036] The sensor arrangement according to the invention serves to monitor a function of at least one component of a machine. The sensor arrangement comprises at least one radio sensor node for detecting an operating state of the component and a central monitoring device. The at least one radio sensor node and the central monitoring device are configured to carry out the method according to the invention. The at least one radio sensor node and the central monitoring device are preferably configured to carry out one of the described preferred embodiments of the method according to the invention. The sensor arrangement preferably also comprises features that are described in connection with the method according to the invention.

[0037] Further advantages, details, and modifications of the invention are described below using preferred embodiments, with reference to the drawings. Fig. 1 is a diagram of a state measurement value over time, which is recorded according to a preferred embodiment of a method according to the invention; Fig. 2 a diagram illustrating a largest singular value for state measurements determined according to the method according to the invention; Fig. 3 a diagram illustrating state measurement values ​​reconstructed according to the method according to the invention; Fig. 4 shows a first diagram illustrating correlation coefficients determined according to the method according to the invention; Fig. 5 a second diagram showing correlation coefficients determined according to the method according to the invention; Fig. 6 shows a flow chart of an initialization phase of a preferred embodiment of the method according to the invention; Fig. 7 shows a flow chart of an observation phase of a preferred embodiment of the method according to the invention; and Fig. 8 shows a flow chart of an application phase of a preferred embodiment of the method according to the invention.

[0038] Fig. 1 shows a diagram of a condition measurement value over time, which is recorded by way of example according to a preferred embodiment of a method according to the invention. Time is plotted in minutes on the abscissa axis. An acceleration in meters per second squared is plotted on the ordinate axis. This is, for example, an acceleration acting on a component (not shown) of a machine (not shown), for example a bearing. In this example, 2,559 measured values ​​were recorded every 1 / 10 of a second. The diagram shows large amplitudes in the range between 25 and 45 minutes as well as increasing amplitudes in the range between 70 and 200 minutes. In fact, the large amplitudes in the range between 25 and 45 minutes are due to external influences that acted on the machine (not shown), but the machine was functioning properly.However, the increasing amplitudes in the range between 70 and 200 minutes are due to increasing wear of the bearing, which is identified according to the invention.

[0039] According to the invention, the state measurements are stored in a radio sensor node 01 (shown in Fig. 6). To reduce the amount of data, subsampling can be performed, for which the data is only collected in selected time periods. For example, 32,768 state measurements can be collected in such a time period of only 1.28 seconds, after which no state measurements are collected in a subsequent time period of 58.72 seconds. The state measurements are denoted as x i Subsampling is performed to divide the state measurements into equal length sections, for example as [x i1 , x i2 , ... x i16 ] where each element contains 2048 state measurements. The state measurements xi are arranged in a two-dimensional state measurement matrix D, for example in the form: D=[xi1xi2xi3⋅⋅⋅xin]with D∈ℤn×2048

[0040] The relationships and sizes given as examples must be adapted to the respective application.

[0041] In an initialization phase of the procedure, a threshold value Ψ schwell to identify condition measurements associated with an irregular condition.

[0042] For this purpose, a singular value decomposition SVD is performed for each of the determined state measurement matrices D: U,S,V=SVD(D)

[0043] Where U and V each represent a unitary matrix and S a diagonal matrix containing the singular values. For each of the state measurement matrices D, the largest singular value σ H These largest singular values ​​σ H can be expressed as a vector Σ = [σ H1 , σH2 ,.. σ Hm ] For the vector Σ, the standard deviation ψ m certainly: ψm=std(∑)

[0044] The determined standard deviation ψ m is evaluated with a predefined factor t > 1, whereby the threshold Ψ schwell = t. ψ m is obtained and this has a first current threshold value Ψ schwell State measurements whose associated largest singular value σ H greater than the current threshold value Ψ schwell are associated with an irregular state by the radio sensor node 01 in an observation phase following the initialization phase, whereupon a warning is transmitted from the radio sensor node 01 by radio to the central monitoring device.

[0045] According to a preferred option, the state measurement values ​​are transmitted from the radio sensor node 01 (shown in Fig. 6) via radio to the central monitoring device, for which the data volume is reduced. This transmission preferably only occurs upon request from an operator of the central monitoring device. This transmission takes advantage of the fact that a singular value decomposition has already been performed. For one of the state measurement matrices D, only the two largest singular values ​​σ1, σ2 and the fast Fourier transforms of the singular vectors u1, u2, v1, v2 assigned to the two largest singular values ​​σ1, σ2 are transmitted. [⋮⋮⋮⋮u1u2⋯un⋮⋮⋮⋮][σ1⋯⋯⋯⋯σ2⋯⋯⋮⋮⋱ ⋮⋯⋯⋯σn][⋯v1⋯⋯v2⋯⋮⋮⋮⋯vn⋮]=SVD(D)

[0046] Another preferred option is to evaluate the condition measurements in the central monitoring device. If these condition measurements are assigned by the central monitoring device to indicate proper machine function, this can then be reported to the respective radio sensor node 01 in the form of a false warning message. The affected condition measurements are then stored in this radio sensor node 01 and used to determine the next current threshold value.

[0047] Fig. 2 shows a diagram illustrating the largest singular values ​​σ H for the condition measurements x i , the determination of which with reference to Fig. 1. The abscissa axis represents a number x of steps. The ordinate axis represents the largest singular values ​​σ H for the recorded condition measurements x i The current threshold value Ψ schwellis represented by a dashed line 02. The largest singular values ​​σ shown in the diagram above line 02 H are those condition measurements x i which are now associated with an irregular state. The largest singular values ​​σ shown in the diagram below line 02 H are those condition measurements x i which are associated with a regular state.

[0048] The data received from the wireless sensor node 01 (shown in Fig. 6) The state measurements transmitted to the central monitoring device are transmitted in a data-reduced form according to the above description of a preferred option, so that they must first be reconstructed after receipt by the central monitoring device. For this purpose, an inverse Fast Fourier transform of the transmitted transformed singular vectors u1, u2, v1, v2 must be determined, which, together with the transmitted two largest singular values ​​σ1, σ2, is used to reconstruct the singular value-decomposed state measurement matrix. From this, the state measurement matrix can be reconstructed, the elements of which are the reconstructed state measurements.

[0049] Fig. 3 shows a diagram showing the reconstructed state measurements x i , whose reconstruction with reference to Fig. 2. The axes are the same as the axes of the Fig. 1. A comparison of the diagram shown in Fig. 3 shown diagram with the one in Fig. The diagram shown in Figure 1 makes it clear that the data reduction described above does not lead to a loss of relevant data.

[0050] Fig. Figure 4 shows a first diagram illustrating correlation coefficients determined in accordance with a preferred embodiment of the method according to the invention during the application phase. A number x of steps is plotted on the abscissa axis. The unitless correlation coefficient CCF is plotted on the ordinate axis. During the application phase, the radio sensor node 01 (shown in Fig. 6) further state measurements are recorded. According to the above explanations, state measurement matrices D are formed, and a singular value decomposition U, S, V = SVD(D) is performed for each of them. The largest singular value σ neuthe current singular value decomposed state measurement matrix is ​​compared with the current threshold value ψ schweıı Insofar as this largest singular value σ neu greater than the current threshold value Ψ schwell , the new largest singular value σ neu underlying state measurements are associated with an irregular state. A correlation coefficient (CCF) is then determined between these newly detected irregular state measurements and the previously recorded state measurements, for which a canonical correlation analysis is performed. The new largest singular value σ neu is a singular vector v fneu ∈ ℝ{ 1×xf The following matrix results: Vc=[Vfvfneu]

[0051] Next, a covariance matrix is ​​determined: C=Covariance(Vc)∈ℝ(m+1)×(m+1)

[0052] The covariance matrix has the structure: C=[Cmmc1mcm1c11]

[0053] The submatrices of the covariance matrix have the dimensions: Cmm=C[:m,:m]ℝm×m cm1=C[:m,:m]ℝm×1 c1m=C[m:, :m]ℝ1×m c11=C[m:, m:]ℝ1×1

[0054] Here, m: stands for the remaining elements after the m elements from the beginning; and : m stands for the first m elements from the beginning.

[0055] Subsequently, a scaling of the covariance matrix C mm made: Cmmmax=max(abs(Cmm)) Cmm=Cmm / Cmmmax

[0056] Subsequently, the covariance matrix C mm all diagonal elements which are below a limit of, for example, 10 -6 are removed by setting them equal to zero.

[0057] It turns out: cf=Cmm−1 / 2⋅(Cm1⋅C11−1)⋅Cf∈ℝm×1 Subsequently, a singular value decomposition of c f made: u, sccf, v=SVD(cf)

[0058] Herein are u and S ccf scalar values ​​and v ∈ ℝ{ m×1 .

[0059] The correlation coefficient to be determined is: CCAcoef=sccf

[0060] The correlation coefficients determined in this way in the radio sensor node 01 are shown for an example in the Fig. The diagram shown in Figure 4 shows the following.

[0061] As long as the largest singular value σ neu not greater than the current threshold value Ψ schwell If the condition measurement values ​​are normal, the condition measurements are associated with a normal state, so that the warehouse is assumed to be in proper condition. For example, a green light is signaled like a traffic light.

[0062] As soon as the largest singular value σ neu greater than the current threshold value Ψ schwellIf the condition is irregular, the measured values ​​are associated with an irregular condition, and the correlation coefficient is determined as described above. A warning is also issued, such as a yellow light, similar to a traffic light.

[0063] As soon as the correlation coefficient is several times smaller than a correlation threshold, the bearing is assumed to be damaged or worn, so that, like a traffic light, a red light is signaled, and the operation of the bearing or machine must be interrupted. Maintenance or replacement of the bearing is now required. The correlation threshold is preferably 0.4, for example, and is Fig. 4 is illustrated by a line 03. Correlation coefficients of at least 0.4 indicate an operational condition of the bearing. Correlation coefficients less than 0.4 indicate an inoperative condition of the bearing.

[0064] Fig. Figure 5 shows a second diagram illustrating correlation coefficients which correspond to the results obtained with reference to Fig. 4 described preferred embodiment of the method according to the invention in the application phase. The second diagram differs from the one in Fig. 4 only insofar as it was recorded for a different exemplary bearing. Here, too, correlation coefficients less than 0.4 indicate a non-operational condition of the bearing.

[0065] Fig. Figure 6 shows a flowchart of the initialization phase of a preferred embodiment of the method according to the invention. The flowchart relates to the schematically illustrated radio sensor node 01, which comprises a sensor, a processor for data preprocessing, and a memory for storing regular features. In a method step referred to as "recording state measured values," the state measured values ​​x are recorded with the radio sensor node 01. i recorded as with reference to Fig. 1 above. In a process step called “data preprocessing”, the state measurements x i the state measurement matrices D, as with respect to Fig. 1 above. In a process step called “feature determination”, the state measurement matrices D are decomposed into singular values ​​and the largest singular value σ Hand the associated singular vector v is extracted, as described with reference to Fig. 1 above. The above steps are repeated m times, so that m time slices are acquired, where the number m is predetermined. In a process step called “storing regular features”, the m largest singular values ​​σ H1 , σ H2 ,... σ Hm and the m associated singular vectors v1, v2, ... v m stored in the memory of the radio sensor node 01, assuming that the bearing (not shown) is working properly in this initial initialization phase, so that the m largest singular values ​​σ H1 , σ H2 ,... σ Hm and the m associated singular vectors v1, v2,... v m are first associated with a regular state. In a process step called “determine and store threshold”, the m largest singular values ​​σ H1 , σH2 ,... σ Hm the threshold Ψ schwell determines how this is done with reference to Fig. 1 above. In an optional step called “Transfer information to central monitor device”, the m largest singular values ​​σ H1 , σ H2 ,... σ Hm and the m associated singular vectors v1, v2,... v m transmitted wirelessly from the wireless sensor node 01 to the central monitoring device (not shown).

[0066] Fig. 7 shows a flow chart of the observation phase of a preferred embodiment of the method according to the invention. The flow chart relates to the schematically illustrated radio sensor node 01, which comprises a sensor, a processor for data preprocessing, and a memory for storing regular features, wherein an update, i.e., an overwriting of the features stored there with newly determined features, takes place during the observation phase. The method step referred to as "recording state measured values," the method step referred to as "data preprocessing," and the method step referred to as "feature determination" are similar to the corresponding steps of the initialization phase, as described with reference to Fig. 6 above. In a decision called “greater than threshold?” it is decided whether a largest singular value σ resulting from the currently recorded state measurements Hgreater than the current threshold value Ψ schwell is, as with reference to Fig. 1 above. If this is the case, a warning is transmitted via radio to the central monitoring device (not shown) in a process step called "Warning to central monitoring device." In an optional process step called "Transmit information to central monitoring device," the m largest singular values ​​σ H1 , σ H2 ,... σ Hm and the m associated singular vectors v1, v2,... v mtransmitted wirelessly from the radio sensor node 01 to the central monitoring device, where they are used to reconstruct the condition measurements. In a decision referred to as "False warning?", a check is made in or on the central monitoring device to determine whether the transmitted warning was not caused by a malfunction of the bearing. "External information about the false warning" is used for this decision. For example, the central monitoring device may have information about external disturbances affecting the machine, from which it can be concluded that the transmitted warning is not caused by a malfunction of the bearing. In this case, a false warning has occurred, whereupon, in a process step referred to as "update of the features", the m largest singular values ​​σ stored in the memory of the radio sensor node 01 are H1 , σ H2 , ... σ Hm and the m associated singular vectors v1, v2,... v mare gradually overwritten with the current largest singular values ​​and associated singular vectors. If there is no false warning, the bearing must be repaired in a process step known as "waiting for repair". After the process step known as "updating the characteristics", a decision must be made as to whether the observation phase should be ended in a decision known as "end of observation phase?". If the observation phase is to be continued, the described process steps must be repeated, starting with the process step known as "recording measured condition values". If the observation phase is to be ended, the status known as "end of observation" is reached, in which the observation phase is concluded. The initialization phase and the observation phase are preferably run through when the bearing or the entire machine is put into operation for the first time.The initialization phase and the observation phase can last, for example, a few minutes, hours or days.

[0067] Fig. Figure 8 shows a flowchart of the application phase of a preferred embodiment of the method according to the invention. The flowchart refers to the schematically illustrated radio sensor node 01, which comprises a sensor, a processor for data preprocessing, and a memory for storing the threshold value Ψ schwell , the m largest singular values ​​σ H1 , σ H2 ,... σ Hm and the m associated singular vectors v1, v2,... v mon the basis of which state measurements associated with an irregular state are detected. The method is preferably designed to operate several of the radio sensor nodes 01. The method step referred to as "recording state measurements", the method step referred to as "data preprocessing" and the method step referred to as "feature determination" are similar to the corresponding steps of the initialization phase, as described with reference to Fig. 6 above. In a decision called “Warning active?” it is decided whether a warning as described in relation to Fig. 7 above is active. If this is the case, the correlation coefficient is determined in a process step referred to as “determining the correlation coefficient”, as described with reference to Fig. 4 above. Otherwise, a decision called “greater than threshold?” is made as to whether a largest singular value σ resulting from the currently recorded state measurements H greater than the current threshold value Ψ schwell is, as with reference to Fig.7 above. If this is the case, a warning is transmitted by radio to the central monitoring device (not shown) in a process step referred to as "warning to central monitoring device", and the process step referred to as "determining the correlation coefficient" follows. In a process step referred to as "transmitting the correlation coefficient", the determined correlation coefficient is transmitted by radio from the respective radio sensor node 01 to the central monitoring device. In a decision referred to as "false warning?", a check is carried out in or on the central monitoring device to determine whether the transmitted warning was not caused by a faulty function of the bearing. "External information about the false warning" is used for this decision.For example, the central monitoring device may have information about external disturbances affecting the machine, from which it can be concluded that the transmitted warning is not caused by a malfunction of the bearing. In this case, a false warning is present, after which, in a process step known as "feature update," the m largest singular values ​​σ stored in the memory of radio sensor node 01 are updated. H1 , σ H2 ,... σ Hm and the m associated singular vectors v1, v2,... v m are gradually overwritten with the current singular values ​​and assigned singular vectors and the threshold value Ψ schwelldetermined. The current warning is then deactivated in a process step known as "deactivating the warning." If there is no false warning, the central monitoring device knows that the bearing is malfunctioning or that wear on the bearing is impairing its function. Otherwise, the application phase continues. The application phase represents the main service life of the bearing and typically extends over several years. Reference symbol 01 Radio sensor node 02 Threshold 03 Correlation limit

Claims

[1] Method for monitoring a function of at least one stored or stored component of a machine, for which the machine has at least one radio sensor node (01) for detecting an operating state of the component and a central monitoring device is assigned to the machine, the method comprising the following steps: - continuously recording the operating state of the at least one component during operation of the machine with the at least one radio sensor node (01), whereby state measurement values ​​are obtained; - Recording the state measurement values ​​during an initialization phase in the radio sensor node (01) in order to determine a threshold value (Ψ schwell , 02) which is suitable for identifying state measurements associated with an irregular state and a current threshold value (Ψ schwell , 02) forms; - Recording the state measurements during an observation phase following the initialization phase, whereby the current threshold value (Ψ schwell , 02) is used to detect state measurements associated with an irregular condition from the state measurements recorded in the observation phase, wherein in a case in which state measurements associated with an irregular condition have been detected, a warning is transmitted from the respective radio sensor node (01) by radio to the central monitoring device; - Checking whether the transmitted warning was caused by a faulty function of the component, wherein in a case where the check has shown that the transmitted warning was not caused by a faulty function of the component, a false warning message is transmitted from the central monitoring device to the respective radio sensor node (01); - Determine a new threshold value (ψ schwell, 02) from the last recorded state measurements in the event of a transmitted false warning message in the respective radio sensor node (01), wherein the new threshold value (ψ schwell , 02) is suitable for identifying state measurements associated with an irregular state and now determines the current threshold value (Ψ schwell , 02) forms; and - Recording the state measurements during an application phase following the observation phase; the current threshold value (Ψ schwell , 02) is applied to detect state measurements associated with an irregular state from the state measurements recorded in the application phase; wherein the step of determining the threshold value (Ψ schwell , 02) comprises the following sub-steps from the recorded state measurement values, which are to be carried out in the respective radio sensor node (01): - Selecting m time periods from the initialization phase, whereby the following substeps are carried out for an i-th time period of the m time periods: ◯ Forming an i-th state measurement matrix from at least a selection of the state measurements recorded in the i-th time period; ◯ Determining a singular value decomposition of the i-th state measurement matrix, thereby obtaining singular values ​​of the i-th state measurement matrix; and o Determine the largest singular value (σ H ) of the i-th state measurement matrix (D), whereby an i-th largest singular value (σ Hi ) is obtained; - Determine a standard deviation (Ψ m ) for the m largest singular values ​​(σ Hi ); and - Evaluate the determined standard deviation (Ψ m ) with a predefined factor (t), whereby the threshold value (Ψ schwell, 02); wherein the i-th state measurement matrix comprises as elements at least the selection of the state measurement values ​​recorded in the i-th time period, wherein the state measurement values ​​are divided into n parts, each forming a row of the state measurement matrix. [2] Method according to claim 1, characterized by that the m largest singular values ​​(σ Hi ) are stored in a memory of the respective radio sensor node (01), wherein after determining a new largest singular value (σ Hm+1 ) the oldest of the largest singular values ​​(σ H1 ) in memory to obtain the new largest singular value (σ Hm+1 ) in memory. [3] Method according to claim 1 or 2, characterized by that the step of applying the current threshold (ψ schwell) for detecting state measurement values ​​associated with an irregular state from the state measurement values ​​recorded in the observation phase or in the application phase, comprises the following sub-steps, which are to be carried out in the respective radio sensor node (01): - forming a state measurement matrix from at least a selection of the state measurement values ​​recorded in a period of the observation phase or the application phase; - Determining a singular value decomposition of the state measurement matrix, whereby singular values ​​of the state measurement matrix are obtained; - Determine the largest singular value (σ H ) of the state measurement matrix; and - Compare the largest singular value (σ H ) with the current threshold (ψ schwell , 02), which determines the condition measurements associated with an irregular condition. [4] Method according to one of claims 1 to 3, characterized bythat at least one of the state measurement matrices is transmitted in a data-reduced form from the respective radio sensor node (01) by radio to the central monitoring device, for which the following sub-steps are carried out: - Determine the k largest singular values ​​(σ H ) of the state measurement matrix to be transmitted, where k ≤ 3; - Determine the k largest singular values ​​(σ H ) associated singular vectors (u, v); - Transformation of the singular vectors (u, v) into a frequency space, thereby obtaining transformed singular vectors; and - transmitting the k largest singular values ​​and the transformed singular vectors from the respective radio sensor node (01) by radio to the central monitoring device. [5] Method according to claim 4, characterized by that it includes the following further sub-steps: - Reconstruction of the state measurement values ​​from the k largest singular values ​​transmitted and the respective transformed singular vectors in the central monitoring device, thereby obtaining reconstructed state measurement values; - Check whether the reconstructed condition measurements correspond to a proper condition of the machine, so that the proper condition of the machine or not the proper condition of the machine is determined; - Reporting the proper condition of the machine from the central monitoring device to the respective radio sensor node (01) when the proper condition of the machine has been determined; - storing the respective condition measurement values ​​previously transmitted from the respective radio sensor node (01) to the central monitoring device in the respective radio sensor node (01) when the proper condition of the machine has been reported to the respective radio sensor node (01); and - Using the stored state measurements to determine the threshold value (ψ schwell , 02) . [6] Method according to one of claims 1 to 5, characterized by that it comprises the following further steps for predictive maintenance of the machine, which are to be carried out in the application phase in the respective radio sensor node (01): - determining a correlation coefficient between a newly detected condition measurement value associated with an irregular condition and the previously recorded condition measurements; and - Report a warning if the correlation coefficient is less than a predefined correlation limit (03). [7] Method according to claim 6, characterized by that it includes the following further steps: - transmitting the correlation coefficient from the respective radio sensor node (01) to the central monitoring device; - determining at the central monitoring device whether the transmitted correlation coefficient is due to a faulty function of the component, wherein, in a case in which it has been determined that the transmitted correlation coefficient is not due to a faulty function of the component, a false warning message is transmitted from the central monitoring device to the respective radio sensor node (01); and - Using the state measurements underlying the transmitted correlation coefficient to determine the threshold value (ψ schwell , 02) if a false warning message was transmitted. [8] Sensor arrangement for monitoring a function of at least one stored or stored component of a machine, comprising at least one radio sensor node (01) for detecting an operating state of the component and a central monitoring device, characterized bythat the at least one radio sensor node (01) and the central monitor device are configured to carry out a method according to one of claims 1 to 7.

Images