Method and component for monitoring the state of a device

Abstract

The present invention relates to a method and a component for state monitoring of a device, wherein the device is state-monitored by measuring a time-continuous and value-continuous analog signal representing the state variable of the device, the analog signal is converted into a digital signal at sampling time points within a measurement interval by means of an analog-to-digital converter, the analog-to-digital converter operates within a measurement range, for the case where the analog signal exceeds the measurement range, the signal portion of the analog signal exceeding the measurement range is cut off in the digital signal, subsequently a spectral analysis is applied to the digital signal in order to determine which frequency components the analog signal includes in the spectrum, and a fault of the device is inferred based on a comparison of the occurring frequency components with known damage frequencies, for the case where the analog signal exceeding the measurement range at the sampling time point is cut off in the digital signal, this event is detected and this event is determined as a quantity, and a signal quality is provided as the quotient of the quantity within the measurement interval and the total number of samples.

Description

Field of the Invention

[0001] The present invention relates to a method for monitoring the state of a device by measuring a time - continuous and value - continuous analog signal representing a state variable of the device, wherein the analog signal is converted into a digital signal at sampling time points within a measurement interval by means of an analog - to - digital converter, wherein the analog - to - digital converter operates within a measurement range, and wherein, for the case where the analog signal exceeds the measurement range, the signal portion of the analog signal exceeding the measurement range is cut off in the digital signal, and subsequently a spectral analysis is applied to the digital signal in order to determine which frequency components the analog signal includes in the spectrum, and a fault of the device is inferred based on a comparison of the occurring frequency components with known damage frequencies.

[0002] The present invention also relates to an electronic component designed for monitoring the state of a device, the electronic component including an input circuit for measuring a time - continuous and value - continuous analog signal representing a state variable of the device; an analog - to - digital converter which scans the analog signal at sampling time points within a measurement interval and converts the analog signal into a digital signal, wherein the analog - to - digital converter is designed to operate within a measurement range, and wherein, for the case where the analog signal exceeds the measurement range, the signal portion of the analog signal exceeding the measurement range is cut off in the digital signal; and a mechanism for spectral analysis in order to apply a spectral analysis to the digital signal in order to determine which frequency components the analog signal includes in the spectrum, and a fault of the device is inferred based on a comparison of the occurring frequency components with known damage frequencies. Background Art

[0003] Such a method and such a component are known, for example, from the operating instruction "Siemens, SIPLUS SM 1281, ConditionMontioring System SM 1281 Condition Monitoring, 06 / 2016, AE536912747 - AB".

[0004] For the purposes of the present invention, the state monitoring of a device is understood as preventive monitoring of a machine or a facility in order to avoid major damage or long downtimes.

[0005] For example, the diagnosis of a bearing housing is carried out based on vibration diagnosis by frequency analysis. The principle of frequency analysis lies in transforming a signal from the time domain to the frequency domain by spectral analysis. For this, a common mathematical method is the Fourier transform.

[0006] The disadvantages of known methods and components are that, for example, in the case of vibration analysis, especially in the case of severe bearing damage, the degree of bearing damage should be diagnosed, and the sensor signal input briefly reaches saturation because the actually occurring amplitude reaches the upper / lower limit of the measurement input range. Other reasons for high amplitudes may be due to faulty components or a changed machine train alignment and the occurring imbalance caused by process influences. As a result, the sensor signals and measurement inputs of the digital system are overmodulated because the actually occurring errors exceed the designed measurement range.

[0007] Sensor signal saturation also occurs in current signature analysis, especially when current converters with ferrite cores, but also Hall sensors or Rogowski current measurement inputs reach saturation in the case of high starting currents or excessive starting currents in the motor, or when a critical fault current occurs in a single branch of a three-conductor system and the measurement system is not designed for the intensity of the critical fault current.

[0008] The disadvantages of the devices and methods known from the prior art are that the detected signals with cut-off signal parts generate additional faulty frequencies in the subsequently to be evaluated spectrum, where the signal parts are caused, in particular, by overmodulation of the measurement input. Summary of the Invention

[0009] The object of the present invention is to provide a method that identifies the overmodulation of parts of state monitoring sensor signals and provides a measure for the overmodulation or a measure for the signal quality.

[0010] For the method initially proposed, the object is achieved in such a way that, for the case of cutting off an analog signal that exceeds the measurement range at a sampling time point in a digital signal, the event is detected and determined as a quantity, where signal quality is provided as the quotient of the quantity and the total number of samples within a measurement interval, and the signal quality is used to evaluate whether known damage frequencies can still be identified from the determined frequency parts of the spectrum despite the possible additional overmodulation parts generated in the spectrum by the cut-off signal parts.

[0011] The method according to the invention provides that at each sampling time point it is determined whether the measurement range has been overmodulated and thus a cut-off signal is formed. Then, some of the cut-off signal events are associated with the total number of samples. Advantageously, if the signal quality is not only determined and communicated to a superior system or application, but is also used for the optimized use of signal correction methods.

[0012] Thus, in an optimized design of the method, it is proposed to correct or estimate a digital signal using a signal correction method, thereby eliminating or at least reducing additional overshoot portions formed by clipping in the estimated signal, wherein known corrupted frequencies are also identified and the corrupted frequencies are mapped into a transfer function, and wherein the transfer function is continuously used according to the signal quality until a preset limit value of the signal quality.

[0013] Since now a plurality of samples outside or at the edge of a certain measurement range are continuously detected according to this method, this knowledge can be used to provide signal quality, wherein the plurality of samples are also defined as low level and high level. The mentioned limit can be defined as the limit for reliably ensuring a certain linearity of the measurement channel, the maximum available limit in an analog-to-digital converter or an amplifier (for example, the operating voltage spacing from the operating voltage) or the limit for reliably allowing an application measurement limit that can be adjusted to a certain extent.

[0014] The maximum signal amplitude that can be applied to the input of the converter before causing "clipping" at the digital output is called the full measurement range (also called full deflection) of the analog-to-digital converter. In the case of full deflection, the minimum and maximum codes of the analog-to-digital converter are used at the output.

[0015] For the present invention, clipping is understood as "cutting off" the excess signal portion by the processing hardware or software. Since most types of corruption in the spectrum can be recognized when typical corrupted frequencies or typical patterns of corrupted frequencies occur, it is disadvantageous if frequency portions already caused by clipping additionally appear in the spectrum to be analyzed.

[0016] However, this method now enables a better evaluation of the observation and assessment of frequency portions through the determined signal quality. If, for example, the vibration is greater than the measurement range due to the actually occurring amplitude, the information on signal quality can be used to perform the evaluation of frequency portions and / or the evaluation of the deteriorated state of the device to be monitored.

[0017] It is also advantageous to modify the transfer function according to the signal quality.

[0018] It is also advantageous to form and archive a series of transfer functions according to the signal quality during a learning phase when the device is in a good state, wherein the transfer function of the monitored device in its original good state is then used as a reference.

[0019] Especially with regard to a suitable estimated signal, in order to reduce the frequency components formed by clipping of the overshoot part, it is advantageous to use a Wiener filter as a signal correction method. The Wiener filter is used in the method to attempt to reconstruct (estimate) missing or noise-distorted or -distorted sampled values. Although the Wiener filter is used to estimate an unknown original signal in the time domain from a noisy signal, and the measured signal distorted by an LDI system with a transfer function is used as the best possible approximation Y in the time domain, the filter can also be used for clipped signals.

[0020] For the above-mentioned electronic component, the above-mentioned object is also achieved in such a way that a detection mechanism is present in the component, which is designed to detect the event in the case of cutting off an analog signal that exceeds the measurement range at the sampling time point in the digital signal and to count this event as a quantity in a counter, where there is a signal quality evaluation device that provides the signal quality as the quotient of the quantity within the measurement interval and the total number of samples, and there is also a decision-making mechanism that is designed to evaluate whether known damage frequencies can still be identified from the determined frequency components of the spectrum despite the additional overshoot components generated in the spectrum by the possibly cut-off signal part, based on the signal quality.

[0021] Clipping mainly produces spectral lines with an amplitude of up to 10% in the case of high frequencies. Transferring this method to bearing damage analysis means that the spectral lines representing bearing damage may gradually become apparent. Either it causes misinterpretation ("false alarm"). But either the bearing damage is classified as more severe than it actually is because the existing lines and harmonics are covered by the clipped lines.

[0022] In an improved version of the component, the component has a filter mechanism in which a signal correction method is run, which corrects or estimates the digital signal and thereby eliminates or at least reduces the additional overshoot components formed by the cut-off in the estimated signal, where the filter mechanism also has a transfer function that also maps known damage frequencies, where the filter mechanism has an input for the signal quality, and the filter mechanism is designed to continue using the transfer function up to a presettable limit value based on the signal quality.

[0023] When using the signal correction method, a reduction in false alarms can be expected; the measurement system can also operate more reliably, for example, within a certain frequency band beyond the designed measurement range. The measurable damage frequencies during normal operation, that is, all the measurable damage frequencies within the measurement limit (signal quality Q = 1), are identified by the Wiener filter and mapped into the transfer function. If a clipped signal (Q < 1) appears now, the transfer function should still be retained, that is, the artifacts caused by clipping should be suppressed first, but the bearing damage frequencies that may already exist will continue to be transmitted. The method for monitoring the state largely retains its spectrogram and classification functions.

[0024] To achieve this, it is advantageous to have an adaptation mechanism within the component, which is designed to modify the transfer function according to the signal quality.

[0025] The component can also be continuously optimized using a learning mechanism, which is designed to form and archive a series of transfer functions according to the signal quality during the learning phase when the device is in good condition. The learning mechanism is also designed to use the transfer function of the monitored device in its original good state as a reference.

[0026] Advantageously, the Wiener filter is implemented in the filter mechanism. Description of the Drawings

[0027] The drawings illustrate embodiments of the present invention, showing here:

[0028] Figure 1 A component for state monitoring is shown,

[0029] Figure 2 Showing Figure 1 An improved version of the component in

[0030] Figure 3 The principle of state monitoring of the device is shown,

[0031] Figure 4 An example of a spectrum with fault frequencies is shown,

[0032] Figure 5 The signal to be detected that transitions to the measurement range limit is shown,

[0033] Figure 6 A chart for spectral power density is shown,

[0034] Figure 7 The transfer function derived from the spectral power density is shown, and

[0035] Figure 8 The signal estimated using the Wiener filter compared to the original signal is shown. Detailed Description of the Invention

[0036] In Figure 1 is shown a component 10 for state monitoring of a device 1. The component 10 includes: an analog signal xa(t) that is time-continuous and value-continuous and represents a state variable of the device 1; an analog-to-digital converter ADU that scans the analog signal xa(t) at sampling time points ts within a measurement interval Tn and converts the analog signal into a digital signal xd[k], wherein the analog-to-digital converter ADU is designed to operate within a measurement range MB, and wherein, for the case where the analog signal xa(t) exceeds the measurement range MB, the signal portion of the analog signal xa(t) that exceeds the measurement range MB is cut off in the digital signal xd[k] (see Figure 5 ). Further, the component 10 has a mechanism 12 for spectral analysis in order to apply spectral analysis to the digital signal xd[k]. From the determined spectral analysis or from the determined spectral FFT, it can be read out which frequency components fi the analog signal xa(t) includes, and based on the comparison of the frequency components fi that occur in a comparison mechanism 21 with known damage frequencies fs, a fault of the device 1 can be inferred.

[0037] According to the invention, there is now a detection mechanism 13 in the component 10, which is designed to detect the event for the case where the analog signal xa(t) that exceeds the measurement range MB at the sampling time point ts is cut off in the digital signal xd[k], and to count this event as a quantity NSat in a counter 14.

[0038] Via a signal quality assessment device 15, a signal quality Q can now be provided as the quotient of the quantity NSat within the measurement interval Tn and the total number of samples N. Further, there is a decision-making mechanism 16 in the component 10, which is designed to evaluate, despite additional overshoot components fk generated in the spectral FFT by possibly cut-off signal portions, whether the known damage frequency fs can still be identified from the determined frequency components fi of the spectral FFT based on the signal quality Q.

[0039] If it is decided in the decision-making mechanism 16 with the aid of the signal quality Q that the damage frequency fs can still be determined from the determined frequency components fi, the valid signal 30 is forwarded to an evaluation unit.

[0040] Figure 2 Shows Figure 1An improved version of the component 10 as shown. The component 10 now additionally has a filter mechanism 17 in which a signal correction method WF is run. The signal correction method corrects or estimates the digital signal xd[k] and thereby eliminates or at least reduces the additional overshoot part fk formed by clipping in the estimated signal y[k]. The filter mechanism 17 also has a transfer function G(w), which also maps known damage frequencies fs. The filter mechanism 17 has an input 18 for the signal quality Q, and the filter mechanism 17 is designed to continue using the transfer function G(w) according to the signal quality Q until a preset limit value Vt.

[0041] According to Figure 2 , the component 10 also has an adaptation mechanism 19, which is designed to modify the transfer function G(w) according to the signal quality Q. In the design variant according to Figure 2 , the component 10 is also provided with a learning mechanism 20, which is designed to form and archive a series of transfer functions Gj(w) according to the signal quality Q during the learning phase when the device 1 is in a good state. The learning mechanism 20 is also designed to use the transfer function P(w) of the monitored device 1 in its original good state as a reference.

[0042] A Wiener filter is implemented in the filter mechanism 17.

[0043] Using Figure 3 Schematically shows the condition monitoring of the device 1. The device 1 includes a motor with a transmission, a drive shaft, and a swinging load. For example, an unbalance 31 is placed on the swinging load. Measurements 33 are recorded via a sensor 32 when in a good state. Due to the unbalance 31 on the flywheel, a measurement 34 with an increased amplitude of machine vibration can be determined as a fault. The associated spectrum 35 of the vibration speed shows three different frequency components for three different rotational frequencies.

[0044] Using Figure 4 Exemplarily shows which faults can be determined according to the damage frequencies on the spectrum. Thus, the spectrum 40 with fault frequencies shows a belt fault at position 41, a rolling bearing damage at position 42, a blade passing frequency at position 43, and a tooth meshing fault at position 44.

[0045] Figure 5 Shows the original signal x'a(t) reaching the limit from top to bottom, where the measurement range MB is plotted in the signal curve x'a(t), and once the signal part of the original signal x'a(t) leaves the measurement range, it is clipped. Below it, the signal quality Q is plotted, which decreases from 100% to 0% as the clipping increases. Below it, the number NSat of the clipped signal parts is plotted inversely.

[0046] Figure 6 The power density PSD of the spectrum is shown, where, in the graph, the power density 61 is plotted for x with x on the one hand and the power density 62 is plotted for x of the clipped signal on the other hand.

[0047] Figure 7 It shows a filter constructed from the curve of Figure 6 according to the transfer function of Wiener. Clipping mainly produces spectral lines with an amplitude of up to 10% in the case of high frequencies. This behavior, when transferred to bearing damage analysis, means that the spectral lines representing bearing damage may gradually become apparent. Either it causes misinterpretation ("false alarm"), but either the bearing damage is classified as more serious than it actually is. This is because the existing spectral lines and harmonics are partially covered by the faults caused by clipping.

[0048] The estimated signal formed from the original signal using the Wiener filter now has almost only the three main components of the original signal. The false part of the estimated signal is in the range of <1%, that is, the spectral lines caused by clipping are suppressed by 10 -2 , and the existing spectral lines are broadened. As long as the measure Q according to the present invention or the measure of the number of similar saturation measurement values exceeds the threshold, the power density PSD of the spectra of PSDxx and PSDXS can be continuously calculated by using the derived transfer function to optimize the Wiener filter.

[0049] Using Figure 8 shows the original signal xa[k], the observed signal xd[k] including the clipped part, and the signal y[k] estimated by the Wiener filter in the FFT spectrum 80.

[0050] The signal Y estimated from X using the Wiener filter has almost only the three main components of the original signal. The wrong part of the estimated signal is in the range of <1%, that is, the spectral lines caused by clipping are suppressed by 10 -2 , and the existing spectral lines are shown broadened.

Claims

1. A method for condition monitoring of a device (1) by measuring a time - continuous and value - continuous analog signal (xa(t)), wherein, The analog signal represents a state variable of the device (1), wherein the analog signal (xa(t)) is converted into a digital signal (xd[k]) at sampling time points (ts) within a measurement interval (TN) by means of an analog-to-digital converter (ADU), wherein the analog-to-digital converter (ADU) operates within a measurement range (MB), and wherein, for the case where the analog signal (xa(t)) exceeds the measurement range (MB), the signal portion of the analog signal (xa(t)) that exceeds the measurement range (MB) is cut off in the digital signal (xd[k]), and wherein subsequently a spectral analysis is applied to the digital signal (xd[k]) in order to determine which frequency components (fi) the analog signal (xa(t)) includes in a spectrum (FFT), and a fault of the device (1) is inferred based on a comparison of the occurring frequency components (fi) with known fault frequencies (fs). It is characterized in that, for the event that the analog signal (xa(t)) exceeding the measurement range (MB) at the sampling time point (ts) is cut off in the digital signal (xd[k]), this event is detected and determined as a quantity (NSat), wherein a signal quality (Q) is provided as the quotient of the quantity (NSat) and the total number of samples (N) within the measurement interval (TN), and the signal quality (Q) is used to evaluate whether the known fault frequencies (fs) can still be identified from the determined frequency components (fi) of the spectrum (FFT), despite additional overshoot components (fk) being generated in the spectrum due to possibly cut-off signal portions.

2. The method according to claim 1, wherein, A signal correction method (WF) is used to correct or estimate the digital signal (xd[k]), and as a result, the additional overshoot components (fk) formed by the cut-off are eliminated or at least reduced in the estimated signal (y[k]), and wherein the known fault frequencies (fs) are also identified and mapped into a transfer function (G(w)), and the transfer function (G(w)) is continued to be used according to the signal quality (Q) until a preset limit value (Vt) of the signal quality (Q).

3. The method according to claim 2, wherein The transfer function (G(w)) is modified according to the signal quality (Q).

4. The method according to claim 2 or 3, wherein In a learning phase when the device (1) is in a good state, a series of transfer functions (Gj(w)) are formed and archived according to the signal quality (Q), and then the transfer function (Gj(w)) of the monitored device in its original good state is used as a reference.

5. The method according to claim 2 or 3, wherein A Wiener filter is used as the signal correction method (WF).

6. An electronic component (10), the electronic component being designed for state monitoring of a device (1), the electronic component comprising: - an input circuit (11) for measuring a time-continuous and value-continuous analog signal (xa(t)) representing a state variable of the device (1). - An analog-to-digital converter (ADU) that scans the analog signal (xa(t)) at sampling time points (ts) within a measurement interval (TN) and converts the analog signal into a digital signal (xd[k]), where the analog-to-digital converter (ADU) is designed for a measurement range (MB), and for the case where the analog signal (xa(t)) exceeds the measurement range (MB), the signal portion of the analog signal (xa(t)) that exceeds the measurement range (MB) is cut off in the digital signal (xd[k]). - An apparatus (12) for spectral analysis to apply spectral analysis to the digital signal (xd[k]) to determine which frequency components (fi) the analog signal (xa(t)) includes in a spectrum (FFT), and to infer a fault of the device (1) based on a comparison of the occurring frequency components (fi) with known fault frequencies (fs). Characterized in that - There is a detection mechanism (13) designed to detect an event where the analog signal (xa(t)) that exceeds the measurement range (MB) at a sampling time point (ts) is cut off in the digital signal (xd[k]), and to count the event as a quantity (NSat) in a counter (14). - Wherein there is a signal quality evaluation device (15) that provides a signal quality (Q) as the quotient of the quantity (NSat) within the measurement interval (TN) and the total number of samples (N). - There is also a decision-making mechanism (16) designed to evaluate, based on the signal quality (Q), whether the known fault frequencies (fs) can still be identified from the determined frequency components (fi) of the spectrum (FFT), despite additional overshoot components (fk) being generated in the spectrum (FFT) due to possibly cut-off signal portions.

7. The component (10) according to claim 6 further has a filter mechanism (17) in which a signal correction method (WF) is run, the signal correction method correcting or estimating the digital signal (xd[k]) and, as a result, eliminating or at least reducing in the estimated signal (y[k]) the additional overshoot portion (fk) formed by clipping, wherein, The filter mechanism (17) also has a transfer function (G(w)) that also maps the known fault frequencies (fs), where the filter mechanism (17) has an input terminal (18) for the signal quality (Q), and the filter mechanism (17) is designed to continue using the transfer function (G(w)) according to the signal quality (Q) until a preset limit value (Vt).

8. The component (10) according to claim 7, having an adaptation mechanism (19) designed to modify the transfer function (G(w)) according to the signal quality (Q).

9. The component (10) according to claim 7 or 8, having a learning mechanism (20) designed to form and archive a series of transfer functions (Gj(w)) according to the signal quality (Q) during a learning phase when the device (1) is in a good state, and the learning mechanism is also designed to use the transfer function (G(w)) of the monitored device (1) in its original good state as a reference.

10. The component (10) according to claim 7 or 8, wherein, A Wiener filter is implemented in the filter mechanism (17).

Images