METHOD AND SYSTEM FOR PROCESSING SENSOR DATA FOR TRANSMISSION

Abstract

In order to compress sensor data for transmission, one or several operational parameters of a machine (20) having a rotating component are received. A compression technique is applied to a spectral representation of the sensor data to generate a compressed sensor data representation. The applied compression technique is dependent on the one or several operational parameters of the machine (20).

Claims

1. A method of processing sensor data for transmission, the method comprising: receiving, by at least one processing device, one or several operational parameters of a machine having a rotating component or having a reciprocating component, the one or several operational parameters of the machine influencing the sensor data and being distinct from the sensor data; applying, by the at least one processing device, a compression technique to a spectral representation of the sensor data to generate a compressed sensor data representation, wherein the applied compression technique is dependent on the one or several operational parameters of the machine; and transmitting, by the at least one processing device, the compressed sensor data representation.

2. The method of claim 1, wherein the sensor data is time-domain data and the method further comprises generating the spectral representation of the sensor data by transforming the time-domain data into a frequency domain.

3. The method of claim 1, wherein the one or several operational parameters of the machine comprise a rotation speed of the rotating component of the machine or a frequency at which the reciprocating component of the machine reciprocates.

4. The method of claim 3, further comprising modifying the applied compression technique responsive to a change in the rotation speed or a change in the frequency at which the reciprocating component of the machine reciprocates.

5. The method of claim 1, wherein the applied compression technique is further dependent on one or several machine specifics of the machine.

6. The method of claim 5, wherein the one or several machine specifics are selected from a group consisting of fault cases, application type of the machine, ambient conditions.

7. The method of claim 1, further comprising receiving, by a data analytics computer, the compressed sensor data representation and analyzing, by the data analytics computer, the compressed sensor data representation.

8. The method of claim 7, wherein the data analytics computer determines at least one key performance indicator, KPI, of the machine.

9. The method of claim 7, further comprising modifying, by the processing device, the applied compression technique responsive to feedback information from the data analytics computer.

10. The method of claim 1, wherein the machine has a rotating component and the one or several operational parameters comprise a rotation speed of the rotating component of the machine, and wherein applying the compression technique comprises: applying an alignment transformation that is dependent on the rotation speed to the spectral representation of the sensor data to align the spectral representation of the sensor data with at least one reference spectrum of a set of reference spectra, and determining a set of decomposition coefficients of a linear decomposition of the spectral representation of the sensor data, wherein the set of decomposition coefficients is transmitted in the compressed sensor data representation.

11. The method of claim 1, wherein the one or several operational parameters comprise a rotation speed of the rotating component of the machine, wherein applying the compression technique comprises identifying, based on the rotation speed, a set of peaks in the spectral representation of the sensor data and determining peak characteristics for each peak in the identified set of peaks, wherein the peak characteristics of the peaks included in the identified set of peaks are transmitted in the compressed sensor data representation.

12. The method of claim 1, wherein the at least one processing device is a field sensor device.

13. The method of claim 1, wherein the machine is a generator or a motor.

14. A device for processing sensor data for transmission, comprising: an interface adapted to receive one or several operational parameters of a machine having a rotating component or having a reciprocating component, the one or several operational parameters being different from the sensor data; at least one processing circuit adapted to determine a compression technique that is to be applied to the sensor data as a function of the one or several operational parameters of the machine, the one or several operational parameters of the machine influencing the sensor data and being distinct from the sensor data, and apply the determined compression technique to a spectral representation of the sensor data to generate a compressed sensor data representation; and output circuitry adapted to transmit the compressed sensor data representation.

15. The device of claim 14, wherein the one or several operational parameters of the machine comprise a rotation speed of the rotating component of the machine or a frequency at which the reciprocating component reciprocates.

16. A system, comprising: a machine having a rotating component, in particular a motor or a generator, or a machine having a reciprocating component; and the device of claim 14.

17. A system, comprising: a machine having a rotating component, in particular a motor or a generator, or a machine having a reciprocating component; and the device of claim 15.

18. The method of claim 6, further comprising receiving, by a data analytics computer, the compressed sensor data representation and analyzing, by the data analytics computer, the compressed sensor data representation.

19. The method of claim 18, wherein the data analytics computer determines at least one key performance indicator, KPI, of the machine.

20. The method of claim 19, further comprising modifying, by the processing device, the applied compression technique responsive to feedback information from the data analytics computer.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0093] The subject-matter of the invention will be explained in more detail with reference to preferred exemplary embodiments which are illustrated in the attached drawings, in which:

[0094] FIG. 1 is a block diagram representation of a system according to an embodiment.

[0095] FIG. 2 is a data flow diagram according to an embodiment.

[0096] FIG. 3 is a flow chart of a method according to an embodiment.

[0097] FIG. 4 is an exemplary time diagram.

[0098] FIG. 5 is a flow chart of a method according to an embodiment.

[0099] FIG. 6 is a flow chart of a method according to an embodiment.

[0100] FIG. 7 illustrates processing of frequency-domain data in an embodiment.

[0101] FIG. 8 is a flow chart of a method according to an embodiment.

[0102] FIG. 9 illustrates processing of frequency-domain data in an embodiment.

DETAILED DESCRIPTION OF EMBODIMENTS

[0103] Exemplary embodiments of the invention will be described with reference to the drawings in which identical or similar reference signs designate identical or similar elements. While some embodiments will be described in the context of specific industrial systems, electric power systems, transportation systems, or vessels, such as wind turbines or motors installed onboard a train or vessel, the embodiments are not limited thereto.

[0104] The features of embodiments may be combined with each other, unless specifically noted otherwise.

[0105] FIG. 1 is a block diagram of a system 10. The system 10 comprises a machine 20. The machine 20 may have an operational parameter that is variable while the machine 20 is running. Variation of the operational parameter may influence sensor data captured using a sensor 32.

[0106] For illustration rather than limitation, the machine 20 may have a moving component, such as a rotating component 21. The rotating component may rotate with a rotation speed, which may be variable while the machine 20 is running. The rotation speed is exemplary for an operational parameter that may be used for controlling a compression algorithm applied to a spectral, i.e., frequency-domain, representation of the sensor data, as will be explained in more detail below.

[0107] While embodiments will be described in the context of a machine having a rotating component (such as a generator or a motor), with the operational parameter(s) including the rotating speed of the rotating component, the invention is not limited thereto. For illustration, the machine 20 may have a component that is reciprocating, and the operational parameter(s) used for controlling a compression algorithm may include the reciprocation frequency. The characteristic speed of a component of the machine 20 that undergoes a cyclic change will also be referred to as “running speed” herein.

[0108] The machine 20 may be a generator, e.g., a wind turbine, or a motor, e.g., a pump motor, that may be stationarily mounted. The machine 20 may be a motor or generator that may be mounted on a boat or train.

[0109] The system 10 includes a device 30, which may be a field sensor device. The device 30 has an interface 31 to receive one or several operational parameters of the machine 20, which are capable of affecting the sensor data sensed by the sensor 32. The interface 31 may be coupled to a measuring unit 23, which may measure the one or several operational parameters. The measuring unit 23 may comprise a rotation speed sensor. The measuring unit 23 may be comprised by the machine 20, may be integrated with the sensor device 30, or may be a device separate from both the machine 20 and the sensor device 30. Alternatively or additionally, the interface 31 may be coupled to a control unit 22 that controls operation of the machine 20 to receive the one or several operational parameters of the machine 20.

[0110] The sensor 32 of the device 30 may be integrated into a housing of the device 30 or may be otherwise associated with the device 30. The sensor 32 may be operative to sense at least one electrical, mechanical, and/or chemical parameter value of the machine 20. Alternatively or additionally, the sensor 32 may be operative to sense at least one electrical, mechanical, and/or chemical parameter value of a component 24 that is electrically coupled to the machine 20, such as a power line. Alternatively or additionally, the sensor 32 may be operative to sense at least one electrical, mechanical, and/or chemical parameter value of a component 25 that is mechanically coupled to the machine 20, such as a support or connecting structure attached to the machine 20.

[0111] The sensor 32 provides sensor data that are distinct from the operational parameter(s) of the machine 20, but which are influenced by the operational parameter(s) of the machine 20.

[0112] The sensor 32 may provide sensor data in the time domain. The sensor data may be time series data. A sampling interval between successive datums of the sensor data may be shorter, in particular much shorter, than the running speed of the machine 20.

[0113] The device 30 comprises one or several integrated circuits (ICs) 33. The ICs may comprise one or several application specific integrated circuits (ASICs), processors, controllers, microprocessors, microcontrollers, or a combination thereof. The one or several ICs 33 operate as a processing device that computes a spectral representation of the sensor data, applies a compression algorithm, and adapts the compression algorithm in dependence on the one or several operating parameters of the machine 20, as will be described in more detail with reference to FIGS. 2 to 9.

[0114] The device 30 comprises a memory 34. The memory 34 may comprise a non-volatile memory. The memory 34 may store frequency-domain data that may be used for compressing the spectral representation of the sensor data. For illustration, one or several reference spectra associated with typical operating conditions (such as normal operation and/or various faults) may be stored in the memory 34, for use in the compression algorithm.

[0115] The one or several ICs 33 receive the time-domain sensor data and process the time-domain sensor data into a compressed sensor data representation, using a compression algorithm that is a function of the one or several operating parameters of the machine 20.

[0116] The device 30 may comprise output circuitry 35. The output circuitry 35 may comprise a transmit circuit, which may include a modulator and an antenna, to transmit the compressed sensor data representation.

[0117] The device 30 may include a further interface that allows machine specifics to be entered to the device 30. The machine specifics may reflect the specific use to which the machine 20 is installed (e.g., waste water pumping, fan, wind turbine), and/or specific fault cases (which may be based on a history of similar machines and uses), and/or specific ambient conditions (such as whether the machine 20 is stationarily mounted, mounted on a train, or mounted on a boat).

[0118] The device 30 may comprise input circuitry to receive from an analytics computer 40 feedback information. The IC(s) may use the feedback information to control the compression algorithm. The input circuitry and the output circuitry 35 in combination may form a wireless interface.

[0119] The system 10 comprises an analytics computer 40, which may be a server. The analytics computer 40 may have one or several processors to process the compressed sensor data representation received from the device 30 via a wired or wireless communication channel 41. The analytics computer 40 is adapted to reconstruct the sensor data (it being understood that there will typically be some information loss due to the compression) from the compressed sensor data representation and to further analyze the sensor data. The further analysis may include aggregation of sensor data from a large number, e.g., at least ten or at least fifty, sensors, and processing the aggregated sensor data.

[0120] Based on the analysis of the sensor data, the analytics computer 40 may identify existing or predicted future operating conditions of the machine 20 or of a system including the machine 20. The analytics computer 40 may suggest control parameters or may automatically set control parameters of the machine 20 or of a system including the machine 20.

[0121] The analytics computer 40 may be operative to determine a key performance indicator (KPI) of the machine 20. Various ways of determining KPIs of machines 20, such as generators or motors, are known to the skilled person, depending on the specific machine 20 and its application.

[0122] The analytics computer 40 may provide feedback to the device 30. The feedback may be indicative of an assessment of the quality of the reconstructed signal data determined by the analytics computer 40. In response to the feedback from the analytics computer 40, the IC(s) 33 of the device 30 may control the compression algorithm.

[0123] Operation of the device 30 will be described in more detail with reference to FIGS. 2 to 9 below.

[0124] FIG. 2 is a data flow diagram representing the data processing by the device 30 and the analytics computer 40.

[0125] The device 30 may capture the sensor data 51 or may receive the sensor data 51. The sensor data 51 may be time series data.

[0126] The device 30 may compute a spectral representation 52 of the sensor data 51. Computing the spectral representation may comprise performing a Fourier transform (which may also be implemented as a Fast Fourier Transform (FFT)), a Laplace transform, or other transforms from the time domain to the frequency domain.

[0127] As used herein, the term “spectral representation” refers to sensor data as a function of frequency, i.e., in the frequency domain. The spectral representation may be the result of a transformation of time-domain sensor data into the frequency domain. The spectral representation may be complex-valued, i.e., may include both a magnitude and a phase of a complex number for each frequency.

[0128] The device 30 applies a compression algorithm 55 to the spectral representation 53. The compression algorithm 55 is executed such that it depends on the operational parameter(s) of the machine 20. The mathematical operations applied to the spectral representation 51 by the compression algorithm 55 may be dependent on the operational parameter(s) of the machine 20. A change in the operational parameter(s), e.g., an increase of the rotation speed of the rotating component of the machine 20 from a first non-zero rotation speed to a second non-zero rotation speed may cause the mathematical operations applied to the spectral representation 51 by the compression algorithm 55 to change.

[0129] The operational parameter(s) 53 of the machine 20 are used to control the compression algorithm 55. The operational parameter(s) 53 may be used for performing spectrum alignment and/or identifying harmonics in the spectral representation of the sensor data, as will be explained in more detail below.

[0130] Optionally, machine specifics 54 may also be used to control the compression algorithm 55. The machine specifics 54 may be dependent on which possible fault cases can occur. The information on possible fault cases may be based on historical sensor data.

[0131] For illustration, reference spectra representing a set of several fault cases may be stored in the sensor device and may be used when executing the compression algorithm 55. The reference spectra can be or can include recorded spectra obtained from measurements in operating systems, spectra obtained by processing measured spectra (such as by averaging) and/or spectra obtained by performing system simulations.

[0132] The machine specifics 54 may be dependent on the application for which the machine is used. For illustration, the compression algorithm 55 may be controlled differently depending on whether the machine is used for wastewater pumping, a fan, or wind turbine. The machine specifics 54 may be dependent on ambient conditions. For illustration, the compression algorithm 55 may be controlled differently depending on whether the machine is stationary or mobile. The compression algorithm 55 may be controlled differently depending on whether the machine is mounted on a boat or train.

[0133] The output of the compression algorithm 55 is a compressed sensor data representation (it being understood that the compression is performed in the frequency domain, rather than the time domain) The compressed sensor data representation is transmitted to an analytics computer or other recipient, which may perform a reconstruction 56.

[0134] In the reconstruction 56, the sensor data 51 are reconstructed, typically with some loss of information due to the compression, from the compressed sensor data representation. The reconstruction involves transforming the uncompressed data back into the time domain, e.g., by applying an inverse Fourier transform, an inverse FFT, or an inverse Laplace transform.

[0135] The reconstructed sensor data may be used for further analysis and/or control functions 57. For illustration, the reconstructed sensor data may be used for determining a KPI of the machine 20 or of a system including the machine 20.

[0136] Optionally, feedback may be provided based on the results of the analytics or KPI determination performed remotely from the device 30. The feedback may be used by the device 30 to control the compression algorithm 55.

[0137] The compression algorithm 55 makes use of operational parameters 53 and, optionally, specifics 54 of the rotating machine to control the compression. The specifics 54 of the machine 20 may be dependent on the setup of the machine 20, e.g., on its installation location etc. When a feedback mechanism is implemented in which feedback information is input to the device 30 from the analytics computer 40, the compression algorithm 55 can be adapted to produce a compressed representation that gives rise to optimal KPI calculation or analytics.

[0138] The feedback can be implemented offline (e.g., during commissioning, development) or online (i.e., during operation).

[0139] FIG. 3 is a flow chart of a method 60 according to an embodiment. The method 60 may be performed automatically by the device 30.

[0140] At step 61, the device 30 retrieves one or several operational parameter(s) of the running machine 20. The operational parameter(s) may be or may include a running speed of the machine. The running speed may be the rotation speed of a component of the machine 20.

[0141] At step 62, the device 30 may control compression of the spectral representation of sensor data in dependence on the one or several operational parameter(s) of a running machine 20. The device 30 may have an integrated sensor 32 or may be coupled to a separate sensor that outputs the sensor data for compression prior to transmission.

[0142] At step 62, the compression of sensor data may optionally also be controlled in dependence on machine specifics, such as the installation location or possible fault types of the machine 20.

[0143] At step 63, the compressed sensor data representation (which is still in the frequency domain) is transmitted.

[0144] Sensor data sampled during a time interval must be used to determine the spectral representation. The one or several operational parameter(s) of the running machine 20 may, but do not need to be measured during that time interval, as will be explained with reference to FIG. 4.

[0145] FIG. 4 illustrates a time diagram. Sensor data sampled in a time interval 71, which may be a moving window, may be transformed into the spectral representation, so as to transform the sensor data into the frequency domain. When the operational parameter(s) of the running machine 20 are measured at a time 73 within the time interval 71, that measured value of the operational parameter(s) of the running machine 20 may determine how the compression algorithm is controlled for compressing the spectral representation of the sensor data captured within time interval 71. When the operational parameter(s) of the running machine 20 are measured at times 72, 74 before or after the time interval 71, extrapolation or interpolation techniques may be used to infer the operational parameter(s) of the running machine 20 at, e.g., the start, the center, or the end of the time interval 71. The compression algorithm may be controlled for compressing the spectral representation of the sensor data captured within time interval 71, using operations that are dependent on the operational parameter(s) of the running machine 20 extrapolated or interpolated from the measurements at times 72, 74.

[0146] The compression applied to the spectral representation of the sensor data may be adjusted as the operational parameter(s) of the running machine 20 vary. This may be done on an ongoing basis, e.g., intermittently at a repeat interval, and/or in response to trigger events such as detected changes in the operational parameter(s) of the running machine 20 and/or receipt of feedback.

[0147] FIG. 5 is a flow chart of a method 65 according to an embodiment. The method 65 may be performed automatically by the device 30.

[0148] At step 66, the device 30 performs compression of the spectral representation of sensor data in dependence on the one or several operational parameter(s) of the running machine 20. This may be done in such a way that the mathematical operations performed on the spectral representation vary when the one or several operational parameter(s) of the running machine 20 vary, e.g., when the running speed varies from a first non-zero running speed to a second non-zero running speed.

[0149] At step 67, the device 30 determines whether a trigger for a change in the compression algorithm is fulfilled. The trigger may be expiry of a timer, detection of a change in at least one of the operational parameters, detection of a change in at least one of the machine specifics, and/or receipt of feedback.

[0150] At step 68, the device 30 adjusts the compression in dependence on the value of the operational parameter(s) at the time at which the sensor data to be processed were captured.

[0151] The method returns to step 66.

[0152] It will be appreciated that the operational parameter(s) of the running machine 20 may be used in various ways to adjust the compression, i.e., the mathematical operations, applied to the spectral representation of sensor data. The operational parameter(s) may be used to perform spectrum registration, as will be explained with reference to FIGS. 6 and 7. When performing spectrum alignment, a reference spectrum that may be stored in memory 34 and/or the spectral representation of the sensor data may be transformed in dependence on a ratio between the actual running speed (e.g., rotation speed) of the machine 20 and a reference speed at which the reference spectrum has been recorded. This facilitates a meaningful comparison of the spectral representation of the sensor data to one or several reference spectra. Such a comparison may be useful in computing a linear decomposition, without being limited thereto.

[0153] Alternatively or additionally, the operational parameter(s) may be used to determine which peaks in the spectral representation of the sensor data should be included in the compressed sensor data representation, as will be explained with reference to FIGS. 8 and 9.

[0154] FIG. 6 is a flow chart of a method 80 according to an embodiment. The method 80 may be automatically performed by the device 30.

[0155] At step 81, a spectrum registration is performed. The spectrum registration, which may also be referred to as spectra warping, has the effect of warping a spectral representation of the sensor data or the magnitude of the spectral representation to a reference speed.

[0156] This operation requires knowledge of the actual rotating speed of the machine. The actual transformation of the spectrum may be represented as

A.sub.T(k)=A(u(k)),  (1)

where A(.Math.) denotes a reference spectrum, A.sub.T(.Math.) denotes the transformed reference spectrum after spectrum registration, and u(k) is a function that maps one frequency k to another frequency. Typically, u(k) may be a linear function

u(k)=k.Math.λ.  (2)

The factor λ is determined by the ratio of the reference speed of the machine 20 at which the reference spectrum has been recorded and the actual rotation speed of the machine 20 at the time at which the sensor data has been captured.

[0157] FIG. 7 illustrates the effects of spectrum alignment. FIG. 7 shows a magnitude of a spectral representation of the sensor data 91 and a magnitude of a reference spectrum 92, respectively as a function of frequency. The alignment may be implemented in such a way that the spectral representation of the sensor data 91 is not modified, but the magnitude of the reference spectrum 92 is transformed in accordance with Equation (1). This has the effect that a peak in the reference spectrum 92 originally located at a frequency ω.sub.ref is shifted, in the transformed reference spectrum 93, to a frequency ω′.sub.ref that corresponds to the actual rotating speed ω.sub.act of the machine 20 when recording the sensor data of the spectral representation 91.

[0158] It will be appreciated that the transform for spectrum registration explained above is dependent on the actual running speed of the machine 20 and will vary when the actual running speed of the machine 20 varies, thereby affecting the compression algorithm.

[0159] An alignment or registration of the spectra allows the spectral representation of the sensor data to be quantitatively compared to one or several reference spectra. This may be particularly useful when determining coefficients of a linear decomposition.

[0160] While FIG. 7 illustrates a case in which the reference spectra are transformed to perform frequency alignment with the spectral representation of the sensor data, in alternative implementations the spectral representation of the sensor data may be transformed to ensure frequency alignment with one or several reference spectra.

[0161] At step 82, the results of the spectrum registration may be used for performing compression. For illustration, a linear decomposition of the spectral representation of the sensor data may be determined. This may be done by determining decomposition coefficients c.sub.i such that

Σ.sub.k∥SD(k)−Σ.sub.i=1.sup.Nc.sub.i.Math.A(u.sub.i(k))∥.sup.2  (3)

is minimum. In Equation (3),
k denotes the frequency (it being understood that the sum would be replaced by an integral for continuous frequencies);
SD(.Math.) denotes the spectral representation of the sensor data;
A.sub.i(.Math.) denotes an i.sup.th reference spectrum out of a set of N reference spectra;
u.sub.i(.Math.) is a function that performs frequency alignment of the i.sup.th reference spectrum with the spectral representation of the sensor data, which depends on the ratio of the reference speed of the machine 20 at which the i.sup.th reference spectrum has been recorded and the actual rotation speed of the machine 20 at the time at which the sensor data has been captured, as has been explained above; and
c.sub.i denote the linear decomposition coefficients that are to be determined and that, in combination, may be transmitted as or in the compressed sensor data representation.

[0162] Optionally, the set of decomposition coefficients may be input to a further compression algorithm, such as lzw, zip, run length encoding or other techniques, to generate the compressed sensor data representation from the set of decomposition coefficients.

[0163] When the set of decomposition coefficients is further compressed, lossless compression may be applied to generate the compressed sensor data representation from the set of decomposition coefficients.

[0164] It will be appreciated that various modifications can be used. For illustration, regularization may be used by determining decomposition coefficients c.sub.i such that

Σ.sub.k∥SD(k)−Σ.sub.i=1.sup.Nc.sub.i.Math.A(u.sub.i(k))∥.sup.2+R(c.sub.1,c.sub.2, . . . )  (4)

is minimum, where R(.Math.) denotes a regularization term. Alternatively or additionally, while spectral decomposition may be computed so as to minimize the L.sub.2 metric, as illustrated with reference to Equations (2) and (3), the linear decomposition may be computed such that it minimizes another metric, such as L.sub.n metrics, without being limited thereto. Equations (5) and (6) are exemplary for cost functions that can be minimized in step 82, where M(.Math.) denotes a metric that quantifies a deviation of the spectral representation of the sensor data from the linear decomposition:

M(SD(k),Σ.sub.i=1.sup.Nc.sub.i.Math.A(u.sub.i(k)))  (5)

M(SD(k),Σ.sub.i=1.sup.Nc.sub.i.Math.A(u.sub.i(k)))+R(c.sub.1,c.sub.2, . . . )  (6)

[0165] Various techniques can be employed to solve the optimization problem of any one of Equations (3) to (6), as will be appreciated. For example, gradient descent techniques may be used.

[0166] Other terms may be used in the cost function that is minimized, e.g., to weight the importance of different fault cases and their associated reference spectra A.sub.i(.Math.).

[0167] It will be appreciated that, by representing each obtained spectrum by a combination of a few carefully selected prototype reference spectra, significant compression can be obtained. It is sufficient to transmit the linear decomposition coefficients. The reference spectra can be chosen so as to represent a set of possible failure cases. This may be done so as to provide maximum information in a small set of linear decomposition coefficients. Various techniques, such as archetype analysis, ICA, PCA, and/or NMF may be used.

[0168] The set of reference spectra A.sub.i(.Math.) denote a basis for the linear decomposition and are chosen such as to allow for informative determination of failure cases and/or other suitable analytics.

[0169] FIG. 8 is a flow chart of a method 100 according to an embodiment. The method 100 may be automatically performed by the device 30.

[0170] The method 100 operates on a spectral representation of the sensor data. The actual rotation speed information of the machine 20 may be used to find all harmonic peaks within the obtained spectral representation of the sensor data. Peak characteristics (such as maximum and peak width and/or peak area) may be transmitted in the compressed sensor data representation.

[0171] Information on the signal background/floor may additionally be included in the compressed sensor data representation.

[0172] Additionally or alternatively, the peak characteristics and/or signal background/floor may be input to a compression algorithm, in particular a lossless compression algorithm, to generate the sensor data representation.

[0173] The following processes may be used.

[0174] Prior to step 101, a background estimation may optionally be performed. This may comprise smoothing of a magnitude of the spectral representation of the sensor data. A width of the smoothing function or filter that is applied to the spectrum magnitude of the spectral representation of the sensor data is broader than the width of the individual peaks.

[0175] At step 101, peaks may be identified. This may comprise determining frequencies at which the magnitude of the spectral representation is greater than the background by at least a certain threshold factor. Alternatively or additionally, harmonic peak locations derived from the running speed of the machine 20 may be used to identify the peaks.

[0176] After step 101 and prior to step 102, the set of peaks identified at step 101 may optionally be reduced. This may include, without limitation, non-maxima suppression and/or reducing multiple neighboring peak values to a single peak representation.

[0177] At step 102, a sub-set of the previously identified peaks may be selected for compression. The size of the sub-set may be determined in dependence on a desired compression ratio. Peak characteristics, such as peak maximum value and peak width (e.g., full width at half max) and/or spectral weight (i.e., the area of the peak) may be determined for the sub-set.

[0178] At step 103, the magnitude of the spectral representation without the sub-set of peaks may be approximated. This may comprise determining a piece-wise constant floor signal, that may optionally have a varying size of the constant portions.

[0179] At step 104, the peak characteristics of the sub-set selected at step 102 is transmitted in the compressed sensor data representation. The peak characteristics may include the peak maximum value and peak width (e.g., full width at half max) or spectral weight (i.e., the area of the peak) for each peak in the sub-set, but not for other peaks. Compressed information on the background may optionally also be included in the compressed sensor data representation. The compressed information on the background may define a piece-wise constant function.

[0180] The compressed sensor data representation may be the peak characteristics and optionally the compressed information on the background. Alternatively or additionally, the peak characteristics and optionally the compressed information on the background may be input to a compression algorithm to generate the compressed sensor data representation.

[0181] FIG. 9 illustrates the use of the actual running speed of the machine 20 when computing a small number of parameters that include the essential information contained in a spectral representation 111 of the sensor data. The spectral representation 111 includes dominant peaks 112-115 and smaller peaks.

[0182] Compression of the spectral representation 111, using the technique of FIG. 8, includes determining a sub-set of all peaks, using the actual rotating speed. For illustration, the dominant peaks 112-115 may be easily identified thereby. The magnitude between these peaks may be approximated by piece-wise constant functions, which may, but do not need to extend from one peak to another peak.

[0183] FIG. 9 also illustrates the reconstructed spectrum 120 that may be determined from a compressed sensor data representation. Peak heights and peak widths of the dominant peaks 122-125 provide an approximation to the dominant peaks 112-115 in the spectral representation. Piece-wise constant functions 126-130 approximate the background between the dominant peaks. The essential characteristics of the spectrum 111 may be provided for remote analysis, while attaining a high compression ratio.

[0184] By applying conventional compression techniques to the peak heights, peak widths and optional the background signal levels 126-130, the amount of data to be transmitted may be reduced further.

[0185] The selection of peaks for which peak characteristics are to be transmitted may also be dependent on the KPI or other analysis that is performed on the receiving side of the transmission.

[0186] Embodiments of the invention are adapted for compression of spectral information of recorded data to reduce the amount of bulk data transfer. The compression is dynamically controlled by operational parameters of the machine, e.g. speed of rotation, to achieve high compression rates and fidelity, adaptive to the actual machine. Compressed data provides richer information for remote analytics than a few predefined parameters, but still reduces the amount of data to be transferred and thus saves energy and bandwidth. Optionally, compression can be specifically optimized to allow for high-fidelity reconstruction of pre-defined condition parameters at the remote end compared to wasteful generic signal reconstruction.

[0187] The compression may be performed on a sensor device for cloud-based or on-premise analytics.

[0188] Additional embodiments are defined by the following list of aspects:

[0189] 1. A method of processing sensor data for transmission, the method comprising:

[0190] receiving, by at least one processing device, one or several operational parameters of a machine, wherein the one or several operational parameters influence the sensor data but are different from the sensor data;

[0191] applying, by the at least one processing device, a compression technique to a spectral representation of the sensor data to generate a compressed sensor data representation, wherein the applied compression technique is dependent on one or several operational parameters of the machine; and

[0192] transmitting, by the at least one processing device, the compressed sensor data representation.

[0193] 2. The method of aspect 1, wherein the sensor data is time-domain data.

[0194] 3. The method of aspect 2, wherein the method further comprises generating the spectral representation of the sensor data by transforming the time-domain data into a frequency domain.

[0195] 4. The method of aspect 3, wherein transforming the time-domain data into a frequency domain comprises performing a Fourier transform, a Fast Fourier transform, or a Laplace transform.

[0196] 5. The method of any one of the preceding aspects, wherein the machine has a movable component.

[0197] 6. The method of aspect 5, wherein the movable component reciprocates or rotates.

[0198] 7. The method of aspect 5 or aspect 6, wherein the one or several operational parameters of the machine comprise a frequency at which the movable component reciprocates or rotates.

[0199] 8. The method of any one of the preceding aspects, further comprising modifying the applied compression technique responsive to a change in the running speed of the machine.

[0200] 9. The method of any one of the preceding aspects, wherein the applied compression technique is further dependent on one or several machine specifics of the machine.

[0201] 10. The method of aspect 9, wherein the one or several machine specifics are selected from a group consisting of fault cases, application type of the machine, ambient conditions.

[0202] 11. The method of aspect 9 or aspect 10, wherein reference spectra representing a set of several fault cases are stored in the processing device and are used when applying the compression technique.

[0203] 12. The method of any one of aspects 9-11, wherein the compression technique is executed differently depending on the specific application of the machine.

[0204] 13. The method of any one of aspects 9-12, wherein the compression technique is executed differently depending on whether the machine is stationary or mobile.

[0205] 14. The method of any one of aspects 9-12, wherein the compression technique is executed differently depending on whether the machine is mounted on a boat or train.

[0206] 15. The method of any one of the preceding aspects, further comprising

[0207] receiving, by a data analytics computer, the compressed sensor data representation and

[0208] analyzing, by the data analytics computer, the compressed sensor data representation.

[0209] 16. The method of aspect 15, wherein the data analytics computer determines at least one key performance indicator, KPI, of the machine.

[0210] 17. The method of any one of aspects 15-16, further comprising

[0211] modifying, by the processing device, the applied compression technique responsive to feedback information from the data analytics computer.

[0212] 18. The method of aspect 17, wherein the feedback information is provided online.

[0213] 19. The method of aspect 17, wherein the feedback information is provided offline, in particular during system engineering or commissioning.

[0214] 20. The method of any one of the preceding aspects,

[0215] wherein the one or several operational parameters comprise a rotation speed of a rotating component of the machine, and

[0216] wherein applying the compression technique comprises:

[0217] applying an alignment transformation that is dependent on the rotation speed to the spectral representation of the sensor data to align the spectral representation of the sensor data with at least one reference spectrum of a set of reference spectra.

[0218] 21. The method of aspect 20,

[0219] wherein applying the compression technique comprises:

[0220] determining a set of decomposition coefficients of a linear decomposition of the spectral representation of the sensor data.

[0221] 22. The method of aspect 21, comprising

[0222] determining decomposition coefficients c.sub.i such that

Σ.sub.k∥SD(k)−Σ.sub.i=1.sup.Nc.sub.i.Math.A(u.sub.i(k))∥.sup.2+R(c.sub.1,c.sub.2, . . . )

is minimum, wherein
k denotes frequency;
SD(.Math.) denotes the spectral representation of the sensor data;
A.sub.i(.Math.) denotes an i.sup.th reference spectrum out of a set of N reference spectra;
u.sub.i(.Math.) is a function that performs frequency alignment of the i.sup.th reference spectrum with the spectral representation of the sensor data and which depends on the ratio of the reference speed of the machine at which the i.sup.th reference spectrum has been recorded and the actual rotation speed of the machine at the time at which the sensor data has been captured;
c.sub.i denote the linear decomposition coefficients that are to be determined and that, in combination, may be transmitted as compressed data; and
R(.Math.) denotes a regularization term that may be present, but which may also be omitted;
or

[0223] determining decomposition coefficients c.sub.i, such that

M(SD(k),Σ.sub.i=1.sup.Nc.sub.i.Math.A(u.sub.i(k)))+R(c.sub.1,c.sub.2, . . . )

is minimum, wherein
M(.Math.,.Math.) denotes a distance metric in frequency space;
k denotes frequency;
SD(.Math.) denotes the spectral representation of the sensor data;
A.sub.i(.Math.) denotes an i.sup.th reference spectrum out of a set of N reference spectra;
u.sub.i(.Math.) is a function that performs frequency alignment of the i.sup.th reference spectrum with the spectral representation of the sensor data and which depends on the ratio of the reference speed of the machine at which the i.sup.th reference spectrum has been recorded and the actual rotation speed of the machine at the time at which the sensor data has been captured;
c.sub.i denote the linear decomposition coefficients that are to be determined and that, in combination, may be transmitted as compressed data; and
R(.Math.) denotes a regularization term that may be present, but which may also be omitted.

[0224] 23. The method of aspect 21 or aspect 22, wherein the set of decomposition coefficients is transmitted as the compressed sensor data representation or wherein the method comprises generating the compressed sensor data representation from the set of decomposition coefficients, in particular by applying a compression algorithm to the set of decomposition coefficients.

[0225] 24. The method of any one of the preceding aspects, wherein the one or several operational parameters comprise a rotation speed of the machine,

[0226] wherein applying the compression technique comprises identifying, based on the rotation speed, a set of peaks in the spectral representation of the sensor data and determining peak characteristics for each peak in the identified set of peaks.

[0227] 25. The method of aspect 24, wherein the peak characteristics of the peaks included in the identified set of peaks are transmitted as the compressed sensor data representation.

[0228] 26. The method of aspect 24 or aspect 25, wherein the peak characteristics comprise a peak maximum and peak width or a peak area or wherein the method comprises generating the compressed sensor data representation from the peak characteristics, in particular by applying a compression algorithm.

[0229] 27. The method of any one of aspects 24-26, wherein applying the compression technique comprises approximating a background of the spectral representation between the set of peaks.

[0230] 28. The method of claim 27, wherein the background is approximated by piece-wise constant functions.

[0231] 29. The method of claim 28, wherein the compressed sensor data representation includes information on the piece-wise constant functions.

[0232] 30. The method of any one of the preceding aspects, wherein the at least one processing device is a field sensor device.

[0233] 31. The method of any one of the preceding aspects, wherein the machine is a generator.

[0234] 32. The method of any one of the preceding aspects, wherein the machine is a motor.

[0235] 33. The method of any one of the preceding aspects, further comprising:

[0236] using the compressed sensor data representation for controlling the machine or a component of a system in which the machine is installed.

[0237] 34. A device for processing sensor data for transmission, comprising:

[0238] an interface adapted to receive one or several operational parameters of a machine, wherein the one or several operational parameters influence the sensor data but are different from the sensor data;

[0239] at least one processing circuit adapted to

[0240] determine a compression technique that is to be applied to the sensor data as a function of the one or several operational parameters of the machine, and

[0241] apply the determined compression technique to a spectral representation of the sensor data to generate a compressed sensor data representation; and

[0242] output circuitry adapted to transmit the compressed sensor data representation.

[0243] 35. The device of aspect 34, wherein the sensor data is time-domain data.

[0244] 36. The device of aspect 35, wherein the at least one processing circuit is adapted to generate the spectral representation of the sensor data by transforming the time-domain data into a frequency domain.

[0245] 37. The device of aspect 36, wherein transforming the time-domain data into a frequency domain comprises performing a Fourier transform, a Fast Fourier transform, or a Laplace transform.

[0246] 38. The device of any one of aspects 34-37, wherein the machine has a movable component.

[0247] 39. The device of aspect 38, wherein the movable component reciprocates or rotates.

[0248] 40. The device of aspect 38 or aspect 39, wherein the one or several operational parameters of the machine comprise a frequency at which the movable component reciprocates or rotates.

[0249] 41. The device of any one of aspects 34-40, wherein the at least one processing circuit is adapted to modify the determined compression technique responsive to a change in the running speed of the machine.

[0250] 42. The device of any one of aspects 34-41, wherein the applied compression technique is further dependent on one or several machine specifics of the machine.

[0251] 43. The device of aspect 42, wherein the one or several machine specifics are selected from a group consisting of fault cases, application type of the machine, ambient conditions.

[0252] 44. The device of aspect 42 or aspect 43, wherein reference spectra representing a set of several fault cases are stored in the processing device and are used when applying the compression technique.

[0253] 45. The device of any one of aspects 42-44, wherein the at least one processing circuit is adapted to execute the compression technique differently depending on the specific application of the machine.

[0254] 46. The device of any one of aspects 42-45, wherein the at least one processing circuit is adapted to execute the compression technique differently depending on whether the machine is stationary or mobile.

[0255] 47. The device of any one of aspects 42-46, wherein the at least one processing circuit is adapted to execute the compression technique differently depending on whether the machine is mounted on a boat or train.

[0256] 48. The device of any one of aspects 34-47, wherein the at least one processing circuit is adapted to modify the applied compression technique responsive to feedback information.

[0257] 49. The device of aspect 48, wherein the feedback information is provided online.

[0258] 50. The device of aspect 49, wherein the feedback information is provided offline, in particular during system engineering or commissioning.

[0259] 51. The device of any one of aspects 34-50,

[0260] wherein the one or several operational parameters comprise a rotation speed of a rotating component of the machine, and

[0261] wherein the at least one processing circuit is adapted to apply an alignment transformation that is dependent on the rotation speed to the spectral representation of the sensor data to align the spectral representation of the sensor data with at least one reference spectrum of a set of reference spectra.

[0262] 52. The device of aspect 51,

[0263] wherein the at least one processing circuit is adapted to determine a set of decomposition coefficients of a linear decomposition of the spectral representation of the sensor data.

[0264] 53. The device of aspect 51 or 52, wherein the at least one processing circuit is adapted to determine decomposition coefficients c.sub.i such that

Σ.sub.k∥SD(k)−Σ.sub.i=1.sup.Nc.sub.i.Math.A(u.sub.i(k))∥.sup.2

is minimum, wherein
k denotes frequency;
SD(.Math.) denotes the spectral representation of the sensor data;
A.sub.i(.Math.) denotes an i.sup.th reference spectrum out of a set of N reference spectra;
u.sub.i(.Math.) is a function that performs frequency alignment of the i.sup.th reference spectrum with the spectral representation of the sensor data and which depends on the ratio of the reference speed of the machine at which the i.sup.th reference spectrum has been recorded and the actual rotation speed of the machine at the time at which the sensor data has been captured; and
c.sub.i denote the linear decomposition coefficients that are to be determined and that, in combination, may be transmitted as compressed data.

[0265] 54. The device of aspect 52 or aspect 53, wherein the device is adapted to transmit the set of decomposition coefficients as the compressed sensor data representation.

[0266] 55. The device of any one of aspects 34-54, wherein the one or several operational parameters comprise a rotation speed of the machine,

[0267] wherein the at least one processing circuit is adapted to identify, based on the rotation speed, a set of peaks in the spectral representation of the sensor data and determine peak characteristics for each peak in the identified set of peaks.

[0268] 56. The device of aspect 55, wherein the at least one processing circuit is adapted to transmit the peak characteristics of the peaks included in the identified set as the compressed sensor data representation.

[0269] 57. The device of aspect 55 or aspect 56, wherein the peak characteristics comprise a peak maximum and peak width or a peak area.

[0270] 58. The device of any one of aspects 56-57, wherein the at least one processing circuit is adapted to approximate a background of the spectral representation between the set of peaks.

[0271] 59. The device of claim 58, wherein the background is approximated by piece-wise constant functions.

[0272] 60. The device of claim 59, wherein the compressed sensor data representation includes information on the piece-wise constant functions.

[0273] 61. The device of any one of aspects 34-60, wherein the at least one processing device is a field sensor device.

[0274] 62. The device of any one of aspects 34-61, wherein the device has a wireless interface comprising the output circuitry.

[0275] 63. The device of any one of aspects 34-62, further comprising a battery.

[0276] 64. The device of any one of aspects 34-63, wherein the device is battery-powered.

[0277] 65. A system, comprising:

[0278] a machine having a rotating component, in particular a motor or a generator; and

[0279] the device of any one of aspects 34-64.

[0280] 66. The system of aspect 65, wherein the machine is a generator.

[0281] 67. The system of aspects 65, wherein the machine is a motor.

[0282] 68. The system of any one of aspects 65-67, further comprising

[0283] a data analytics computer adapted to receive the compressed sensor data representation and to analyze the compressed sensor data representation.

[0284] 69. The system of aspect 68, wherein the data analytics computer is adapted to determine at least one key performance indicator, KPI, of the machine.

[0285] 70. The system of any one of aspects 65-69, further comprising:

[0286] a control center adapted to control the machine or a component of a system in which the machine is installed using the compressed sensor data representation.

[0287] While the invention has been described in detail in the drawings and foregoing description, such description is to be considered illustrative or exemplary and not restrictive. Variations to the disclosed embodiments can be understood and effected by those skilled in the art and practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims. In the claims, the word “comprising” does not exclude other elements or steps, and the indefinite article “a” or “an” does not exclude a plurality. The mere fact that certain elements or steps are recited in distinct claims does not indicate that a combination of these elements or steps cannot be used to advantage, specifically, in addition to the actual claim dependency, any further meaningful claim combination shall be considered disclosed.