BR112019017435B1 - OPTICAL ACOUSTIC DETECTION SYSTEM AND METHOD - Google Patents

Abstract

A method of measuring the state or condition of a plurality of spatially spaced machine parts subject to wear and emitting an acoustic characteristic, the method including the steps of: (a) optically detecting the acoustic properties of the plurality of machine parts subject to wear, and generating detected signals therefrom; (b) dividing the detected signals into a first series of corresponding spatial segments along spaced machine parts and, for each spatial segment, dividing the detected signal into a temporal segment recording the acoustic properties for the spatial segment over an extended period of time; (c) dividing each temporal segment into a series of frequency domain subsegments, transforming the subsegments into corresponding frequency domain subsegments; (d) combining the frequency domain subsegments within a spatial segment to produce a corresponding combined frequency domain subsegment of lower noise level; and (e) determining the fundamental frequency of the emitted acoustic signatures present in the combined frequency domain subsegment and associated harmonic.

Description

FIELD OF INVENTION

[0001] The present invention provides systems and methods for the distributed detection of mechanical equipment and, in particular, for the distributed optical detection of rollers and bearings or moving parts and mechanisms, including those contained in conveyor belts.

BACKGROUND OF THE INVENTION

[0002] Any discussion of the background of the invention through the descriptive report should in no way be considered as an admission that such technique is widely known, or forms part of the general knowledge in the field.

[0003] Remote monitoring of equipment such as rollers and bearings on conveyor belts is a significant problem in the mining and other industries. Conveyor belt maintenance can be expensive. For example, a 1km length of conveyor belt could contain 6,700 or more bearings, the failure of any one of which could lead to catastrophic consequences such as damage or fire to the conveyor belt. Rollers and bearings can fail in many ways, including lining failure, pizza cutter failure and bearing failure.

[0004] Previous methods of monitoring the condition of rollers have included visual and auditory inspection by physically walking along the conveyor belt and manually listening to the bearings, or for example, thermal imaging.

[0005] PCT publication WO 2005/042661 and PCT publication WO 2015/115591, the contents of which are incorporated herein by cross-reference, propose a form of conveyor belt idler bearing monitoring system that involves having wi-fi enabled monitors inside each bearing roller that monitor the condition of the bearing. Such a solution is likely to be unduly expensive and is prone to failure due to the large number of bearings and a large system requiring complex interactions.

[0006] The present invention is applied to the monitoring of other equipment containing bearings and other moving parts, such as motors, pumps, gearboxes, turbines, transmissions, crushing and shearing machinery, and bulk material handling equipment. The present invention is also applied to the monitoring of other equipment in a distribution mode using fiber optic acoustics.

SUMMARY OF THE INVENTION

[0007] It is an object of the invention, in its preferred form, to provide an improved form of monitoring equipment, including machine parts subject to wear.

[0008] According to a first aspect of the present invention, there is provided a method for measuring the state or condition of a plurality of spatially spaced apart machine parts subject to wear and which emit an acoustic signature, the method including the steps of: (a) optically detecting the acoustic properties of the plurality of machine parts along their length, and deriving detected signals therefrom; (b) dividing the detected signals into a first series of corresponding spatial segments along spaced apart bearings, and for each spatial segment, dividing the detected signal into a temporal segment recording the acoustic properties for the spatial segment over an extended period of time; (c) dividing each temporal segment into a series of frequency domain subsegments and transforming the subsegments into corresponding frequency domain subsegments; (d) combining the frequency domain subsegments within a spatial segment to produce a corresponding lower noise level combined frequency domain subsegment; and (e) determine the fundamental frequency of the emitted acoustic signatures, present in the combined frequency domain subsegment, and associated harmonics.

[0009] In some embodiments, the machine parts may include bearings, and may form part of a conveyor belt. Preferably, optical detection occurs substantially simultaneously for the acoustic properties of the plurality of machine parts.

[0010] Step (d) may preferably include subtracting a noise floor measure from the combined frequency domain subsegment. The noise floor measure may preferably include a curve interpolated through substantially the lowest background noise levels of the combined frequency domain subsegment. The interpolated curve may preferably include one of a polynomial, exponential, logarithmic or piecewise curve.

[0011] The detection step (a) may comprise optically detecting the acoustic properties of the bearings. Optical detection may preferably include using diffusion along an optical fiber to detect the acoustic properties.

[0012] Step (a) may preferably further include filtering the detected signals to take into account the frequency-based attenuation characteristics of the detection system.

[0013] Step (e) may preferably include the step of convolving a Gaussian spike-shaped structure, or other suitable spike-shaped structure with the combined frequency domain subsegment.

[0014] According to another aspect of the present invention, there is provided a system for measuring the state or condition of a plurality of spatially spaced bearings, the system including: a first sensor unit for optically detecting acoustic properties surrounding an extended optical waveguide disposed proximate to the spaced bearings; processing means for processing the detected signals into a series of corresponding spatial segments; with each spatial segment having an extended temporal segment, recording the acoustic properties of the spatial segment for an extended period of time; the processing means further performing frequency domain processing of the extended temporal segment to produce a corresponding lower noise level frequency domain segment, and the processing means further reviewing the frequency domain segment to determine the existence of any fundamental frequency and associated harmonics of any bearing emitting an acoustic signature.

[0015] According to another aspect of the present invention, there is provided a method for measuring the state or condition of a conveyor belt system, the conveyor belt system including a plurality of roller units, also known as idler gears, each having a series of bearings on the rollers, the method including the steps of: (a) optically detecting acoustic properties of the conveyor belt system along its length, and deriving detected signals therefrom; (b) dividing the detected signals into a first series of spatial segments along the conveyor belt and for each spatial segment, dividing the signal into a temporal segment that records the acoustic properties for the spatial segment over an extended period of time; (c) optionally filtering the temporal segment to account for the frequency-based attenuation characteristics of the optical detection system; (d) dividing each temporal segment into a series of frequency domain subsegments, transforming the subsegments into corresponding frequency domain subsegments; (e) combine the frequency domain subsegments within a spatial segment to produce a corresponding combined frequency domain subsegment of lower noise level; (f) determine the fundamental rotational frequency of any rolls present in the combined frequency domain subsegment, and associated harmonics; and (g) determine frequency patterns indicative of different types of fault.

[0016] In some embodiments, the method may further include the steps of: (g) subtracting a model of the fundamental rotational frequency and harmonics from the combined frequency domain subsegment to obtain a modified frequency domain subsegment; and (h) analyzing the modified frequency domain subsegment to determine bearing wear conditions.

[0017] Step (f) may preferably include cross-correlating the combined frequency domain subsegment with a Gaussian or any other similar response function. Step (e) may preferably include averaging or summing the frequency domain subsegments. Step (e) may preferably include regressing the background noise within the frequency domain subsegment using a polynomial, exponential, logarithmic or other linear or piecewise curve function, plus an algebraic addition or subtraction.

[0018] Bearing wear conditions may include at least one of: the fundamental roller rotation frequency and harmonic frequencies; bearing spalling frequencies, if any; bearing half-life wear patterns; or bearing late-stage wear patterns; bearing haystack patterns.

[0019] Time segments are preferably about 5 to 30 minutes in duration, depending on the carrier, but may be shorter or longer. Subsegments are preferably about 3 seconds in duration, but vary depending on the carrier. Flaking frequencies are preferably below 100Hz.

BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Embodiments of the invention will now be described by way of example only, with reference to the accompanying drawings, in which:

[0021] Fig. 1 schematically illustrates the optical detection system of one embodiment;

[0022] Fig. 2 illustrates a photograph of an exemplary connection of a fiber optic sensing cable to a conveyor belt system;

[0023] Fig. 3 illustrates an example of a ‘waterfall’ chart as received by an interrogator unit;

[0024] Fig. 4 illustrates a form of high-pass filter;

[0025] Fig. 5 illustrates the bearing frequency calculation equations;

[0026] Fig. 6 illustrates a graph of operating frequencies on an exemplary conveyor belt system; and

[0027] Fig. 7 illustrates a graph of different bearing operating characteristics;

DETAILED DESCRIPTION

[0028] Embodiments provide a system and method that utilize an acousto-optic monitoring system with complex backend processing to accurately monitor the condition of equipment along a conveyor belt.

[0029] Referring initially to Fig. 1, the operating environment of one embodiment is shown schematically. In this arrangement 1, a series of rollers, e.g. 3-8, are monitored by a single fiber optic cable 11 that is disposed along the rollers. The fiber optic cable 11 includes an acoustic sensing environment and is connected to a series of advanced processing units 12 that process the acoustic signal to determine the status or health of each roller. In practical systems, many thousands of sets of rollers can be effectively monitored simultaneously.

[0030] Through meaningful processing of the received optical signal, the condition of the rollers and their bearings can be determined, in addition to the condition of the conveyor belt.

[0031] Examples of optical monitoring systems 12 include the Future Fibre Technologies FFT Secure Pipe product and the Optasense Pipeline monitoring system. Also included are fiber acousto-optical interrogators from other manufacturers, such as Fotech and Hawk.

[0032] With such systems, a single fiber optic cable may be laid out along the conveyor belt and secured near each roller. Fig. 2 illustrates a photograph of such an arrangement 20, in which a cable 21 has been secured to a conveyor belt infrastructure 22 using cable ties 23, 24, or the like. Alternative securing arrangements include screw or spring clamps.

[0033] The interrogator unit collects responses from light pulses from glass fiber. Glass fiber contains microscopic imperfections that scatter light. The imperfections cause “Rayleigh Scattering,” the same effect that makes the sky blue. Vibrations and temperature variations in the fiber cause changes in the refractive index of the light, inducing readable signals in the reflected light. High-intensity, short-pulse lasers are used, coupled to sensitive reflection detectors and fast analog-to-digital converters.

[0034] The fiber is connected to an interrogator unit 12 of one of the proprietary companies mentioned above, and outputs a digital signal equivalent to the received acoustic signal for storage. The acoustic signal is output in a data format as follows. For the emission time period, an acoustic signal for each channel or unit distance along the carrier is output for each received audio frequency.

[0035] One form of representation of each time unit of data is known as a waterfall chart, an example of which is illustrated 30 in Fig. 3. The interrogator unit outputs data equivalent to waterfall charts at a predetermined rate. For example, 20,000 times per second (20 kHz).

[0036] The captured data needs to undergo processing before it can be used to effectively generate information on the state of the conveyor belt system.

[0037] In one embodiment, the emission data has been processed with the following sequence of operations to generate meaningful details about the conveyor belt operations. The following list is a series of operations in the order in which they are programmed:

[0038] 1. Selection of the length of time over which raw data is collected.

[0039] 2. Converting the raw data to the standard HDF5 data format or other data format suitable for processing

[0040] 3. Separating data in each channel to separate time series

[0041] 4. Establishing the high pass filter or other type of time constant for the filter.

[0042] 5. Applying a high-pass filter or other type of filter to the time series data sets.

[0043] 6. Establishment of Fourier Transform parameters to determine frequency range and resolution.

[0044] 7. Broadcast the input data set into monitorable file sizes that are about 3 seconds in length, so as not to exceed a monitorable number of samples, typically, but not restricted to, 65,536 samples.

[0045] 8. Application of a Fast Fourier Transform operation to extract frequency responses.

[0046] 9. Average the filter to combine all short files within 5 minutes or 10 minutes or longer as required.

[0047] 10. For each long-averaged frequency plot regression of the lower background noise envelope to a multiple order polynomial equation or other function, such as exponential, logarithmic, linear, piecewise linear, or piecewise curved or other suitable type of function.

[0048] 11. Subtract the regression equation curve from the frequency plot to normalize it.

[0049] 12. Plot the frequency curve for each channel for visual inspection, if required.

[0050] 13. Supply the belt speed and roller diameter to calculate the fundamental roller frequency.

[0051] 14. Select the fixed parameters of a Gaussian curve equation or some other standard matching curve.

[0052] 15. Generate the standard matching curve equation with adjustable center frequency.

[0053] 16. Select the maximum frequency and frequency range to be used in cross-correlation

[0054] 17. Cross-correlate the standard frequency domain matching curve with the frequency spectral plot for each channel. Then generate a very high resolution averaged cross-correlation plot for each channel.

[0055] 18. Use the high-resolution chart to precisely identify the fundamental frequency of the roll and all of its harmonics up to about 200 Hz, or higher if required. Measure the center frequency and pitch of each corresponding pattern.

[0056] 19. Use the standard matching curve equation and the known central harmonic frequency and amplitude to recreate a rough mathematical model graph of just the fundamental and its harmonics.

[0057] 20. Use this graphical model to generate a simple quality metric for the fundamental harmonic condition of the belt.

[0058] 21. Subtract the harmonic mathematical coarse model plot from the channel normalized frequency spectral plot.

[0059] 22. Use the frequency subtracted graph up to 100 Hz to generate a simple quality metric of roller bearing spalling frequencies to assess early stage bearing wear conditions.

[0060] 23. Record the fundamental harmonic conditions and bearing chipping wear frequency condition on a log sheet, one line for each channel. Initially, the frequency chart is examined over the range of 100 Hz to 300 Hz to identify bearing conditions in terms of mid-life failure. Next, a review of frequencies above 300 Hz is determined to identify failure at a later stage and at the end of life. Additional steps are to match (1) knock frequencies below the fundamental, (2) rattle frequencies above 100 Hz, (3) squeak frequencies above 150 Hz, and haystack charts at any frequency above 100 Hz. The results are saved as a csv file for future processing.

[0061] 24. Combine channel reports from spreadsheets with rows and numbers or idler gear frame numbers for conveyor belt and publish as a new spreadsheet.

Detailed Description of Data Processing Steps

[0062] The data processing steps listed above will now be described in greater detail.

[0063] 1. Selection of the time duration over which raw data is collected

[0064] Other vibration data are typically collected from fiber optic cables for short periods of time only, on the order of milliseconds or seconds. There are several reasons for this, primarily arising from the normal use of fiber optic vibration sensing as used in seismic work, intrusion detection, and the like. These other uses of fiber optic vibration sensing typically look for occurrences of seismic vibration.

[0065] In the embodiment, a long-term acoustic vibrational signal rather than a short-term vibrational signal is collected. The embodiments seek to operate at more or less continuous acoustic frequencies, not instances of vibration. Unfortunately, the marker frequencies of interest also tend to be buried in significant noise. To extract the marker frequencies from the noise and improve the signal-to-noise ratio for useful frequency identification, a data collection time of several minutes is best utilized.

[0066] Under ideal conditions, a collection time of 5 minutes has been found to be adequate. Under less than ideal conditions, where high noise levels are present, the collection time may be extended to 10 minutes or more. If the sampling frequency is 20 kHz, the number of samples collected is about a minimum of 6,000,000 per 5 minutes. In the embodiment, a range of data collection times and sampling frequencies are used, depending on the application, such as the length of the carrier and other factors.

[0067] 2. Converting raw data to a suitable data format for further processing

[0068] The FFT and Optasense interrogator products store their data in proprietary formats. As a first step in aiding further processing of the raw data collected by these instruments is to convert this data to an open standard format. It has been found that each brand of interrogator unit has its own data format, and it is necessary to transfer this data to a standard data format. It has been found that one, but not the only, suitable format is the HDF5 data file format.

[0069] 3. Separating the data in each channel to separate the time series.

[0070] The HDF5 or other data format file can then be partitioned into a separate file for each channel, also known as a “distance bin”. A channel is a discrete length of fiber that effectively acts as an individual microphone. The channel length can be set at the interrogator and is usually set at the most sensitive discrimination - currently around 1 meter or less. Partitioning is a first step so that preliminary filtering can be done on each individual channel. Depending on how the system is configured, this operation can occur at any time before a subsequent Fourier Transform.

[0071] Using an appropriate scripting code, such as a Matlab Prototype Scripting Code, the data can be formatted appropriately from the proprietary format. Future software can be in any other programming language.

[0072] 4. Setting the preprocessing filter type and its time constant and drift removal

[0073] The raw time series data files, as stored in HDF5, or other suitable data formats are first filtered to modify the frequency response. The frequency response is modified differently depending on the type of data collected and the specific interrogator used. Any type of filter can be used, including but not limited to high-pass, low-pass, band-pass, band-reject, integration, differentiation, moving window, and ARMA (autoregressive moving average) filters. In practice, it has been found that each proprietary data type and each interrogator tends to have a frequency response in its own way, as follows:

[0074] OptaSense Data Type 1: Single Pulse Technology: Data from this source has been found to have a high DC component (at zero frequency) and a strong bias towards lower frequencies with a very high decay as the frequency rises above 50 to 100 Hz. As a result, a high-pass filter is required to suppress the low frequencies and amplify the frequencies above about 100 Hz. In general, the time constant of the filter is set to a value of about 2000.

[0075] OptaSense 6 data types: Dual pulse technology: Data from this source was found to have a much flatter frequency response than data type 1 and was relatively consistent up to several kilohertz. The theoretical upper limit is about 10 kHz, but frequencies above 1 kHz are rarely seen, with occasional frequency spikes in the 2 to 3 kHz region. The raw data was found to have a very heavy downward drift towards the negative. It was stored as a 32-bit integer, and then when the value reached -32767 it immediately jumped to +32768, a process referred to as “integer roll-over”. Integer roll-over and drift posed significant problems. The solution was to use a differencing filter on the raw time series signal. The spikes resulting from integer roll-over could then simply be deleted from the

[0076] Future Fiber Technology Data (Type F): Single Pulse Technology: Data from this source has been found to be similar to Type 1 data, except that the frequency response required. Consequently, it has been found that lower frequencies, and a time constant has been chosen is established.

[0077] 5. Applying a frequency modification filter to time series data sets.

[0078] While there are many filters that can be used, an exemplary filter was a first-order high-pass filter. A typically simple filter used experimentally was a first-order high-pass filter that suppresses low frequencies and amplifies frequencies above the “knee” frequency response. The transfer function of such a filter is analogous to:

[0079] where the RC constant is formed from the resistance in ohms and the capacitance in farads if the filter is an analog filter constructed from an operational amplifier, a resistor and a capacitor. If the transfer function is changed to the following equation below, it can be modeled using the "digitally simulated analog method" © Paul Wilson. The digitally simulated analog method consists of a conversion of a transfer function to an analog computer program. The analog computer program is then digitally simulated in a computer program, element by element. The technique was first described by Paul Wilson in his master's thesis in 1986, and he holds the copyright. The method can be used for any type of analog filter or any other analog process.

[0080] The filter time constant = RC: the larger the value, the slower the filter. The filter can be modeled as an analog form, as illustrated in Fig. 4. The analog-to-digital computer program conversion of this program (written in pseudocode) can be as follows:

[0081] Filtering of Type 6 data and other interrogators: It was found that filtering for Type 6 data was of a different form and required a different pre-processing filter. Each interrogator manufactured by different vendors modifies the frequency dynamics in a different way, which requires a different type of pre-processing filter customized for the interrogator.

[0082] The end result of the pre-transformation portion of the process is a modified frequency response graph that becomes apparent after it has been transformed. If the input data time series is differenced, it does not affect the frequency response except to remove the extremely low frequency component - the DC component. Differentiating filtering of an incoming analog signal only shifts the phase of each frequency by +90 degrees, but does not change the frequencies. Since an input signal is a digital time series rather than an analog one, the cutoff frequency is the alternate time at which the data is sampled. In one example, a 3 second sample was taken at 20 kHz, and then the differentiator removed the DC component below 1/3 Hz or 0.333 Hz.

[0083] This ideal was used for data type 6, as it removes the problematic drift but still retains the value frequencies that have been found to be typically 5 Hz and upstream. Therefore, a direct differentiation of the digital input data stream replaces the high-pass filter. Depending on how the interrogator captures the data, either type of filter can be used.

[0084] 7. Establishment of Fourier Transform parameters.

[0085] To determine the frequency range and frequency resolution required for further analysis it is necessary to select the correct parameters for a Fourier Transform operation. The equations governing the selection of the parameters are known. The key equations are as follows:

[0086] where T = time at which samples are collected; N = number of samples in a collection; Δt = the sampling interval (time between adjacent samples); Fmax = is the maximum observable frequency; and Δf = Frequency resolution (the smallest frequency that can be discriminated).

[0087] Thus, it is possible to decide the sampling interval and the time at which sampling occurs in order to determine the other observable factors. The settings depend on choices affecting the particular carrier, the application, the carrier owner's requirements, the length of the carrier, and other factors.

[0088] The Fourier transform is a powerful tool, but it can also be subject to data artifacts, both mirroring and edge effects, so care must be taken to eliminate such artifacts and extract reliable results. To this end, the data was processed into slices of 65536 time series samples. Then, to remove edge effects such as envelope effects, this was reduced to 32768 samples after processing. At a sampling frequency of 20,000 Hz, the resulting time slices were about 3 seconds long.

[0089] The sampling frequency of 20,000 can be adjusted at the interrogator, mainly to cover a longer distance along the fiber. It is therefore a maximum observable frequency shift to prevent covering more than one length of the carrier in a single measurement. In practice, the sampling frequency is adjusted by the interrogator settings, but the value is decided according to the application, the carrier and the owner's requirements.

[0090] 7. Split the saved data file (HDF5 or others) into monitorable file sizes of about 3 seconds.

[0091] To meet the demands of the Fourier transform equations above, the sample for each 5 minute or 10 minute data channel is now divided into shorter (time) slices. At a sampling frequency of 20 kHz, this becomes about 100 time slices of 3 seconds each. This gives 65536 time series samples in each time slice. Each of these can now be processed to generate a Fourier Transform. At other sampling frequencies, the number of samples required per processing defines the length of each time slice and the number of time slices.

[0092] 8. Application of a Fast Fourier Transform operation

[0093] Each 3-second time slice of data is processed using a modern Fast Fourier Transform algorithm (experimentally Matlab was used, but one of several math library functions such as GSL could be used). Other types of discrete or fast Fourier Transform algorithms could be used and are included here as an option. Thus, for each 3-second time slice and for a fiber channel (about 1 meter long) 100 to 200 Fourier Transform plots are obtained in the form of a data set or list of values. Then the top half of the data set is removed to remove the mirror-wrapping effect, reducing the data set to 32768 samples.

[0094] 9. Averaging the filter to improve the signal to noise ratio.

[0095] In most cases, a few seconds of data were found to be insufficient. The signal-to-noise ratio was found to be generally so poor that it was very difficult to obtain any meaningful analysis from a frequency plot covering less than a few minutes of data.

[0096] The Fourier transform has already done a partial job of extracting useful spectral signatures, but they tend to get buried in noise. To take advantage of the extended time period (5 minutes or more) of data collection, hundreds of frequency plots for each channel were averaged and summed, and the necessary frequency peaks were amplified and noise suppressed during the averaging. This also removes occasional spurious noise, such as passing vehicles or something impacting the carrier structure.

[0097] 10. Noise floor removal

[0098] It was found that the frequency plot obtained after the average filter operation still presented artifacts that interfered with the analysis of the results, especially any form of automated analysis. It was found that the noise floor level varies considerably from channel to channel, depending on spurious environmental noise and acoustic resonances in and around the structure of the equipment. The design of the conveyor structure and the external environment can have a considerable impact on the noise level if, for example, the conveyor crosses a narrow bridge.

[0099] It has been found that the noise floor increases greatly at the lower end of the spectrum due to contamination by low hum caused by metallic resonances and other factors, and sometimes a DC component, although the pre-processing filter can remove much of this noise. The rise at the lower end of the graph should be removed. However, where there is a sharp rise in noise floor in the higher frequency range (usually above 200 Hz) this should be retained as it is an important marker of bearing wear at a later stage. The shape of the noise floor does not necessarily follow a mathematical analytical curve described by a single equation, and is different for each frequency graph.

[0100] The process of removing background noise can proceed as follows. Identify the envelope that runs along the lower side of the graph that joins all the lower peaks. To this envelope, a curve is fitted. Several different envelope curves have been used, and recently a 5th order polynomial equation, an exponential or logarithmic curve, a piecewise linear curve, and a piecewise conglomeration of several curves have been used. The library incorporates a variety of envelope equations, including all of the above and others. The envelope curve is currently found using standard mathematical packages or mathematical software libraries (Matlab, Octave, Scientific Library, or GSL). It has been found that 5th order polynomial fitting and exponential fitting are best for mapping background noise in certain types of data sets without fitting (and therefore removing) very close to the required, important markers of late-stage wear. The selection of a suitable envelope equation library will depend in the future on the application parameters and requirements. The envelope curve is then subtracted from the 5 minute (or 10 minute or more) frequency plot for each channel. Normalization of each channel is an essential step in automating the pattern analysis process.

[0101] 12. Storing normalized frequency graphs

[0102] At this point in the procedure, all frequency plots, one for each channel (or distance range) are stored for later analysis and manual inspection if required.

[0103] Preparation for automated analysis and pattern recognition.

[0104] The very volume of processed frequency graphs is such as to make manual interpretation difficult. Fully automated graph analysis is therefore important. The following steps are part of the ongoing development of fully automated pattern recognition and fault diagnosis.

[0105] There are six main patterns to be solved from the frequency graphs. They are:

[0106] 1. The fundamental frequency of rotation of the rollers and its harmonic frequencies, 2nd, 3rd, 4th, 5th and higher harmonics. If the fundamental frequency is 10 Hz (typical), then harmonic frequencies occur at 20, 30, 40, 50, 60, 70 and sometimes higher than 150 Hz.

[0107] 2. Spalling frequencies representing early stage wear on bearing surfaces such as ball surfaces and raceway surfaces. Also included are cracks in the ball carrier. These frequencies can be calculated and typically occur below 100 Hz.

[0108] 3. Half-life wear patterns, represented by what is called a “haystack” graph since it resembles a disorganized medieval haystack. The haystack is a series of rattling frequencies caused by looseness in the bearing races as they wear and begin to rattle. These frequencies are typically seen from about 50 Hz up to about 200 to 250 Hz.

[0109] 4. Later stage wear patterns are seen as haystack plots at higher frequencies. They usually occur above 250 Hz, frequently occur in a range of 400 Hz to 800 Hz, and can sometimes be seen in a range as high as 4000 Hz. These haystack plots are typical of bearings that are sized and require immediate replacement. There is often more than one haystack plot across the frequency range.

[0110] 5. Knock wear patterns. Knock is a large frequency peak below the fundamental frequency, indicating a bearing that is too loose or is failing.

[0111] 6. Squeaking: Squeaking is a single frequency peak in the range above 150 Hz, indicating a stuck ball or ball carrier.

[0112] Spalling Frequency Calculations: Frequency indicators for spalling frequency are known and also haystack graphs are widely discussed in the literature and trade warning notes. The present embodiment allows them to be extracted from fiber optic cable sampling technology. The frequencies and equations are provided in Fig. 5.

[0113] Fig. 6 illustrates chipping frequencies. Fig. 7 illustrates a general range of marker frequencies.

[0114] Known frequency inputs: To trigger calculations relating to the fundamental frequency and its harmonics, it is necessary to establish the fundamental frequency. This can be calculated from the belt speed and the roller diameter.

[0115] 14. Select the fixed parameters of a Gaussian curve equation or other envelope equations.

[0116] There are many peak frequency envelope equations that can be used, including a triangular Gaussian curve, a rectangular one, and any of the waveforms. The shape of the fundamental frequency and its peak harmonic frequency closely resembles a Gaussian curve, and this shape has been shown to be effective for the purpose. The embodiment can make use of other envelope shapes. A Gaussian envelope can be generated from an exponential equation of the usual form:

[0117] The values of a, b, and c are constant parameters. The value of “a” determines the height, the value of “b” determines the location of the center point, and “c” determines how steep the slope is. The values of “b” and “c” are more or less constant for all harmonic peaks and can be set to fixed values. The value of “a” varies, depending on the use. For the cross-correlation procedure below, the value of “a” is a nominal value (say, 1). When the overall model is created, the value of “a” is the height of the peak being modeled.

[0118] 15. Establish the peak frequency envelope curve equation for cross-correlation.

[0119] A cross-correlation procedure is used in the frequency domain to correlate the frequency spectrum with the peak frequency envelope. Cross-correlation is typically used for time-series data in the time domain, so this is an unusual use. The goal is to find locations where the frequency plot corresponds to Gaussian-shaped peaks and thus identify the fundamental frequency and its harmonics. The important aspect of the cross-correlation process here is that it produces a plot that appears smoother filtered, with considerably higher frequency accuracy than the original frequency plot. The resolution is somewhat arbitrarily set at 100 times greater than the resolution of the original frequency plot derived from the Fourier Transform. The resolution magnification factor is a parameter to be selected in an operational embodiment.

[0120] The higher resolution plot allows for more accurate identification of the location of frequency peaks ready for modeling. So at this point the cross-correlation parameters, a, b, c and resolution are defined. Resolution is defined by the incremental movement of the frequency peak envelope curve as it passes across the frequency plot - how small a movement is made between each successive step.

[0121] Another decision that is made at this point is the maximum frequency of interest. This can be any value, but it has been proven that harmonics tend to die out around 150 Hz to 300 Hz, so there is little value in proceeding beyond that. It has been found that each carrier has its own range of harmonic frequencies, and the upper value needs to be established for the carrier when it is put into service.

[0122] 17. Cross-correlation of the peak frequency envelope curve with the frequency plot.

[0123] Cross-correlation is done by passing the peak frequency envelope curve across the frequency plot in one direction. Typically this would be from left to right, from low frequencies to high frequencies, but it could be in the opposite direction. Passing the curve from left to right could be in small increments. Because the resolution of the basic frequency plot is considerably higher than the correlation increments, at each incremental step the peak frequency envelope model needs to be recalculated at the coarse resolution and, if plotted, looks like a gradual peak.

[0124] The recalculation procedure is as follows: Let F(f) be the frequency plot data set. Let G(f) be the Gaussian curve plot at a coarse resolution at a specific fine resolution location. Let Y be the resulting correlated value at a single step. Let C(ff) be the resulting fine resolution cross-correlation plot. : Then c(ff) is the resulting graph of Y for 0 to 60,000

[0125] 18. Accurately identify the fundamental frequency and harmonics.

[0126] Now that there is a high resolution frequency graph with noise suppressed, the Gaussian peaks are amplified. The location of the fundamental has been calculated and then this location is determined and a precise measurement of the maximum point is taken, thus giving a very precise fundamental frequency. The harmonics will be integer multiples of this value and are then visited as a check. The maximum point of each harmonic peak is measured precisely and a list is made of all the harmonics and their amplitudes.

[0127] 19. Reshape the fundamental and harmonics in coarse resolution

[0128] Using the harmonic list and a precise location of each peak, the locations are used to determine the value of “a” in the Gaussian equation for each of the harmonics. The next step is to reconstruct a coarse-grained model of just the fundamental and its harmonics at the resolution of the original frequency plot. This plot is devoid of other signals, such as chip frequency plots and haystack plots.

[0129] 20. Using the fundamental and harmonic model to generate a report of only harmonic content in the data

[0130] The fundamental model and its harmonics (fine or coarse grains) are then used to create a somewhat subjective metric of roller condition. This metric gives an indication of the belt hitting the rollers, any roller deformation conditions, and any material adhering to the rollers. The metric is added to a belt condition report that could be as a column in a spreadsheet.

[0131] 21. Subtraction of the fundamental-harmonic model from the frequency graph.

[0132] The coarse-grained fundamental frequency and harmonics plot model is now subtracted from the original coarse-grained frequency plot. This removes the fundamental and harmonics, leaving behind the other spectral signatures of interest, specifically the chipping frequency patterns, the 100-200 Hz rattle and loose haystack frequencies, the knock frequencies below the fundamental, the squeak frequencies above 200 Hz, and the late-stage wear plots or haystack capture plots above 200 Hz.

[0133] 22. Generate a chipping failure metric from the new (subtracted) graph.

[0134] The new frequency subtracted graph, devoid of fundamental and harmonics, is used to examine and measure peaks below 100 Hz. A metric is generated and averaged to give an overall assessment of the wear condition of the bearing surfaces. This metric is added to the report for each channel.

[0135] 23. Generate a Haystack Failure Metric from the New (Subtracted) Graph

[0136] The new frequency subtracted graph, stripped of fundamental and harmonics, is used to examine and measure any haystack graphs above 100 Hz. A haystack or series of peaks between 100 Hz and 200 Hz usually indicates a squealing or grinding noise in the bearings and indicates mid-life failure. Haystack graphs above 200 Hz are indicative of late-stage failure or seized bearings. Metrics are generated and averaged to give an overall assessment of the bearing wear condition. The metrics are added to the report for each channel.

[0137] In summary, the embodiments provide a system and method for the automated detection of equipment and systems, such as conveyor belt systems using a simple fiber optic sensing arrangement. This has significant potential to provide an effective, low cost monitoring system, and can be applied for the simultaneous detection of machine wear in a variety of environments.

Interpretation

[0138] Reference throughout the specification to “an embodiment”, “some embodiments” means that a particular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment of the present invention. Thus, the occurrence of the phrases “in an embodiment”, “in some embodiments” in various places throughout the specification do not necessarily refer to the same embodiment, but may do so. Furthermore, the features, structures or characteristics may be combined in any suitable manner, as would be apparent to those skilled in the art from this disclosure, in one or more than one embodiment. During various trials using different interrogations on different carriers, all of these methods have been used in a range of different configurations or embodiments.

[0139] As used herein, unless otherwise specified, the use of ordinal numerals "first," "second," "third," etc., to describe a common object, merely indicates that reference is made to different examples of similar objects and is not intended to imply that the objects so described must be in a particular sequence, whether temporally, spatially, in a row, or otherwise.

[0140] In the claims below and in the present description, any of the terms comprising, comprised of or comprising is an open term meaning including at least the elements/features that follow, but not excluding others. Thus, the term comprising, when used in the claims, should not be construed as being limiting to the meaning or elements or steps listed below. For example, the scope of the expression a device comprising A and B should not be limited to devices consisting only of elements A and B. Any of the terms including or including, as used herein, is also an open term meaning also including at least the elements/features that follow the term, but not excluding others. Thus, including is synonymous with and means comprising.

[0141] As used herein, the term “exemplary” is used in the sense of providing examples, as opposed to indicating quality. That is, an “exemplary embodiment” is an embodiment provided as an example, as opposed to necessarily being an embodiment of exemplary quality.

[0142] It should be noted that in the above description of exemplary embodiments of the invention, various features of the invention are sometimes grouped into a single embodiment, FIG., or description thereof for the purpose of accompanying the disclosure and aiding in understanding one or more of its various inventive aspects. This method of disclosure, however, is not to be construed as reflecting an intent that the claimed invention requires more features than are expressly recited in each claim. Rather, as the following claims reflect, the inventive aspects reside in fewer than all of the features of a single disclosed embodiment. Accordingly, the claims that follow the Detailed Description are hereby expressly incorporated into the Detailed Description, with each claim constituting a separate embodiment of the present invention.

[0143] Furthermore, while some embodiments described herein include some but not other features included in other embodiments, combinations of the features of different embodiments are intended to be within the scope of the invention, and form different embodiments, as would be understood by those skilled in the art. For example, in the following claims, any of the claimed embodiments may be used in any combination.

[0144] Furthermore, some of the embodiments are described herein as a method or combination of elements of a method that may be implemented by a processor of a computer system or by other means of performing the function. Thus, a processor with the instructions necessary to perform such a method or element of a method forms a means for performing the method or element of a method. Furthermore, an element described herein of an apparatus embodiment is an example of a means of performing the function performed by the element for the purpose of performing the invention.

[0145] In the description provided herein, various specific details are set forth. However, it is understood that embodiments of the invention may be practiced without such specific details. In other examples, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of the present description.

[0146] Similarly, it will be appreciated that the term coupled, when used in the claims, is not to be construed as being limited to direct connections only. The terms "coupled" and "connected," together with their derivatives, may be used. It should be understood that these terms are not intended to be synonymous with each other. Thus, the scope of the expression a device A coupled to a device B is not to be limited to devices or systems in which an output of device A is directly connected to an input of device B. It means that there is a path between an output of A and an input of B which may be a path including other devices or means. "Coupled" may mean that two or more elements are either in direct physical or electrical contact, or that two or more elements are not in direct contact with each other but nevertheless cooperate or interact with each other.

[0147] Thus, while what are believed to be preferred embodiments of the invention have been described, those skilled in the art will recognize that other modifications may be made herein without departing from the spirit of the invention, and intend to claim such changes and modifications as being within the scope of the invention. For example, any formulas given above are merely representative of procedures that may be used. Functionality may be added or deleted from the block diagrams, and operations may be interchanged between functional blocks. Steps may be added or deleted to methods described within the scope of the present invention.

Claims (15)

1. A method for measuring the state or condition of a plurality of spatially spaced machine parts (3, 4, 5, 6, 7, 8) subject to wear and emitting an acoustic signature, the method comprising the steps of: (a) optically detecting along a waveguide the acoustic properties of the plurality of machine parts (3, 4, 5, 6, 7, 8) subject to wear, and deriving detected signals therefrom; (b) dividing the detected signals into a first series of corresponding spatial segments along spaced machine parts (3, 4, 5, 6, 7, 8) and, for each spatial segment, dividing the detected signal into a temporal segment recording the acoustic properties for the spatial segment over an extended period of time; (c) dividing each temporal segment into a series of frequency domain subsegments, transforming the subsegments into corresponding frequency domain subsegments; (d) combining the frequency domain subsegments within a spatial segment to produce a corresponding lower noise level combined frequency domain subsegment; and (e) determining the fundamental frequency of the emitted acoustic signatures present in the combined frequency domain subsegment, and associated harmonics. 2. Method according to claim 1, characterized in that such optical detection occurs substantially simultaneously for the acoustic properties of the plurality of machine parts (3, 4, 5, 6, 7, 8). 3. The method of claim 1, wherein said step (d) further includes subtracting a background noise measure from the combined frequency domain subsegment. 4. The method of claim 3, wherein said background noise measure includes a curve interpolated through substantially the lower background noise levels of the combined frequency domain subsegment. 5. The method of claim 4, wherein said interpolated curve includes one of a polynomial, exponential, logarithmic segment curve or piecewise segment curve. 6. The method of claim 1, wherein such optical detection includes using diffusion along an optical fiber to detect such acoustic properties. 7. The method of any one of claims 1 to 6, wherein said step (a) further includes filtering the detected signals to take into account the frequency-based attenuation characteristics of the detection system. 8. The method of any one of claims 1 to 7, wherein said step (e) includes the step of convolving a Gaussian-shaped structure with the combined frequency-domain subsegment. 9. A system for measuring the state or condition of a plurality of spatially spaced bearings (3, 4, 5, 6, 7, 8), the system comprising a first sensor unit for optically detecting along a waveguide the acoustic properties surrounding an extended optical waveguide disposed proximate to said spaced bearings (3, 4, 5, 6, 7, 8); processing means for processing such detected signals into a series of corresponding spatial segments; with each spatial segment having an extended temporal segment that records the acoustic properties of the spatial segment for an extended period of time; such processing means further performing frequency domain processing of such extended temporal segment to produce a corresponding lower noise level frequency domain segment; and such processing means further reviewing such frequency domain segment to determine the existence of any fundamental frequency and associated harmonics of any bearing (3, 4, 5, 6, 7, 8) that emits an acoustic signature. 10. The method of claim 1, wherein said machine parts form part of a conveyor belt, and said machine parts include bearings (3, 4, 5, 6, 7, 8). 11. The method of claim 10, further comprising the steps of: (f) subtracting a model of the fundamental rotational frequency and harmonics from the combined frequency domain subsegment to obtain a modified frequency domain subsegment; and (g) analyzing the modified frequency domain subsegment to determine bearing wear conditions. 12. The method of either claim 10 or claim 11, wherein said step (d) includes averaging or adding the frequency domain subsegments. 13. The method of claim 11, wherein the bearing wear conditions include at least one of: the fundamental roller rotation frequency and harmonic frequencies; bearing spalling frequencies, if any, bearing half-life wear patterns, or bearing late-stage wear patterns, bearing haystack patterns, and wherein said spalling frequencies are preferably below 100 Hz. 14. Method according to any one of claims 10 to 13, characterized in that such time segments are two to ten minutes in duration. 15. Method according to any one of claims 10 to 13, characterized in that such sub-segments are approximately 3 seconds in duration.