METHOD AND SYSTEM FOR DETECTION OF FAULTS IN PUMP ASSEMBLY VIA HANDHELD COMMUNICATION DEVICE

Abstract

A method for detecting faults or operational parameters in a pump assembly by use of a handheld communication device is described. The pump assembly includes an electric motor and a pump, wherein the pump assembly or electric motor has at least one rotating shaft The method comprises the steps of: a) measuring a sound signal emanating from the pump assembly by use of a microphone connected to or implemented in the handheld communication device, b) processing the measured sound signal, and c) recognising one or more sound emanating condition including any possible faults by way of the processed sound signal. The app automatically repeats at least steps b) and c) for a plurality of preselected frequency ranges in order to detect different fault states.

Claims

1. A method for detecting faults in a pump assembly including an electric motor and a pump by use of a handheld communication device running an app, the method comprising: a) measuring a sound signal emanating from the pump assembly by use of a microphone connected to or implemented in the handheld communication device, b) processing the measured sound signal, and c) recognizing one emanating condition including a possible fault state by way of the processed sound signal, wherein the app automatically repeats at least steps b) and c) for a plurality of preselected frequency ranges in order to detect different fault states.

2. A method according to claim 1, wherein the method is carried out by carrying out a first measurement for a first duration of time for detecting a first fault state, and then carrying out a second measurement for a second duration of time for detecting a second fault state.

3. A method according to claim 2, wherein the first measurement and the second measurement are carried out at two different positions relative to the pump assembly.

4. A method according to claim 1, wherein the method is carried out by sequentially executing scans for individual fault states.

5. A method according to claim 4, wherein the method is carried out by carrying out a first scan in a first frequency range for detecting a first fault state and then carrying out a second scan in a second frequency range for detecting a second fault state.

6. A method according to claim 5, wherein the first frequency range is located in a first band near or in the kHz-range, advantageously up to 1 kHz, and the second frequency range is located in a second band in the kHz-range, advantageously 10-20 kHz.

7. A method according to claim 1, wherein steps b) and/or c) are carried out via a processing unit, such as a DSP, implemented in the handheld communication device and/or a software app installed on the handheld communication device.

8. A method according to claim 1, wherein fault states are identified among the group of: bearing faults, cavitation, dry running, water hammering and unbalance.

9. A method according to claim 8, wherein the method at least carries out sequential detection of water hammering and cavitation.

10. A method according to claim 1, wherein the processing step of step b) comprises the sub-steps of processing the measured sound signal so as to estimate the rotational speed of the rotating shaft and optionally normalizing the measured sound signals or processed sound signals and/or wherein the processing step of step b) optionally comprises the step of filtering out periodic signals of the processed signal, and wherein the recognition of step c) is carried out by use of the periodic signals.

11. A method according to claim 1, wherein the handheld communication device provides a feedback, e.g. an audial, visual or vibrational feedback, when an acceptable measurement position of the microphone has been found.

12. A fault detection system for detecting faults in a pump assembly, the fault detection system comprising: a pump; an electric motor; a handheld communication device, which includes a microphone for measuring sound emanating from the pump assembly or electric motor; a processing unit implemented in the handheld communication device; a software app installed on the handheld communication device for processing a measured sound signal measured via said microphone; a recognition module for recognizing a fault condition by way of the processed sound signal, wherein the fault detection system is configured to automatically repeat a sequence of fault detection steps by processing the measured sound signal and recognizing one emanating condition including a possible fault state in a plurality of preselected frequency ranges in order to detect different fault states.

13. A fault detection system according to claim 12, wherein the recognition module is implemented on the handheld communication device.

14. A fault detection system according to claim 12, wherein the processing unit and/or filter module comprises an analysis module chosen from the group of: an RMS level detection module; a spectral analysis module; an envelope analysis module; or a Cepstral analysis module.

15. A non-transitory computer readable storage medium storing one or more programs, which when executed by a handheld communication device cause the handheld communication device to perform a method for a pump assembly including an electric motor and a pump by use of the handheld, the method comprising: a) measuring a sound signal emanating from the pump assembly by use of a microphone connected to or implemented in the handheld communication device, b) processing the measured sound signal, and c) recognizing one emanating condition including a possible fault state by way of the processed sound signal, wherein steps a)-c) are repeated for a plurality of different frequency ranges in order to detect different fault states.

Description

BRIEF DESCRIPTION OF THE FIGURES

[0085] The invention is explained in detail below with reference to embodiments shown in the drawings, in which

[0086] FIG. 1 shows an embodiment of a fault detection system according to the invention,

[0087] FIG. 2 shows an embodiment of the fault detection system with a first type of pump assembly,

[0088] FIG. 3 shows an embodiment of the fault detection system with a second type of pump assembly,

[0089] FIG. 4 illustrates components of a handheld communication device part of the fault detection system, and components and sound sources of a pump assembly,

[0090] FIG. 5 shows a handheld communication device with a Graphical User Interface,

[0091] FIG. 6 illustrates steps involved in estimating the rotational speed of a shaft in a pump assembly by use of measured sound signals,

[0092] FIG. 7 illustrates the steps in a condition monitoring system,

[0093] FIG. 8 illustrates a fault in a bearing of a pump assembly and the associated sounds emitted due said fault,

[0094] FIG. 9 shows a bearing in perspective,

[0095] FIG. 10 illustrates the steps in an envelope analysis method,

[0096] FIG. 11 illustrates diagrams obtained from spectral analyses and Cepstral analyses,

[0097] FIG. 12 illustrates a normalisation pre-process for use in Cepstral analysis,

[0098] FIG. 13 illustrates the steps in a method of detecting a fault in a pump assembly, and

[0099] FIG. 14 illustrates the steps in a method of sequentially detecting individual fault states in a pump assembly.

DETAILED DESCRIPTION OF THE INVENTION

[0100] FIG. 1 shows a fault detection system 10 according to the invention for detecting faults in a pump assembly 20, which comprises a pump and an electrical motor (not shown). The pump assembly 20 is installed in a pipe system with pipes 15, 16. The system comprises a handheld communication device 30, e.g. in form of a smart phone as illustrated or a tablet computer or the like. The handheld communication device 30 comprises an externally coupled microphone 32, which may be used for contactless measuring sound signals emanating from the pump assembly 20. The system further comprises an external server 40 or cloud. The pump assembly 20 is a circular pump typically used for circulating water in heating systems.

[0101] In a first embodiment, the pump assembly 20 is of a type as shown in FIG. 2, where the pump and electrical motor are implemented into a common housing 21. In a second embodiment shown in FIG. 3, the pump assembly 20′ is of a type, wherein the pump and the electrical motor are separated into a pump housing 22′ and an electrical motor housing 24′ respectively.

[0102] Thus, in difference to the pump assembly of FIGS. 1 and 2, the pump assembly 20′ has distinct and separated sources of acoustical noise. Noises from ball bearings, stator, rotor and other parts will come from housing 24′, and electrical noise will emanate from the terminal box 25′, which incorporates control electronics for the pump assembly. Noise from the fluid in the pipes, i.e. flow noise, emanates from the pump housing 22′ or from the pipes (not shown) connected to the pump housing. Noise from cavitation, from the impeller of the pump or from water hammer will also emanate from this area of the pump assembly.

[0103] The handheld communication device 30 may comprises a processing unit, such as a microcomputer or a digital sound processor and/or a software application installed on the handheld communication device 30 for processing a sound signal measured via the microphone 32. The measured sound signals or the processed sound signals may be compared to known sound emanating conditions, which are indicative of a fault in the pump assembly by use of a recognition module.

[0104] In FIGS. 1-3, the microphone 32 has been depicted as an externally coupled microphone. However, the invention also contemplates using an internal microphone. The external microphone may for instance be necessary, if the internal microphone of the handheld communication device 30 has particular restrictions on the measurable audio bandwidth or directional restraints.

[0105] FIG. 4 shows one embodiment of a handheld device 130 and a pump assembly 120 for use in the fault detection system 110 according to the invention.

[0106] The handheld communication device 130 comprises a microcomputer 131 for assisting and interpreting data. The fault detection system may comprise a dedicated system, e.g. a microcomputer or digital signal processor particularly designed for detecting fault conditions, but in a preferred embodiment, the fault detection system comprises a dedicated software application, which is installed on the handheld communication device 130.

[0107] As previously mentioned, the handheld communication device 130 comprises a microphone 132 for contactless measuring sounds (illustrated with waves between the handheld communication device and pump assembly) emanating from the pump assembly 120 and which may be indicative of a particular fault condition of the pump assembly 120. The handheld communication device 130 further comprises a display 133 for providing a visual feedback to a service worker using the handheld communication device 130. The handheld communication device 130 additionally comprises an input module 134, whereby the service worker may input information into the handheld communication device 130 or the software application installed on the communication device 130. The input module 134 may be part of the display via for instance a touch display, which is now part of most smart phones and tablet computers. However, in principle the input module may also be a keypad of the communication device 130.

[0108] The handheld communication device 130 further comprises a communication module 135 which allows the handheld communication device 135 to communicate with the pump assembly 120 and/or the external server. The communication device may comprise a number of different communication types, such as but not limited to GSM, CDMA, 3G, 4G, infrared, and Bluetooth®.

[0109] The handheld communication device 130 may further comprise a library 136 stored in the software application. The library may comprise a database over known pump assembly types and models, known faults and related sound emanating conditions, e.g. linked to the particular pump assembly type or model, instruction guides on how to obtain measurements from the particular pump assembly type or model, and instruction guides on how to rectify an identified fault.

[0110] In addition to the microphone 132, the handheld communication device 130 may further comprise a number of other sensors, such as for instance an accelerometer 137. The sensors are advantageously integrated in the handheld communication device 130, however; as with the microphone, it is also possible to use externally coupled sensors. The additional sensors may for instance be used to obtain a secondary verification of the sound measurement. The internal accelerometer 137 may for instance be used for a secondary measurement, where the handheld communication device 130 physically contacts the housing of the pump assembly 120 so as to obtain vibration measurements.

[0111] As further shown in FIG. 4, the pump assembly 120 comprises a number of different features, which may lead to faults and noise sources. The pump assembly 120 comprises a motor and rotor 123 for driving the pump, and which may be the source of emitted sounds and faults. Further, the pump assembly 120 may comprise bearings 126, which along with mechanical seals, are often the source of occurring faults in pump assemblies. The pump may further comprise an impeller 125 with impeller blades or vanes that cause the movement of fluid in the pump. The pump assembly 120 further comprises electronic parts 129 for controlling the pump assembly 120 and acting as an interface between a service worker or external server and the pump assembly 120.

[0112] Further, noise from water in the pipe 127 may be the source of sounds emitted from or near the pump assembly and may vary from turbulent or laminar flows. Further, the pump assembly 120 may be prone to cavitation 128 at certain drive conditions. Cavitation is the formation of vapour cavities in a liquid, such as small air bubbles or voids in water, which are caused by forcing acting upon the liquid. Cavitation usually occurs when a fluid is subjected to rapid pressure change that cause the formation of cavities, when the pressure is relatively low. When subjected to a relative high 5 pressure, the voids implode and can generate intense shockwaves, which may cause wear to for instance the impeller 125 of the pump assembly 120. Noise from cavitation typically appears as small “pops” e.g. in the kHz-band, e.g. 10 kHz-20 kHz. Additionally, water hammering 124 may occur.

[0113] Each noise source has its specific frequency range. Noise from the impeller in the pump typically is the same as the rotational speed of the shaft, i.e. around 3000 RPM. In addition to this, the blades on the impeller also generate noise, the blade frequency being a multiple of the rotational speed. Flow noise, i.e. noise stemming from the liquid flowing in the pipes, is typically white noise, also called 1/f noise, and is in the range of 1 Hz to 25 kHz. The electronics of the pump assembly often generates a broadband noise spectrum due to the use of switching electronics. A typical range is 50 kHz to 200 kHz. Cavitation noise is as mentioned in the range of 10 kHz to 20 kHz. Ball bearings generated noise in the area of 1 kHz to 15 kHz, while the electrical motor and the mechanical rotor creates noise in the range of the rotational speed of the shaft. Water hammer, which in extreme cases can destroy the pump, is detectable in the range of 1 Hz to 300 Hz.

[0114] Overall, the system provides a particularly simple fault detection system 10, 110, where a service worker only needs to bring a handheld communication device 30, 130, such as a smart phone, and possibly an externally connected microphone for identifying faults in a pump assembly 20, 120. The system allows sound signals to be measured by use of an internal or external microphone. Further, by processing the sound signal, it is made possible to detect faults and/or identify fault causes via a non-invasive, contactless method, since no sensors have to be fitted to the pump assembly, which otherwise could influence the measurements, and which are sensitive to the position on the pump assembly.

[0115] The analysis tool may be implemented in a processing unit or software application that is installed on the handheld communication device. Alternatively or in addition thereto, the processing unit or software (or part thereof) may be implemented on an external server unit. Thus, the measured sound signal or partially processed signal may be uploaded to a server or a cloud, where the detected sound signal is processed.

[0116] The system may be configured to identify a number of different fault states, e.g. identified among the group of: bearing faults, cavitation, dry running, water hammering and unbalance.

[0117] The software application installed on e.g. a mobile phone 30 having a display 34 may be provided with a Graphical User Interface (GUI) as for instance shown in FIG. 5, which may show the status of the afore-mentioned fault states. Further, the system may be adapted to detect the rotational speed of the shaft of the pump assembly and may display this on the GUI as well. Further, the GUI may show the measured sound frequency spectrum. The measured sound signals may be normalised in accordance with the detected rotational speed, whereby it is made easier to compare the fault spectrum to know sound emanating conditions.

Estimating the Rotation Speed Based on Spectral Analysis

[0118] When working with a pump assembly as in the present invention, it is often important to know the physical rotation speed of the shaft or RPM. Knowing the electrical frequency of the electrical motor is not enough, as there exist a slip between the electrical and the physical speed for asynchronous motors.

[0119] The physical speed of the shaft can be measured by using a tachometer. The principle is to beam a laser to a reflecting tape on the shaft, and thereby counting the number of reflections during a time interval. However, this will require that a reflecting tape to be mounted on every shaft, and often it will not be practical or even possible to get into contact with a rotating shaft. This is also not aligned with the principle behind the present invention, which is intended to provide a simple tool for the service worker and preferably a non-invasive and contactless method of detecting the rotational speed and a possible fault.

[0120] Accordingly, the fault detection system comprises an algorithm for estimating the rotational speed of the shaft based on sound signals measured with the system according to the invention. The outline of the algorithm is illustrated in FIG. 6.

[0121] The sound signal is measured (in step 210) via the microphone of the handheld communication device and is converted to a digital sample signal (in step 220), which provides an array of samples having a number of samples produced at a given sampling frequency, which is set by the handheld communication device. In order to process the sampled signal, it may be necessary to resample (in step 230) the vector sample to a power of two.

[0122] The frequency band that will contain a signal component related to the RPM will typically be in the range of 10 to 60 Hz. To restrict the frequency analysis to this band, the signal is down-sampled to for instance 128 Hz (in step 240) and keeping the frequency above the Nyquist rate. For some handheld communication devices, it is only possible to detect audible frequencies above 20 Hz. In such a case, it may be necessary to increase the RPM to a speed, where it can be detected.

[0123] In each loop of the program, the down-sampling provides a limited number of samples. To obtain a sufficient frequency resolution, the samples are collected into a circular buffer with a size of for instance 512 samples (in step 250).

[0124] If a frequency analysis is performed on the buffer vector, the corresponding RPM resolution will be ΔRPM=60 (128 Hz/512 points)=15 rpm. This resolution may in some circumstances be too low. To increase the resolution, the length of the signal vector must be increased. However, increasing the signal vector will slow down the estimator. Another solution will be to keep the frequency resolution, but interpolate the number of points in the spectra using zero padding (step 260). The new signal vector will contain a windowed version of the buffer vector—zero padded to for instance 8192 points—which provides a resolution of approximately 1 rpm.

[0125] The interpolated signal is Fourier transformed (in step 270) and the resulting Fourier spectrum will show a clear peak corresponding to the RPM. Thus, the RPM may (in step 290) be found by finding a maximum or peak of the Fourier spectrum (in step 280).

[0126] The vibration or sound spectrum from a running motor will always produce a small signal component corresponding to the rotation frequency because no rotor can be made without a small imbalance. However, other signal components can be generated from other parts of the motor, pump or fluid. To make the estimation of the RPM more robust, a small search window may be introduced. The position of the search window will be controlled from known motor parameters, or if the operator has some knowledge about the expected RPM. The size of the window may for instance be fixed to a width of 200 rpm and the height of the window is set to 1.

[0127] It should be noted that the algorithm for detecting the RPM is generic in nature and not only restricted to sound measurements via a handheld communication device. The estimation may for instance also be based on vibration measurements as for instance described in EP1972793.

[0128] Accordingly, the invention also provides a method of estimating the rotational speed of a shaft of a pump assembly, where a vibration or sound is measured and sampled, resampled, down-sampled, buffered, zero-padded, Fourier transformed, and peak detected.

Fault Detection

[0129] In the following, the fault detection system and method of identifying faults are explained with reference to bearing fault detection. However, the system is also applicable for detecting other types of faults.

[0130] The concept of Condition Monitoring (CM) is the use of advanced technologies in order to determine equipment conditions, and potentially predict failures. A perspective of this early prediction is to change load conditions in order to delay a potential break down until a scheduled maintenance. This will in general improve equipment reliability, minimizing downtime and maximizing component life.

[0131] In the current application, the objectives is to develop a CM system to detect bearing faults based on measured sound signals 310 via a handheld communication device. A typical detector for a CM system could be constructed by the two elements shown in FIG. 7, viz. a feature extraction 320 and a classification 330 of the fault.

[0132] Feature extraction is the process of generating a set of descriptors or characteristic attributes from the sound measurements. The Classification is the process of interpreting and comparing the feature against a set of pre-analyzed features with known causes, in order to estimate a proper diagnosis. This division is advantageous, as it underline the fact, that the best classification algorithm based on a large neural network will make a wrong diagnosis, if the feature extraction algorithm is poorly made.

[0133] A bearing 426 as shown in FIGS. 8 and 9 is a high precision mechanical construction with two raceways and a finite number of balls 440. As the inner 455 and outer raceway 460 are rotating with a relative speed, the corresponding emitted sound signal obtained with a microphone will be a low power white noise-like signal. However, in the case of any abnormality, such as a crack 450 in the outer raceway 460, a transient response will be measured by the microphone every time a ball 440 hits this defect 450. This will make the corresponding sound signal 470 significant with a periodic structure 480 reflecting the time interval between each impact.

[0134] The signal analysis of sound signal data has currently resulted in four different approaches to determine the level of damages, viz. RMS level detection, spectral analysis, envelope analysis, and Cepstral analysis.

RMS Level Detection

[0135] There is a strong negative correlation between the overall sound level for a bearing and the expected lifetime of that bearing. By assessing the sound levels (RMS) and comparing these with some warning levels or threshold levels, a damage to the bearings can be determined. Thus, the RMS level detection provides a particular simple method of determining a fault stage based on signal levels. However, this method has the disadvantage that it is “frequency blind” and can thus not distinguish between different noise sources. Further, the method cannot distinguish between for instance the basis signal and the first harmonic. Thus, it is difficult to provide a robust system and method based on RMS level detection, since the method is not very selective.

Spectral Analysis

[0136] Spectral analysis of a vibration signal makes it possible to differentiate between signal components with different frequencies, as they often relates to different sources of vibration. The analysis is often performed using the Power Spectrum of the signal, which calculate how the signal energy is distributed in the frequency domain

[0137] Based on the bearing geometry, it is possible to calculate four fundamental frequencies, which represent the most common occurrence bearing faults, viz. faults to the outer ring 460, an inner ring 455, a retainer 465 or the balls 440 of the bearings, see also FIG. 9.

[0138] The measured frequency will depends on the number of balls of the bearing, the shaft frequency f.sub.shaft, the ball diameter, the retainer diameter and the ball contact angle. For a given bearing geometry, such as a NSK6305 bearing, faults to the outer ring 460, the inner ring 455, the retainer 465 or the ball 440 may for instance be 3.06, 4.93, 0.38, and 2.03 times the shaft frequency f.sub.shaft. This means that the faults will be located in different parts of the sound frequency spectrum and may thus be used to identify the type of fault.

[0139] Accordingly, the library of the fault detection system may comprise a database of expected spectrums based on bearing types of known pump assembly models.

Envelope Analysis

[0140] This is original an analog approach with roots in communication technology. The sound signal can be interpreted as the result of an Amplitude Modulation (AM) of a carrier in accordance with a modulating wave. The modulated wave/baseband signal will be the impact impulse train, and the carrier will be the ringing/characteristic sound of the bearing and corresponding motor construction. In the case of a narrowband AM wave (large carrier frequency compared with the message bandwidth), the demodulation can be accomplished by using a simple, yet highly effective device known as the envelope detector. Ideally, an envelope detector produces an output signal that follows the envelope of the input signal waveform exactly. The analog version consists of a diode and a resistor-capacitor filter to down mixing and lowpass filtering the signal, but a digital version can be obtained using similar operations.

[0141] The construction of the envelope detector can also be argued using the Fourier Spectrogram. As this procedure provide some insight into the properties and limitations of the envelope approach, a short presentation will be given next. A property of the wideband Fourier Spectrogram is a high time resolution, as the underlying short-time Fourier transform (STFT) utilize a very short window length. In the case of a periodic signal, the sliding window will alternately pass through high energy and low energy signal areas as a function of time. The corresponding spectra will also be alternately high and low energy spectra, and the corresponding Fourier Spectrogram will be dominated by vertical stripes (the 1/T line spectrum). The wideband Fourier Spectrogram of a vibration signal from a bearing with an outer fault will reveal the envelope curve (impact pulse train) in a certain frequency band.

[0142] If the Spectrogram is calculated for a bandpass filtered signal that only contained the frequencies around aforementioned certain frequency band, the envelope curve can be extracted by calculating the time marginal of the spectrogram and may as illustrated in FIG. 10 be carried out without calculation of the spectrogram by bandpass filtering the signal (step 510), absolute square the result (step 520) followed by a low-pass filter (step 530).

[0143] The envelope analysis is then obtained when a frequency analysis is performed on the envelope curve from the detector by a Fourier transformation (step 540) and absolute squaring the result (Step 550).

[0144] The four fundamental frequencies relating to the four bearing fault types will emerge in the frequency spectrum similar to the spectral analysis.

[0145] The envelope analysis approach has the advantage that calculation complexity is simple as it can be done using two filtering operations followed by an FFT. However, the method has the disadvantage in limitation of the selection of the specifications for the bandpass filter. The carrier frequency is not related to the bearing, but depends on the motor construction and placement. It might also change depending on the shaft frequency. This will demand individual setup for each application and may thus be complex.

Cepstral Analysis

[0146] A common sign of all bearing faults are the presence of a periodic structure of the measured sound signal. A property of the Narrowband Fourier Spectrogram is a high frequency resolution, as the underlying STFT utilize a long window length. In the case of a periodic signal, the window will catch several periods of the signal, and the corresponding spectrum will be discrete, thus showing a line spectrum.

[0147] This means that the Fourier representation of a periodic signal will be a pulse train with equidistant contributions. In the Fourier Spectrogram, this pulse train will show up as horizontal stripes as seen in FIG. 11a. A measure of the impact rate will be the distance between any two stripes. But Instead of selecting a pair of frequencies (this would introduce a limitation), an overall estimate of the rate can be obtained by performing a frequency transformation of all the stripes, i.e. a frequency analysis of a frequency representation. This approach is called a Cepstral analysis and the result is illustrated in the Cepstrum shown in FIG. 11b. The Cepstrum is generally found by taking the Inverse Fourier Transformation of the logarithm of the estimated spectrum of the measured sound signal.

[0148] It is interesting to note, that the Cepstral representation not only illustrate the expected contribution for the aforementioned fundamental frequency 610, but also indicate periodic structures with other impact rates 620, 630, 640. This gives a more descriptive signature of a bearing fault.

[0149] The sound signal used in FIG. 11 has been recorded on a motor with a defect bearing (outer raceway). In FIGS. 11c and 11d, the shaft speed is swept from 10 Hz to 50 Hz (500 to 2,500 rpm) during 60 seconds. It is very interesting to inspect the corresponding spectrogram in FIG. 11c. It illustrate that some part of the spectrum are independent of the shaft frequency, as they are placed as horizontal contributions.

[0150] They constitute the resonance area of the motor construction. The spectral contributions that increase proportional with the shaft frequency are related to the bearing. For a fixed time position, the frequency slide constitute a discrete spectrum that are related to a periodic signal, and a bearing signal is only periodic when the bearing are defect. To obtain the Cepstral representation in FIG. 11d, each column of the spectrogram is high pass filtered and Fourier transformed. The Cepstral representation shows how the impact rate change as a function of time. Often there are more than one impact rates as the vibration signal contain multiple vibration sources. Their position (rate) makes it possible to distinct between different sources.

[0151] As the Cepstral representation is based on a frequency analysis of the frequency representation, it measures the frequency content through the complete domain. If some passages of the spectra are attenuated/amplified as result of resonance in the construction, the level of the corresponding Cepstral will only experience a minor change and no peaks will disappear. This property underline the fact, that the Cepstral domain is a measure of periodicity not frequency.

[0152] The location of the peaks in the Cepstral domain are depending on the shaft RPM (as illustrated in FIG. 11d), and in order to interpreted the Cepstral representation in connection with bearing fault detection, the RPM must be provided in parallel with the Cepstral. This will complicate the task of diagnosing/classification the Cepstral coefficients unnecessary. The location of peaks in a feature for a particular bearing fault must be fixed for variable RPM. This scaling of the Cepstral domain relative to the RPM can in general be obtained in three ways: preprocessing of the vibration data before the Cepstral analysis, scaling one of the frequency axis in the Cepstral calculation or post processing of the Cepstral data. In a preferred embodiment, it has been selected to preprocess the vibration data by resampling in accordance with the detected rotational speed of the shaft, see also previous explanation. The elements of the normalization algorithm is illustrated in FIG. 12.

[0153] The measured sound signal is run through a pre-processing step 700, where the measured signal is normalized in relation to the shaft frequency f.sub.shaft.

[0154] The purpose of the preprocessing is to scale the frequency axis of the sound signal. In order to avoid aliasing, the signal must be properly lowpass filtered (in step 710), before down-sampling. This filtering may be performed by a 20th order Butterworth filter. The classical approach to resampling is to filtering the time discrete signal using an non-causal sinc function as impulse response function (step 720), after which the pre-processed signal may be sent through the Cepstral analysis (in step 730).

[0155] The RPM can in the case of a synchronous motor be obtained from the frequency converter, but in the case of an asynchronous motor, the RPM has to be measured separately or determined in accordance with the previously explained algorithm from the sound signal.

[0156] Once the signal has been normalized and run through the Cepstral analysis, the resulting Cepstrum may be compared to Cepstrums of known operating conditions including fault conditions in order to identify a particular fault. This may for instance be carried out by pattern recognition.

[0157] The fault detection system has now been described for four different analysis methods. However, it is recognized that the fault detection may also use a combination of the various analysis methods and that particular fault types may better be detected with one of said methods.

[0158] The analysis methods have been described for bearing faults, which typically show themselves at relative low sound frequencies. However, other types of faults may be identified in other frequency bands. Cavitation may for instance be detected via sounds in the kHz band, e.g. between 10 kHz and 20 kHz.

Method for Detecting Faults

[0159] The method for detecting faults in a pump assembly may for instance be carried out in accordance with the steps illustrated in FIG. 13.

[0160] In a first step 810, the pump assembly model is input into a software app on the handheld communication device. The pump assembly is optionally run through a pre-routine 820, where the rotational speed of the shaft is increased or set to a speed, where sounds emanating from the pump assembly may be detected via the microphone of the handheld communication device.

[0161] In a third step 830, sounds emanating from the pump assembly are contactless measured via the microphone of the handheld communication device. Between step 820 and 830, the software app may for instance provide the user with a guide on the display showing at which positions the microphone and/or the handheld communication device should be arranged so as to obtain the sound measurements. This step may for instance be carried out by simultaneously carrying out a sweep of the rotational speed of the shaft, such that sound measurements are carried out for an interval of shaft speeds.

[0162] Positioning of the handheld device, and measurement with the device, is done in two distinct steps. The step of positioning may take between 10 seconds and 1 minute, and measurement may take from 5 seconds to 1 minute depending on the number of parameters to be measured. The distance from the device to the pump assembly is from around 1 meter to a few centimetres, typically in the range from 5 cm to 30 cm from the pump assembly.

[0163] When positioning the handheld communication device or microphone, the app will advantageously show in the app on the screen of the device if the signal(s) received from the pump aggregate is adequate and sufficient in amplitude and quality in order to perform a measurement. The user will move the device closer to the pump assembly, or away from the pump assembly, or move the device to the sides, above, or below the pump assembly. While doing this, the device and the app will detect the signal, and once an optimum position is reached, the device may give an acoustic signal, vibrate or give a visual indication in the display of the device. In this way, the handheld communication device has given signal feedback to the user. This signal and device positioning procedure can be made for one single parameter, e.g. for measuring cavitation, or it can be made for several parameters at the same time. In the latter case, the app of the device may find an optimum position for measuring a plurality of parameters at the same time. As the frequency ranges of some of the fault parameters differ from each other, the handheld communication device measures for a first duration of time, e.g. the flow noise in the area of 1 Hz to 25 kHz, and then switches or a second duration of time to e.g. measuring noise generated by the electronics in the range 50 kHz to 200 kHz. Thus, the handheld device, or more precisely the app of the handheld communication device sequentially executes scans of one or more of the noise sources 120 shown in FIG. 4. The order can be set by the app or be random.

[0164] The method can also detect dry running of the pump, i.e. the case where the rotor and impeller rotates, but where no liquid is in the pipe. This situation is detrimental to the bearings. By looking at the difference in the sound signal measured when the pump is running with liquid, and when running without liquid, a statement as to “Dry run: Yes” or “Dry run: No” can be made and shown on the display of the handheld device as shown in FIG. 5. The system may also as later explained learn to recognise a dry run state, e.g. via a neural network or a database linking a dry run state with associated sound signal patterns or spectrums and possibly the pump type or model.

[0165] The method can likewise also detect unbalance in the pump assembly. The unbalance may for instance be caused from the rotor or the impeller blades. The unbalance may also occur, if the impeller blades are damaged, e.g. from erosion or cavitation. The system may also as later explained learn to recognise a unbalance state, e.g. via a neural network or a database linking a unbalance state with associated sound signal patterns or spectrums and possibly the pump type or model.

[0166] The method can further detect water hammering (or hydraulic shock), which is the momentary increase in pressure inside a pipe caused by a sudden change of direction or velocity of the liquid in the pipe. Water hammer can be particularly dangerous because the increase in pressure can be severe enough to rupture a pipe or cause damage to pump equipment. From a pump warranty perspective, it would be desirable if the method or system would be cable of detecting and logging every water hammer incident. Any phenomenon that manipulates mechanical energy cannot switch abruptly from one energy state to the next. This means that a sound cannot switch suddenly from silence to its maximum amplitude. A finite time, however brief, is needed, during which the sound can evolve to its new state, This transitional time is called the attack transient. By the same terminology, there is a release transient at the time during which the sound return back to silence. In general, the evolution of the amplitude of a sound can be divided into four basic parts—Attack, Delay, Sustain and Release. The evolution of the amplitude of a sound represented as an idealized line that links the positive peaks of its waveform is called the envelope of the wave. Using this terminology, the water hammer sound have an envelope curve like a piano with a frequency content below 1 kHz—the valve sound has en envelope like a trumpet with a frequency content from 1 kHz to 15 kHz. A signature for water hammer is a low frequency fast attack and long release sound. An algorithm for water hammer detection must perform both an frequency analysis and measure the envelope shape of the sound. The envelope curve can be found from the vibration or sound signal by bandpass filter the signal, absolute square the result followed by a lowpass filter.

[0167] Finally, the method can detect cavitation, which is the formation of bubbles or cavities in liquid, developed in areas of relatively low pressure around an impeller in a pump. The imploding or collapsing of these bubbles trigger intense shockwaves inside the pump, causing significant damage to the impeller and/or the pump housing. The sound of cavitation is per nature a very high frequency sound, and is not overlapping with the sound of the mechanical components of the motor and pump. A robust signature of cavitation can be obtained by measuring the energy of the spectrum from 10 kHz to 20 kHz. This can be performed in the time domain using a bandpass filter, and sum up the filtered signal, or in the frequency domain by performing a frequency transformation and sum up the component in the spectrum from 10 kHz to 20 Khz. The energy level will often be compared to a baseline, and when the baseline is exceeded by some levels, an cavitation alarm can be issued.

[0168] In a fourth step 840, the measured sound signals are processed according to the previous routine, i.e. run through an algorithm to estimate the rotational speed of the shaft and further analysed in accordance with one or more of the previously described analysis methods. The analysis may be carried out on the handheld communication device, on an external server or a combination thereof.

[0169] In a fifth step 850, the processed signals are compared to stored sound emanating conditions in order to identify the operational condition of the pump assembly and to identify any possible faults. The results of the analysis are displayed on the GUI to the service worker in a sixth step 860.

[0170] Based on the possible identified fault, the pump may be instructed in step 870 to not drive the pump assembly in rotational speed regions, where the pump assembly is faulty. This may prolong the time before parts need to be replaced and the lifetime of the pump assembly. The pump assembly may be instructed directly via the handheld communication device or via the external server.

[0171] Alternatively, the software app may instruct the service worker to replace the pump assembly or a part of the pump assembly in step 880. The software app may provide a guide on the GUI, which instructs the service worker on how to replace the identified damaged part.

[0172] The sound signals measured in step 830 and/or the processed signals 840 may be uploaded to an external server and stored in a library or server. Thereby, it is possible to provide a library of known sound measurements or processed spectrums, whereby the system may better learn to identify different types of faults and optionally dependent on the particular pump assembly type or model. The learning process can for instance be carried out via the use of a neural network. The sound measurements may for instance be linked to a certain types of identified faults, which may also encompass fault types, which have not previously been encountered. This can also facilitate the learning process. Thereby, the fault detection system will continuously be better at identifying fault conditions.

[0173] According to the invention, the fault detection system will sequentially carry out measurements or individual scans to identify individual fault states of the pump assembly, e.g. from the aforementioned fault states. This is illustrated in FIG. 14. In a first step 910, the system probes for a first fault state and in a following second step 920, the system probes for a second fault state. Thereby, the system sequentially carries out dedicated measurements and analyses to detect specific fault states for the pump assembly. This provides a more accurate measurement system, where individual fault states more accurately can be detected. This can for instance be carried out by first carrying out a first dedicated measurement to detect if the pump assembly has a particular fault state and then carrying out a second measurement to detect if the pump assembly as another particular fault state. It is also possible to sequentially scan for a plurality of different noise sources, where the individual noise sources are located in different frequency bands or ranges. It is also possible to sequentially run measurements through different signal processing algorithms in order to identify individual fault states.

[0174] The system advantageously at least probes for at least water hammering faults and cavitation faults in two consecutive measurements or scans. The associated noise from these faults are as previously described located in different frequency bands and may also require different signal processing steps in order to detect the faults. In an advantageous embodiment, the system probes for faults in a first predetermined frequency range, which is located in a first band near the kHz-range, and in a second predetermined frequency range, which is located in a second band in the kHz-range. These ranges are particularly suited for detection of water hammering and cavitation faults. Each of the probe steps may be run through separate dedicated signal processing algorithms associated with the particular fault state.

[0175] The fault detection system may as illustrated in FIG. 14 also carry out a third step 930, wherein the system probes for a third fault state. The system may of course also carry out additional steps for probing for additional fault steps.

[0176] Each of the probing steps 910-930 may comprise a number of the steps 810-880 from the individual fault detection methods as shown in FIG. 13, in particular steps 830-860.

[0177] The invention has been described with reference to advantageous embodiments. However, the scope of the invention is not limited to the illustrated embodiments, and alterations and modifications can be carried out without deviating from the scope of the invention, which is defined by the following claims.

TABLE-US-00001 Reference Numerals 10, 110 Fault detection system 15, 16  Pipes 20, 20′, 110 Pump assembly  21 Common housing  .sup. 22′ Pump housing 123 Motor and rotor  .sup. 24′ Electrical motor housing 124 Water hammer  .sup. 25′ Terminal box 125 Impeller 126 Bearings 127 Water in pipe noise 128 Cavitation 129 Electronic 30, 130 Handheld communication device/smart phone 131 Signal processor/micro computer 32, 132 Microphone 133 Display 134 Input module 135 Communication module 136 Library 137 Additional sensor(s)/Accelerometer  40 Server/cloud 210-290 Steps in algorithm for detecting shaft speed 310-340 Parts of condition monitoring system 426 Bearing 440 Ball of bearing 450 Crack/fault 455 Inner raceway 460 Outer raceway 465 Retainer 470 Sound signal 480 Periodic structure 510-550 Steps in envelope analysis 610-640 Lines referring to fundamental frequency and other impact rates 700-730 Steps in normalisation and Cepstral analysis 810-880 Steps in fault detection method according to the invention 910-930 Steps in sequentially probing for different fault states