WO2016039658A1

WO2016039658A1 - Monitoring device for monitoring a machine that vibrates

Info

Publication number: WO2016039658A1
Application number: PCT/RU2014/000685
Authority: WO
Inventors: Alexey Petrovich KOZIONOV; Oleg Vladimirovich MANGUTOV; Ilya Igorevich MOKHOV
Original assignee: Siemens Aktiengesellschaft
Priority date: 2014-09-12
Filing date: 2014-09-12
Publication date: 2016-03-17

Abstract

The invention is concerned with a method for operating a remote monitoring device (MD) that monitors a machine (10) that vibrates. The method comprises the step of receiving a vibration signal (V1, V2, V3, V4) from a vibration transducer unit (26) coupled to the machine (10). The object of the invention is to handle the vibration signal (V1, V2, V3, V4) efficiently with regard to the amount of data to be stored or transmitted. The method solves this problem by segmenting (S10) the vibration signal (V1, V2, V3, V4) into segments (S1), generating (S12) transformation coefficients (C) by applying a transform (T) to each segment (S1), and generating (S14) model parameter values (P) for a process model (M) of the vibrating machine (10) on the basis of the transformation coefficients (C). Finally, the model parameter values (P) are output to a date storage (36) or a central monitoring system (40).

Description

MONITORING DEVICE FOR MONITORING A MACHINE THAT VIBRATES

Description The invention is concerned with a monitoring device for monitoring a machine that vibrates. The vibration is observed by generating a vibration signal by means of a vibration transducer unit. As the vibration signal comprises a large amount of data, handling the vibration signal is technically challenging.

In remote condition monitoring and diagnostic systems, tools for analysing vibration data for machine health estimation are usually dealing with a huge amount of data to be stored and transferred to a central monitoring system for detailed analysis. The amount of data is caused by the high sampling rates required to make a diagnosis e.g. in case of monitoring of a rotating machinery. Typically, for the vibration analysis experts use spectral information obtained by e.g. a fast Fourier transform (FFT) calculated out of time series signals received from vibration transducers. For example, for a single machine equipped with ten vibration transducers and a processing unit which calculates one FFT (1024 lines) per second from measured vibration data at every channel (i.e. from every transducer) , the overall amount of data to be stored or transferred is close to 3.5 gigabytes per machine per day (it is assumed that every amplitude is stored by a double value, i.e. 4 bytes, resulting in lx 10 x 60 x 60 x 24 x 1024 x 4 bytes = 3 538 944 000 bytes) . The object underlying the invention is to efficiently handle the amount of data needed for monitoring a vibrating machine . The problem is solved by the subject-matter of the independent claims. Further advantageous embodiments of the invention result from the features of the dependent claims.

The invention provides a method for operating a remote monitoring device that monitors a machine that vibrates. The monitoring device may be placed at the machine in e.g. a production plant. The machine can be e.g. a rotating machine, like an electrical machine or a combustion engine. The inventive method comprises the following steps. A vibration signal is received from a vibration transducer unit coupled to the machine. The vibration signal is correlated to vibration movements of the machine, i.e. a shaking movement and/or a body sound. The vibration signal is segmented into pieces or sections or segments. The segments may be overlapping segments or adjacent segments. For example, the vibration signal may be digitized and each segment may comprise a predefined number of sample values of the digitized vibration signal, wherein the number of sample values can be in a range from 128 to 4096 samples. The more values are comprised in each segment, the higher the frequency resolution for analyzing the vibration signal may be .

By applying a transform to each segment, transformation coefficients are generated. In case that the transform is a spectral transform, the transformation coefficients are called spectral coefficients. In case that the transform is a cepstral transform, the transformation coefficients are called cepstral coefficients. In a further step, a process model of the vibrating machine is matched to the transformation coefficients. Such a model may be, e.g., an autoregressive (AR) model. Matching the model to the transformation coefficients yields parameter values for the model parameters. For example, the model parameter values of the model are varied until values are found, such that the process model fulfils a predefined similarity criterion with regard to the transformation coefficients. Next the model parameters are output.

The invention also provides a monitoring device for monitoring a machine that vibrates . The monitoring device is suitable for performing the inventive method. The monitoring device comprises a vibration transducer unit coupleable to the machine and configured to generate a vibration signal correlated to the vibration of the machine. Further, a processing unit is provided that is coupled to the transducer unit and that is configured to process the vibration signal by applying a method according to the invention. The processing unit can comprise a microcontroller or a digital signal processor. An output unit is coupled to the processing unit and configured to receive model parameter values from the processing unit and configured to transmit the model parameter values to a local storage and/or a central monitoring system. The output unit can comprise a storage, like e.g. a hard drive or a flash drive, or a sending unit, like e.g. LAN- communication module f.WLAN - wireless local area network) or a mobile communication unit, like a UMTS -module (UMTS - universal mobile telecommunications system) . The output unit may also comprise a controller for accessing a communications network, like e.g. an ethernet or a communication bus, like e.g. PROFINET. The invention provides the advantage that expressing a spectrum or a capstrum by means of model parameter values of a process model takes significantly less data than expressing the spectrum or capstrum by the transformation coefficients. In other words, by expressing the signal using model parameters, the amount of data needed to store the relevant information or transmit this information to the central monitoring system may be far less without the risk of missing vital information needed to sufficiently monitor the machine.

As was already described, in one embodiment of the invention the step of outputting the model parameter values comprises the step of saving the model parameter in a local data storage. In other words, the data storage may be included or integrated into the monitoring device itself. This allows for monitoring the machine without the need of continuous access to the monitoring device, like can be the case e.g. with a submarine robot device. In one embodiment, the step of outputting the model parameter values comprises sending the model parameter values to a central monitoring system. This allows for centrally observing the condition of the machine and immediate reaction, when the vibration signal exhibits a predefined pattern that indicates e.g. a damage of the machine .

The vibration transducer unit may comprise one or more vibration transducers. In the vibration transducer unit, at least one transducer may be configured to measure acceleration. This is especially advantageous for measuring high frequency vibrations with low amplitude. In one embodiment, the vibration transducer unit comprises at least one transducer configured to measure velocity. This is particularly advantageous for measuring vibrations with large amplitudes. In one embodiment the vibration transducer unit comprises at least one transducer configured to measure displacement. This is especially advantageous for measuring vibrations with low frequency.

As was already described, the transformation coefficients provide different forms of representing each segment depending on the type of transform use. In one embodiment, the transform is configured to generate a spectrum. Such a transform may be realized as a digital Fourier transform, like it may be obtained using an FFT. One embodiment provides a transform that is configured to generate a spectrum envelope. This already provides less data than describing a full, complex valued spectrum. In one embodiment the transform is configured to generate a cepstrum._. Using a cepstrum for analyzing the vibrations of a machine provides the special advantage that a fundamental frequency, i.e. the stimulus of a vibration, is separated from the transfer characteristics of body parts of the machines transmitting the vibration stimulus to the vibration transducer.

Once the segments are transformed, the model parameter values are generated. Basically there are two different ways .

In one embodiment, the step of generating the model parameter values comprises generating model parameter values for each segment separately. In other words, the process model is matched to the transformation coefficients in one segment, which results in a first set of model parameter values. Additionally, the model is matched to the transformation parameters of the next segment. This results in a further set of model parameter values. Collecting the model parameter values for all segments yields the output.

In another embodiment, the step of generating the model parameter values comprises combining corresponding transformation coefficients of each segment into separate time series of coefficients. For example, if the transform is a spectral transform, the transformation coefficients for the frequency 100 Hz may be taken from each segment. This results in a time series of transformation coefficients representing the spectral amplitude at the frequency 100 Hz. This time series may than be modelled by the process model. The coefficients for the other frequencies may be treated likewise. In other words, the model parameter values are generated for each time series separately .

The better the process model describes the behaviour of the vibrating machine, the fewer model parameters are needed to adapt the model to the actual vibration signals. In one embodiment, the process model is configured to describe a resonant body. In other words, the model may completely based on a second order differential equation. A process model based on a resonant body provides the advantage, that a specific resonance frequency of the body is easily expressed by a single model parameter. In one embodiment, the model comprises a linear predictive coding model. Such a model yield an especially high compression factor with regard to the amount of model parameters needed to describe the vibration signals. An especially preferred embodiment is based on a model that comprises the already-mentioned auto-regressive (AR) model. An auto-regressive model is suitable for describing several resonance frequencies with only a few model parameters . In one embodiment, the number of parameters of the model is set in dependence on at least some of the transformation coefficients. In the case, that an auto-regressive model is used, this means, that the order of the AR-model is set in dependence on transformation coefficients. This provides the advantage that the degree of detail that is preserved may be adapted to the monitoring tasks.

In one embodiment, the number of parameters is determined on the basis of a Bayesian information criterion, BIC, or the Akaike information criterion, AIC. These two criteria allow for a model selection among a finite set of models. In other words, several different process models may be provided with different model order or number of parameters,, and by using one of the mentioned criteria, one of the process models may be selected automatically.

In one embodiment, the number of parameters is determined by stepwise increasing the number until an R2 -coefficient of the match between the model and the transformation coefficients is above a predefined threshold value. The R2- coefficient is also known as coefficient of determination or R- squared (R²) . It describes, how well a spectrum or cepstrum is described or fit by the process model. This embodiment allows for finding the minimum number of parameters that provides for a certain R2 -coefficient value .

In one embodiment the number of parameters or model order is adapted in predefined time intervals. In other words, the number of parameters is updated regularly. This allows for minimizing the amount of data adaptively without loosing essential or vital details with regard to the monitored vibration signals. In one embodiment, a configuration interface receives the number of parameters from a user. In other words, the user may manually set the number of parameters. This provides for considering or taking into account the judgment of an operator.

In one embodiment the number of process parameters is set such that the process model excludes a predefined noise that is different from the vibration of the machine. In other words, the process model is provided with a number of process parameter that is suitable for denoising the vibration signal. For finding a suitable number, simple experiments on the basis of vibration signals may be performed. It has shown that an AR-model is especially suitable for denoising.

As has been already pointed out, the number of model parameter values is preferably smaller than the number of transformation coefficients. Thus, the data amount of the model parameter values is smaller than the data amount of the transformation coefficients, leading to the compression of monitoring data.

In the following, an example embodiment is described on the basis of the figures. The figures show:

FIG 1 a schematic illustration of an embodiment of the inventive monitoring device,

FIG 2 a flow diagram illustrating an embodiment of the inventive method that may be performed by the monitoring device shown in FIG 1, FIG 3 a sketch for illustrating transformation coefficients generated by the monitoring device of FIG 1, Fig 4 a spectrum described by the transformation coefficients and a process model matched to the transformation coefficients,

FIG 5 a diagram illustrating the compression capability of the process model and

FIG 6 a diagram illustrating the value of the R2- coefficient for different numbers of model parameters that may be used.

The embodiment explained in the following is a preferred embodiment of the invention. However, in the embodiment, the described components of the embodiment each represent individual features of the invention to be considered independently of each other, which each develop the invention also independently of each other and thereby are also to be regarded as a component of the invention in individual manner or in another than the shown combination. Furthermore, the described embodiment can also be supplemented by further features of the invention already described .

FIG 1 shows a machine 10. The machine can be e.g. a rotating machine, like an electrical motor or an electrical generator or a combustion engine. As an example, the machine can comprise a drive shaft 12 that is performing a rotation 14. The drive shaft 12 can be supported by a bearing 16 that can be part of a bearing shield 18. While the machine 10 is running, its body 20 and the bearing 16 may perform a vibration 22 that is illustrated in FIG 1 by lines indicating the shaking movement. The vibration 22 can depend on the state or condition that the bearing 16 is in. In other words, by analyzing the vibration 22, it is possible to monitor or survey the state of the bearing 16. Similarly, the vibrations at other parts of the machine 10 can be measured in order to monitor the condition or state of other parts of the machine 10. In order to measure the vibration 22, vibration transducers 24 can be mechanically coupled or attached to the machine 10. Each vibration transducer 24 can be configured to measure acceleration, velocity or displacement. The vibration transducers 24 together represent a vibration transducer unit 26.

The vibration transducers 24 can be connected to a processing unit 28. Each vibration transducer 24 can generate or output a vibration signal VI, V2, V3 , V4 that can be transmitted to the processing unit 28. The processing unit 28 can comprise a micro-controller or a digital signal processor (DSP) . The processing unit 28 can be located in a proximity of the machine 10, for example as close as less than 5 meters. The processing unit 28 can be attached to the machine 10. The processing unit 28 can be configured to digitize the vibration signals VI, V2 , V3 , V4 by means of an analogue- to-digital-converter 30. Each digital digitized vibration signal VI, V2, V3 , V4 can be processed separately by a signal processing path or chain 32. In FIG 1 only one processing chain 32 for one of the vibration signals VI, V2 , V3 , V4 is shown. By means of the signal processing chain 32, the processing unit 28 can generate model parameter value P for model parameters of a process model M which can be an auto-regressive model AR. The model parameter values P describe the vibration signals VI, V2 , V3 , V4. However, the amount of data represented by the model parameter values P is significantly less than the amount of data representing the digital vibration signals VI , V2 , V3 , V4.

The vibration transducer unit 26, the processing unit 28 and the output unit 34 represent a monitoring device MD for remotely the monitoring machine 10.

The model parameter values P can be output to an output unit 34 that can be attached or connected to the process unit 28. The output unit 34 can be configured to store the model parameter values P in a local storage 36 that can be e.g. a hard drive or a flash drive. The output unit 3 can also be configured to transmit the model parameter values P via a communication channel 38 to a central monitoring system 40 that can be e.g. located in a master control station of a production plant.

In the following, it is explained, how the processing unit 28 can generate model parameter values P from one of the vibration signals VI, V2 , V3 , V4 , e.g. the vibration signal VI .

In FIG 2, the processing chain 32 is illustrated. A first step S10 of the processing chain comprises segmenting the digitized vibration signal that is output by the analogue- to-digital-converter 30. Each segment may comprise a predefined number of sample value, wherein the number can be in the range of 128 to 4096. This segmentation step S10 generates segments of the vibration signal VI. In FIG. 2, a single segment SI is shown for illustrating purposes only. The diagram show sample values over time t. Each segment SI, is transformed in a transformation step S12 by means of a transform T. The transform T can be e.g. an FFT. Transforming the segment SI yields transformation coefficients C that describe the frequency components comprised or contained in the segment SI . The diagram shows spectral amplitudes over frequency f. In a matching step S14, the process model M is matched or adapted to the transformation coefficients C. By matching the process model M to the transformation coefficients C, the graph of the model M takes on a shape that emulates the spectrum of the segments SI. Matching the process model M to the transformation coefficient C is performed by adapting model parameter values P of model parameters of the model M. Algorithms for finding model parameter values P for model parameters of an auto-regressive model or an linear predictive coefficient model or a second order differential equation for matching a given series of are well known in the prior art . In an outputting step S16 the model parameters P are stored or transmitted in the described way by the output unit 34.

In FIG 3 it is illustrated how the signal processing chain 32 can process several segments of the vibration signal VI. From data acquisition, we get a set containing NoS (number of spectrums) of transformed segments. Each spectrum contains two vectors: a vector of frequency 42 values fi to f_NoP and a vector of amplitudes 44 al to a_NoP of transformation coefficients C, each vector being of the length NoP (number of points), i.e. the segments length. Each vector 44 corresponds to one segment. In case of subsequent measurements 1... NoS, frequencies vectors are always the same and there is no need to store it every time. So it can be stored only once. Amplitude values a of transformation coefficients C be specified as a matrix 46 of size NoP X NoS . The indices of each amplitude value a indicate its position in the matrix 46. In addition there can be a vector 48 of time stamps T¹ to T¹⁰³ corresponding to available spectrums of NoS length. In the matrix 46, each line 50 represents corresponding transformation coefficients of each segment. Corresponding means, that each line 50 represents transformation coefficients for a given frequency value f .

Based on these acquisition data the signal processing chain 32 takes each frequency component over time, estimates AR- model of component, saves the spectrum data C as AR- coefficient, i.e. model parameter values P.

Not all information contained in the vibration signals VI, V2 , V3 , V4 is essential for analysis. Thus, the overall amount of data can be reduced. However it can be done only by taking into account the importance of particular components for correct detection of potential faults of the machine. Effective compression of data to be stored and taking into account the spectral data specifics, allows significant reduction of the amount of data to be stored and transferred. The approach realized by the signal processing chain can be used for compression of data used for vibration diagnostic (spectra, enveloping spectra and cepstrum) and allows major optimization for bandwidth and storage utilization with in remote monitoring and diagnostic systems.

For the task of data analysis and visualization, spectral data can be reconstructed from AR- coefficients or in general model parameters P (for one time stamp) . FIG 4 illustrates, how the spectral coefficients C are matched by the process model M, once the correct model parameters P have been set. In FIG 4, the spectral coefficients C are plotted over normalized frequency f. A further advantage is that the method can be used for denoising the spectrum signal, as is shown in FIG 4.

The degree of compression depends on the order of the AR- model, or in general, by the number of model parameters used in the process model M. the order of AR-models can be estimated by means of the Akaike information criterion, AIC, or Bayesian information criterion, BIC. Manually setting the number of parameters by judgement of an operator allows the operator to select in order of AR-model which equals the number of base frequencies. This results in the highest possible degree of detail representation by means of the AR-model.

The efficiency of the compressing method can be evaluated by two coefficients: compression coefficient and R2- coefficient (coefficient of determination) .

FIG 5 show the compression coefficient K. The compression coefficient K is the ratio of initial number of input values to number of values to be stored after compression. It can be calculated according to the following formula:

K = (size of input data) /(size of output data).

The higher the order 0 of the AR-model or in general the number of model parameters of the process model M, the lower the compression coefficient K (see FIG 5) .

FIG 6 illustrates values of the R2-coefficient in dependence on the chosen order 0. In statistics, the coefficient of determination, denoted R2 , indicates, how well data points fit a statistical model - sometimes simply a line or curve. R² = 1 - s² /s² _y, wherein s is the variance of model error of a model and s_y is the variance of the real FFT spectrum. FIG 6 shows that the method allows significance spectrum compression with a small number 0 of AR-coefficients .

By means of the monitoring device MD, spectral data can be compressed taking into account data specifics and features. The compression can be adaptive (based on the described accuracy metrics) and also dependent on a manual configuration. Also the described compression provides desnoising capabilities. Overall, the example shows, how a method for compression and denoising of spectra is provided by the invention.

Claims

1. Method for operating a remote monitoring device (MD) that monitors a machine (10) that vibrates, comprising the steps of :

- receiving a vibration signal (VI, V2 , V3 , V4 ) from a vibration transducer unit (26) coupled to the machine (10) ,

- segmenting (S10) the vibration signal (VI, V2 , V3 , V4) into segments (SI) ,

- generating (S12) transformation coefficients (C) by applying a transform (T) to each segment (SI) ,

- generating (S14) model parameter values (P) for a process model (M) of the vibrating machine (10) by matching the model (M) to the transformation coefficients (C) ,

- outputting (S16) the model parameter values (P) .

2. Method according to claim 1, wherein the step of outputting the model parameter values (P) comprises:

- saving the model parameter values (P) in a local data storage (36) and/or

send, the model parameter values (P) to a central monitoring system (40) .

3. Method according to any of the preceding claims, wherein the vibration transducer unit (26) comprises at least one transducer (24) configured to measure acceleration and/or velocity and/or displacement.

4. Method according to any of the preceding claims, wherein the transform (T) is configured to generate a spectrum and/or a spectrum envelope and/or a cepstrum.

5. Method according to any of the preceding claims, wherein the step of generating the model parameter values (P) comprises :

- generating model parameter values (P) for each segment (SI) separately; or

- combining corresponding transformation coefficients (50) of each segment into separate time series of coefficients and generating model parameter values (P) for each time series separately.

6. Method according to any of the preceding claims, wherein the model (M) is configured to describe a resonant body and/or the model comprises a linear predictive coding model and/or the model comprises an autoregressive model.

7. Method according to any of the preceding claims, wherein the number (O) of parameters of the model (M) is set in dependence on at least some of the transformation coefficients (C) .

8. Method according to claim 7, wherein the number (O) of parameters is determined:

- on the basis of a Bayesian information criterion, BIC, or the Akaike information criterion, AIC, and/or

- by stepwise increasing the number (0) until an R2- coefficient of the match between the model (M) and the transformation parameter values (P) is above a predefined threshold value.

9. Method according to claim 7 or 8 , wherein the number (O) of parameters is adapted in predefined time intervals.

10. Method according to any of the preceding claims, wherein a configuration interface receives the number (O) of parameters of the model (M) from a user.

11. Method according to any of the preceding claims, wherein the number (O) of process parameters is set such that the process model (M) excludes a predefined noise that is different from the vibration (22) of the machine (10) .

12. Method according to any of the preceding claims, wherein number (0) of model parameter values (P) is smaller than the number of transformation coefficients (C) .

13. Method according to any of the preceding claims, wherein data amount of the model parameter values (P) is smaller than the data amount of the transformation coefficients (C) .

14. Method according to any of the preceding claims, wherein each segment (SI) comprises a predefined number of sample values of the vibration signal, wherein the number of sample values is in a range from 128 to 4096 values.

15. Monitoring device (MD) for monitoring a machine (10) that vibrates, comprising:

a vibration transducer unit (26) coupleable to the machine (10) and configured to generate a vibration signal (VI, V2, V3, V4) correlated to the vibration (22) of the machine ( 10 ) ,

- a processing unit (28) coupled to the transducer unit (26) and configured to process the vibration signal (VI, V2, V3 , V4) by applying a method according to any of the preceding claims, - an output unit (34) coupled to the processing unit (28) and configured to receive model parameter values (P) from the processing unit (28) and configured to transmit the model parameter value (P) to a local storage (36) and/or a central monitoring system (40) .