WO1990004230A1 - Signal processing method and apparatus for flowmeters - Google Patents

Abstract

A flagpole type of flow barrier is inserted in a pipeline. High pressure swirling vortices or eddies flow downstream from the flagpole barrier. The quantity of vortices per unit time is sensed by a small wing placed downstream in the flagpole barrier wake. The wing movements cause electronic pulses. Many other forces in the pipeline including pipe vibrations and flow anomalies also cause the wing to move thereby creating additional electric pulses known as noise pulses. The combined signals of the vortice pulses and the noise pulses are transmitted into an electronic computer processor. The electronic computer processor amplifies and processes the incoming pulses. Fourier analysis (1.4-1.7) and cross-correlation (2.2) techniques are used to eliminate the noise pulses from the incoming vortex pulses (1.3). The true vortex count per unit time or frequency is isolated from the noise pulses. The true vortex frequency is then used to calculate the fluid flow through the pipeline.

Description

A

SIGNAL PROCESSING METHOD & APPARATUS FOR FLO METERS

TECHNICAL FIELD The present invention relates to flowmeters and digital signal processing.

BACKGROUND ART The vortex shedding flowmeter is emerging as one of the most popular flowmeters. No moving parts are subjected to wear and tear. It has a rugged and sturdy construction suitable for applications involving extreme temperatures and pressures. It provides a highly accurate and reliable flow measurement on the volume flow rate of all types of fluids. Indeed, the vortex shedding flowmeter has the potential of the best kind of flowmeter for many applications if several persistent problems can be solved, some of which are solved by this invention.

Vortex shedding is a natural phenomenon that occurs when a fluid passes a non-streamlined body. Such bodies (bluff bodies) include smokestacks, flagpoles and bodies inserted into flowing fluids in pipelines.

Vortices are localized zones of increased velocity. They form in the fluid stream at the initiating or bluff body. These vortices detach themselves from the initiating body and are "shed" downstream in a "street". The vortices alternately shed from the opposite sides of the initiating body at a rate which is directly proportional and linear with the fluid flow. —-

When the non-streamlined object or bluff body is placed in the fluid path inside a pipeline, this same principle can be used to detect the flowrate of the fluid. By counting the vortices shed from the bluff body over a period of time, one can compute the flowrate, and, by use of the pipe dimensions and fluid characteristics, one can also compute the volumetric fluid flow in the—pipe itself. Such techniques and computations are well-known to persons skilled in this field. In some vortex flowmeter structures, such as that described in U.S. Patent No. 4,699,012, issued to H. Lew, et. al., a wing is placed downstream from the bluff body. The wing moves or vibrates from side-to-side in a direction transverse to the direction of fluid flow with each passing vortex. The rate at which the vortices are shed by the bluff body is proportional to the flowrate of the fluid. A strain gauge, piezoelectric element, or other similar transducer, attached to the wing creates an electric pulse for each passing vortex. The resulting pulse rate forms the proportional basis for the computation of the flow rate. One significant and persistent problem in modern vortex shedding flowmeters has been the inability to eliminate extraneous noise signals from vortex frequency signals. Since the vortex sensor detects the mechanical reaction of the wing to the action of vortices, the wing also picks up all other mechanical actions including structural vibrations of pipe lines, low frequency acoustical noises penetrating across the pipe wall, noises associated with flow fluctuations unrelated to the vortices, and the like. The noise problem becomes particularly serious when the bluff body and wing are mounted on the end of an elongated stem or probe inserted into the mid-portion of the fluid stream in a cantilever mount, and it is further exacerbated when it has to detect a low velocity flow of a low density fluid.

An ultrasonic means for detecting the wake created by the vortex generator provides one attempted solution to the aforementioned weakness in the present-day vortex shedding flowmeters. However, the vortex shedding flowmeter with ultrasonic wake detection has its other weaknesses of its own. For example, the error introduced by the bubbles and particles suspended in the fluid medium distort the ultrasonic signals. Also, the ruggedness needed for durability, temperature, and pressure extremes in many practical applications further limits ultrasonic vortex sensors. DISCLOSURE OF THE INVENTION

Accordingly, it is a general object of this invention to provide a more effective method and apparatus for utilizing vortex shedding and other flowmeters by deriving a more useable flowrate signal from which to measure flow.

It is also a general object of the present invention to provide a vortex or other flowmeter system having electronic means to remove or minimize noise signals from the flow vortex frequency signals, wherein said flow or vortex frequency signals are used to compute the flow.

Another general object of the present invention is to provide an insertion, probe-mounted vortex or other flowmeter system which comprises a naturally high noise environment, with electronic means to remove the noise signals from the vortex frequency signals, such that the vortex or other flow frequency signals can be used more effectively to compute the flow.

Another object of the present invention is to provide a vortex flowmeter system having a strain gauge or other appropriate transducer and integral firmware capable of eliminating noise signals from the vortex frequency signals.

Another object of the present invention is to provide a vortex flowmeter system having integral firmware that utilizes a combination of frequency domain conversions and cross correlational techniques to eliminate noise signals from the vortex frequency signal.

A more specific object of the present invention is to provide a digital signal processing technique capable of recognizing and sorting out the frequency domain, pattern of a vortex or other flow transducer signal as being different relative to the pattern of noise.

Additional objects, advantages, and novel features of this invention are set forth in part in the description that follows, and in part will become apparent to those skilled in the art upon examination of the following specification or may be learned by the practice of the invention. The objects and advantages of the invention may be realized and obtained by means of the instrumentalities and in combinations particularly pointed out in the appended claims. To achieve the foregoing and other objects and in accordance with the purposes of the present invention, as embodied and broadly described herein, the method and apparatus of this invention addresses the noise problem by applying a unique digital signal processing program to any vortex or other flow transducer signal. The program uses a combination of frequency domain conversions and comparative processing to isolate the transducer signal from the noise signals. Thus, the vortex shedding flowmeter can provide accurate readings in high noise environments.

A significant feature of this invention is the discovery and recognition of a signature characteristic waveform of the flowrate signal as distinct from noise signals in the same frequency range and then utilizing digital processing techniques to separate out and eliminate the noise signals, which do not have that signature characterJ-Sj ic. In doing so, a large number of samples, such as preferably 1024, of the vortex sensor signals are taken over a period of time, such as up to 4 seconds, at a precisely controlled sampling interval. This time-domain sampled data is then transformed into a frequency-domain power spectrum (relative power versus frequency) by means of a fast Fourier transform calculation. In order to improve the signal-to-noise ratio and reduce spectral jitter, several power spectra are taken from prior sampling periods and averaged before further signal processing is done. This results in the average power spectrum density (average PSD) . Very low frequencies are then removed from the average PSD. The resulting average PSD is then cross-correlated with a series of templates (selected noise patterns) that have bandwidths representative of noise. This results in the noise template cross-correlation. At the frequency of each peak in the noise template cross-correlation, the average PSD is checked to determine if it more closely approximates a selected characteristic vortex or flow signal, such as wide bandwidth with a Q of 3, or a noise signal, such as narrow bandwidth with a Q of 30, where Q is a measure of the 3-dB bandwidth of a signal. If the peak in the average PSD looks more like noise, then the peak, and surrounding points are removed, thus filtering out or removing the noise components of the signal. This sanitized signal results in the filtered PSD. The frequency of the vortices is then estimated by first cross-correlating the filtered PSD with a series of templates (signal templates) which have selected bandwidths similar to a vortex or other flow signal. The first estimate of the frequency is the frequency at which the maximum value occurs in this cross-correlation. The estimate can be further refined by cross-correlating another signal template centered at the estimated frequency with the filtered PSD, repeating this process if necessary. The result is an accurate estimate of the average frequency of the vortices.

The power at the estimated frequency is then calculated and compared with a minimum acceptable power level or threshold based upon the fluid density and estimated frequency. • If the power is below this threshold power, the result is set to zero, and the entire process is repeated. If the power is above this threshold level, then the electronic output circuitry is set to output a value that is proportional to the resultant estimated frequency. The flowrate is directly computed therefrom.

BRIEF DESCRIPTION OF DRAWINGS The accompanying^"drawings, which are incorporated in, and form a part of, the specifications illustrate the preferred embodiments of the invention, and together with the description serve to explain the principles of the invention. In the drawings:

Figure 1 is a schematic representation of the concepts and component parts of a vortex shedding flowmeter;

Figure 2 is a schematic view showing an insertion type vortex shedding flowmeter sensor in the center of a pipeline; Figure 3 is a side elevational view of an insertion-type vortex shedding flowmeter as it is typically mounted in a pipeline with the pipeline shown in cross section;

Figure 4(a) is a front elevational view of an insertion-type vortex shedding flowmeter sensor apparatus illustrated in Figure 4;

Figure 4(b) is a side elevational view of the vortex shedding flowmeter sensor in Figure 4;

Figure 4(c) is a longitudinal section of the sensor in Figure 4(b) with a partial cutaway of the sensor stem taken along line 4c-4c of Figure 4;

Figure 4(d) is a longitudinal section of,the sensor in Figure 4 taken along line 4d-4d of Figure 4b showing the strain gauge and wires; Figure 4(e) is a bottom cross sectional view of the sensor in Figure 4 taken along line 4e-4e of Figure 4b;

Figure 4(f) is a bottom cross sectional view of the sensor in Figure 4(a) taken along line 4f-4f of Figure 4b showing the strain gauge; Figure 5a and 5b together illustrate in .schematic how the vortex sensor produces an analog output, with Figure 5a representing a side view of the blu'ff body and- the sensor wing, and Figure 5b representing the front view of the vibrating wing and the strain gauge transducer connected to a potentiometer and outputting an analog signal;

Figures 6(a), 6(b), and 6(c) are graphs showing the relationship between time domain and frequency domain signals;

Figure 7 is a graph illustrating the concept of bandwidth and signal signatures;

Figures 8(a), 8(b), and 8(c) are graphs illustrating the concept of cross-correlations; Figure 9 is a flow diagram of the noise removal steps according to the present invention;

Figure 10 is a flow diagram of the signal measurement steps of the electronic processor according to the present invention;

Figure 11 is a flow diagram of the electronic components in the electronic processor of the present invention;

Figure 12 is a graph showing a representative raw frequency domain signal from a vortex sensor;

Figure 13 is a graph showing the average of several representative raw frequency domain signals like that of Figure 12;

Figure 14 is a graph showing the signal of Figure 13 after a cross-correlation with a vortex signal signature; Figure 15 is a graph showing the signal of Figure 14 after a second cross-correlation.

Figure 16 is a graph showing a vortex template; Figure 17 is a graph showing a noise template; Figure 18 is a graph showing a combination vortex and noise signal compared to a series of noise templates;

Figure 19 is a graph showing resultant signal derived from the cross-correlation of Figure 18;

Figure 20 is a graph showing the isolated vortex signal compared to a series of vortex templates; and

Figure 21 is a graph showing the exact location of the peak of the vortex signal which resulted from the cross-correlation of Figure 20.

BEST MODE FOR CARRYING OUT THE INVENTION The signal identification and processing techniques of this invention can be used for signals derived from any of a number of flow meter types and transducers that generate electronic signals that are indicative of flow rate and masked in extraneous noise signals. For purposes of this description, but not for limitation such signals produced by an insertion vortex shedding flow meter will be described, because they are particularly subject to noise problems .

In accordance with the above described purpose, but not for limiting the application of this invention to any particular flow meter type or transducer, the present invention can utilize an integral bluff body and vortex wing sensor assembly wherein the wing sensor reacts to the vortices created by the bluff body. The resultant alternating lift forces of the vortices generate a frequency which is sensed by the wing sensor, which produces electronic signals proportional in amplitude to the pressure peaks, and in frequencies indicative of, the vortices as well as the extraneous noises in the system. These signals are amplified, then electronically filtered to remove high frequency content, since the vortex flow signals are relatively low frequency. The signal which includes both vortex and remaining lower frequency noise components, is carried preferably to a firmware circuit where noise signals are separated from the vortex frequency signals by digital signal processing techniques. The measured frequency output of the sensor is updated periodically, such as once every ten seconds, as either a current (4-20MA) or frequency (Hz) signal directly corresponding to the flow. A final computation of flow is then made. Cross correlational techniques are widely used in the art of digital signal processing to denote the similarities of two wave forms. Specifically, the similarities between two waveforms is found by summing the lagged products of each waveform. Functionally, correlation can be thought of as a matching up of waveform components or a similarity test between waveforms. As used herein, cross correlation is used to determine whether the known shape of a vortex frequency signature is hidden in the complex waveforms of a signal having both vortex and noise components. The first step mathematically is to convert by means of FFT the combined vortex and noise waveform from the time domain to the frequency domain. The combined waveform is now stored in frequency domain patterns. The combined waveform is then multiplied by delayed versions of known narrow bandwidth noise waveforms. The resulting cross-correlation waveform contains only those frequency components common to both waveforms. Thus, the noise waveform component can be pulled from the jumbled hash of the combination vortex and noise waveform.

One type of cross-correlation mathematics may use the delayed waveform of the same waveform as a product. This product is called an autocorrelation. When the autocorrelation is processed by means of an FFT, the result is called a power spectral density (PSD) .

Referring first to Figure 1, a fluid stream in a pipe (not shown in Figure 1) flows in the direction of the arrow 44. A bluff body 1 is positioned in the flow stream. The bluff body 1 acts as a flow barrier and has a tendency to create vortices 2, 3, 4, 5, 6, 7 downstream therefrom. The vortices 2, 3, 4, 5, 6, 7 are shed from the bluff body in alternating fashion from sides 110 and 120. For a constant flow, the time differentials between vortices 2 through 7 are not exactly the same. However, the time differentials between the vortices 2 through 7 can be averaged into a fairly accurate quantity of vortices per unit time for any constant flow rate. For example, a typical average might be sixty four vortices per second (64 Hertz) . The flow rate of fluid for this 64 Hertz vortex frequency can be computed from the following formula:

Flow = (K) (Vortex Frequency in Hertz) = (K) (64) (1) K = calibrated pipeline constant

A method must be provided to sense the quantity of vortices shedding from the bluff body 1. For purposes of describing this invention, an elongated, slender wing 9 is shown in plain view in Figure 1 placed in the wake of vortices downstream from the bluff body 1. The vortices 2, 3, 4, 5, 6, 7 each have a high pressure outer wall 2', 3¹, 4¹, 5¹, 6¹, 7', respectively. These high pressure outer walls 2¹, 3', 4¹, 5' , 6' , 7¹ move or vibrate the wing 9 from side to side as indicated by arrow 445 as they pass by. The high pressure outer walls 2', 3¹, 4', 5¹, 6¹, 7' move the wing 9 in direct proportion to their relative strengths. Additionally, the vortices 2, 3, 4, 5, 6, 7 move the wing 9 once for each vortex passing by. Therefore, it is possible to measure both the strength of the vortices and the frequency of the vortices. The present invention relies primarily on the frequency of the vortices to measure flowrate, but the strength of the vortice forces on the wing 9 is used as a credibility check. A strain gauge 49 (not shown in Figure 1, but shown in Figure 4(d) ) or some other appropriate transducer device, such as a piezoelectric element (not shown) can be attached or connected to the wing 9 to produce electronic signals indicative of the frequency and strength of the shedding vortices.

In the typical operating environment of a vortex shedding flowmeter there exist many extraneous forces in the flow stream, as depicted by noise waves 10 in Figure 1. These noise waves 10, as well as mechanical vibrations in the pipe transmitted through the wing mounting structure, also move the wing 9. Such noise waves 10 are generated primarily by machinery attached to the pipeline carrying the flow stream and/or bubbles and turbulence in the flow stream. The noise waves 10 and other structural vibrations combine with the vortices 2, 3, 4, 5, 6, 7 to produce a complex electronic signal based on the movements of wing 9, such as that shown for example in Figure 8b. The present invention provides a method of separating and eliminating the noise signals from the vortex signals in order to obtain clear vortex signals from which to measure flowrate.

The bluff body/wing assembly 20 is shown in Figure 2 supported in the middle of a pipeline 21 by the stem or probe 22. This type of bluff body/wing assembly 20 and probe mounting is referred to as an insertion type flowmeter. The assembly 20 can be inserted into and removed from an existing large pipe 21, such as through a gate valve body as will be described in more detail below, without disassembling the pipe. Other embodiments of bluff body/wing assemblies (not shown) stretch across the entire diameter of the pipe and are mounted in housings or couplings positioned between, and fastened to, two sections of the pipe. The latter type of devices are referred to as in-line vortex flowmeters. The apparatus and methods of this invention can be used for signals produced by both the in-line and insertion flowmeters, but the K constant used in formula (1) above would be different. It is appropriate to note that the methods for determining the K constants for various installations and flowmeters are well-known in this field and need not be described in detail here for an understanding of this invention. It may, however, be helpful to note that calculations for the K constant for the insertion type flowmeter such as that depicted in Figure 2, add the factors of computing the average flow velocity from the.maximum flow velocity, which is measured in the center of the pipeline 21 by bluff body/wing assembly 20. The shape of the fluid flow velocity curve 23 differs for various fluids and pipe diameters. These differences are entered into the present invention by the user, preferably by use of an-input module 4.15 shown in Figure 11.

Referring next to Figure 3, an insertion-type vortex flowmeter 30 is shown mounted in a pipeline 21. The bluff body/wing sensor assembly 20 is supported in the center of the pipeline 21 by the sensor stem 22. Flanges 34 and 35 support an isolation valve 36. The isolation valve 36 can be a gate valve and allows removal of the insertion type vortex flowmeter 30 for maintenance.

A second isolation valve 37 permits the optional addition of a pressure transducer 36 to the insertion type vortex flowmeter 30. A threaded mounting assembly 38, 39 allows the handle 38 to accurately position the sensor assembly 20 in the center of the pipeline 21. Wires inside sensor stem 22 transmit the signals from the sensor assembly 20 to the electronic processor 40.

The electronic processor 40 contains all the circuitry necessary to execute the present invention. Variable user calibration parameters may also be entered into the electronic processor by means of switches (not shown) in a conventional manner.

Referring to Figure 4(a) and 4(b), the sensor stem 22 supports a sensor assembly 20. A cylindrical flow shield 43 conditions the flow around the bluff body/wing assembly 20 and isolates it from flow turbulence. The longitudinal sectional view of Figure 4(b) shows the flow from left to right, as indicated by the arrow 44. The bluff body 45 creates the vortices as described above. The flow of the vortices past the wing 48, as well as the other noises and system vibrations, cause the wing 48 to move slightly or vibrate from side-to-side as described above.

The wing 48 in this embodiment, as well as the bluff body 45, is actually fashioned out of a monolithic, unitary metal body 41 mounted on the distal end of the sensor stem 22. The shape of the wing 48, which is preferably a body in the form of a fixed free beam extending between ends fixed to upper and lower body portions 51, 52, respectively, can best be seen in Figures 4(c), 4(d), and 4(e). The free sides 54, 55 of wing 48 are separated from the body 51 by narrow cuts 57, 58, respectively. A strain gauge 49 or a pair of strain gages to detect bi-directional movement, is shown in Figure 4(f) mounted to the bottom of the wing 48 in a recess 50. The transmittal wires 47 from the strain gages 49 run through a conduit 46 extending upwardly through the center of the bluff body 45.

Figure 5 depicts the sequence of how the bluff body 45 and wing 48 produce the analog signal 500 which is indicative of the passing vortices and noise waves. The wing 48 flexes from side-to-side as shown by the arrow 445. A DC power supply 300 powers a potentiometer 400 which senses the resistance changes of the strain gauge 49. These resistance changes occur as the wing 48 vibrates from side-to-side. The potentiometer outputs an analog signal 500 proportional to, and indicative of, the movements of the wing 48.

Figure 6(a) illustrates an example ideal analog signal 500 from Figure 5 as it might look if there existed no noise and each vortex was shed exactly 1/55 second apart. The amplitude varies plus and minus because the vortices are shed alternately from sides 110 and 120 of the bluff body 1 as shown in Figure 1. Figure 6(b) shows a somewhat more realistic representation of the slightly varying instantaneous frequencies of vortices during a constant flow rate, since in reality t.he vortices are not shed at exactly the same time intervals. However, as discussed above, the average frequency over fairly short time intervals does remain quite constant for a given flow rate of a given fluid. The time domain signal of 6(b) can be converted from the time domain to the frequency domain by means of Fourier analysis in a manner well-known to persons skilled in this field, so such conversion techniques are not described herein. However, for purposes of this invention, two critical facts are shown in the frequency domain plot of the signal in Figure 6(c). First, the frequencies of Figure 6(b) can be seen to average about 55 vortices per second. Second, the signal signature of Figure 6(c) can be seen to have a broad bandwidth. The first fact that vortex frequency can be averaged to a constant frequency (55 Hertz in Figure 6(c)) allows the use of Formula (1): Flow = (K) x (Frequency) , as discussed above to compute flow. The second fact that the vortex frequency in Figure 6(c) , i.e. , flow rate frequency as produced by the vortex-shedding bluff body, has a characteristic broad bandwidth provides a recognizable signature that is distinct from typical noise signals in the frequency domain. This characteristic signature of the vortex-generated signals provides the basis for the application of mathematical templates to identify vortex generated signals from noise signals according to this invention. Specifically, a significant feature of this invention is the recognition that vortex generated frequency domain signals have relatively broad band characteristic signatures, whereas noise signals within the frequency range vicinity of the vortex generated signals have been found by extensive experimentation and analysis to have narrow bandwidths (spiked curves) and/or a complete lack of repetitiveness. In other words, noise signatures are spiked and/or random. The present invention utilizes these respective noise and vortex generated signal signatures with digital signal processing techniques to separate or pull the noise signals out of the vortex signals, thereby leaving a specifically identified, vortex generated signal similar to that shown in Figure 6(c) for use as the basis of flow measurement. For example. Figure 18 illustrates what could be a typical signal 181 derived from the strain gages 49 due to the movement of the wing 48 in actual conditions of vortex flow plus extraneous noise and system vibrations, where the signal 181 has only had high end noise frequencies filtered out electronically and where it has been converted from time domain to frequency domain by application of Fourier analysis. By mathematical comparisons and digital techniques that will be described in more detail below, a noise signal template, such as that illustrated in Figure 17, is compared at various selected frequencies to the actual frequency domain signal 181 in Figure 18. For example, the noise template 171 of Figure 17 is created with a large Q of 30, where Q is the measure of the 3-dB bandwidth of a signal as defined by the rates of the center frequency of the band or template to the bandwidth at one-half maximum amplitude. This large Q of 30 has been found to be an appropriate and effective approximation of noise signal characteristics for purposes of the digital processing described below. This noise template 171 is applied in series 182, 183, 184, 185, 186 to the actual signal 181 at selected frequency center points f_1f f₂, f₃, f₄, f₅, etc. By mathematical cross-correlation techniques between the noise templates 182, 183, 184, 185, 186 and the actual signal 181, the peaks of actual noise signals can be identified, or at least suspected, such as those represented for example by f6, fO, f7, and f8 in Figure 9. Further cross-correllation computations of each peak at f6, fO, f7, f8 are made against both a noise template 171 of Figure 17 and a vortex template 161 of Figure 16. It can be determined by this comparison which peaks more closely resemble noise signal components and which more closely resemble vortex signal components. For example, the companison described above might determine that peaks at f6, f7, and fδ more closely resemble noise signals. These noise peaks at f6, f7, and f8 can then be removed. A very close approximation of the signal 197 without the noise components is then reproduced mathematically, such as that shown in Figure 20. The frequencies f f₂, f₃, f , f₅, etc. , at which the noise template 171 is applied is somewhat arbitrary, but should be in a range where the actual vortex generated frequency is expected, in close enough intervals to accurately identify actual noise signal peaks, but not so close as to overburden the digital processing equipment or to become unnecessarily cumbersome.

Then, a vortex template 161, such as that shown in

Figure 16 is applied at various selected frequencies, such as f₁₀, f , f₁₂, etc. to the resultant curve 197 in Figure 20. This vortex template 161 is constructed with a small Q, such as, e.g., Q=3, which is more representative of the characteristic broad bandwidth signature of actual vortex generated, frequency domain signals. The f₁₀ f_n f₁₂ etc., frequencies at which the vortex template 161 is applied to the curve 197 are also chosen somewhat arbitrarily, but they should be close together enough to get sufficiently accurate cross-correlation results to be useful, yet not so close together as to overburden the digital processing equipment or to become unnecessarily cumbersome. By mathematical cross-correlation techniques between the vortex templates 191, 192, 193 and the signal 197, a fairly accurate approximation of the actual peak 195, i.e., center frequency f₀, of the vortex-generated signal, can be identified. The resultant vortex generated curve 198 with a now known and identified peak 195 and center frequency f_Q can be represented as shown in Figure 21.

There are a number of additional preferred steps, parameters, and cross checks that will be described in more detail below. However, suffice it to say that for the purposes of the above overview, once the center frequency f₀ of the actual vortex generated signal is identified as described above, the flow rate of the fluid can then be calculated in a conventional manner, as also described above.

Now, for purposes of more detailed description, Figure 7 illustrates how signal signatures are described mathematically. The bandwidth (BW) 70 of signal 71 is measured at one-half of the signal 71's maximum amplitude. The signal signature shape is represented by the symbol Q. Q for any particular signal signature equals the horizontal center point (f₀ in Figure 7) divided by the bandwidth BW.

As mentioned above, we have determined by experimentation and analysis that vortex signal signatures have a broad bandwidth which has a correspondingly low value Q. We have found that a Q of 2 to 3 is common for a vortex signature in the frequency domain. Noise was found almost always to have a spiked signature with a relatively high Q of 30. Also, some noise was found to have random patterns which do not result in any signature such as that shown by signature 71. Noise was also found to have very low and very high frequencies which show up at the extremities of the horizontal frequency axis. The noises at very high and very low frequencies can be removed by conventional electronic high pass and low pass filters. The present invention uses digital signal processing techniques to remove all other types of noise signals that have frequencies closer to, and which mask more effectively, the vortex signal.

Figure 8 illustrates the use of cross-correlation to determine if a vortex signal, which might be similar to that shown in such as Figure 8(a), lies imbedded in a complex noise signal, such as that shown in Figure 8(b). The mathematical definition of correlation is

r⁽™⁾ = ^_> ^~2T^~ _<i ^x<^t> Y^(t+W> ^{dt (2)} where r(w) = the correlation function formed by summing the lagged products of two waveforms x(t) and y(t+w) ; W = the time lag between x(t) and y(t+w) . In actuality, correlation is a similarity test between waveforms. If the two waveforms are the same, x(t) = y(t) , i.e., w=0, then their correlation is referred to as an autocorrelation. If the two waveforms are different as in Figures 8(a) and 8(b), then their correlation is referred to as cross-correlation. Thus, in the present invention a narrow band noise signal signature or template as in Figure 8(a), is cross-correlated with the noise signal of Figure 8(b) . This results in something similar to Figure 8(c) , which shows us that indeed many noise signal signatures of Figure 8(a) are imbedded in the signal of Figure 8(b).

The process of cross-correlation multiplies the signal of Figure 8(b) with delayed versions of the signal signature of Figure 8(a). The resultant cross-correlation shown in Figure 8(c) contains only those frequency components common to both waveforms.

It is appropriate to note at this point that the fundament mathematical principles utilized in the present inventio including Fourier analysis and cross correlational techniques a well-known and are not described herein. Also, Fourier analys includes fast Fourier transforms (FFT) , which formulas ha developed since about 1822. They allow the transformation o physically recognizable time-domain waveforms, such as the numbe of vortices per second in this application, into frequency domai waveforms, such as relating the amplitude of the vortices to th frequencies of their occurrences. FFT formulas are well known i the art of digital signal processing.

Cross correlation techniques are known and used by person skilled in the art of digital signal processing to denote th similarities of two waveforms. Specifically, the similaritie between two waveforms can be found by summing the lagged product of each waveform. Functionally, correlation can be thought of a a matching up of waveform components or a similarity test betwee waveforms.

As used herein, cross correlation is applied to determin whether the known shape of a vortex frequency signature is hidde in the complex waveforms of a signal having both vortex and nois components. The first step mathematically is to convert by mean of FFT the combined vortex and noise waveform from the time domai to the frequency domain. The combined waveform is then stored i ^'frequency, domain patterns. The combined waveform is the multiplied by delayed versions of known narrow bandwidth nois waveforms or templates. The resulting cross-correlation wavefor contains only those frequency components common to both waveforms Thus, the noise waveform component can be pulled from the jumble hash of the combination vortex and noise waveform. One type o cross-correlation mathematics may use the delayed waveform of th same waveform as a product. This product is called a autocorrelation. When the autocorrelation is processed by mean of an FFT, the result is called a power spectral density (PSD) Various computations using FFT, PSD and cross-correlatio methodology are used in the present invention. Numerous treatise are available to describe the precise formulas necessary t execute those well-known techniques that are applied .in th present invention. The description herein is limited to the new useful and non-obvious embodiments of FFT and cross-correlationa methodologies as applied to flow measurement.

Figure 11 illustrates the hardware utilized in the presen invention to execute the process steps of Figures 9 and 10. Constant current source 4.1 provides a constant current of approximately 1.0 mA to the strain gauge sensor 4.2. The resistance of the strain gauge sensor 4.2, which was shown as 49 in Figures 4(c)-(f), varies with the deflection of the wing 48 (described above) in response to the vortex signals and noise. With a constant current flowing through the strain gauge 49, a signal is produced, which has a voltage proportional to its resistance and a frequency indicative of the vibrations or movements of the wing 48 (not shown in Figure 11) . Preamplifier 4.3 amplifies the strain gauge signal to a level acceptable to the low pass filter. The gain of this preamplifier is preferably in the range of about 1000, and the output voltage level can be 1 mV to 2 volts peak-to-peak. Low pass filter. 4.4 can be, for example, a 4-pole low pass active filter with a cutoff frequency of 1500 Hz. This is the first stage of signal filtering, and it suppresses high frequencies that are outside of the normal expected range of frequencies generated by the vortices in th flowmeter. The output of the low pass filter 4.4 is amplified b an adjustable gain amplifier 4.5, which may have a gain adjuste under firmware control to any value from 1 to about 1023. Thi gain adjustment is desirable to accommodate a wide range of vorte signal amplitudes.

The signal is then fed to a programmable anti-aliasing lo pass filter, 4.6, which can be a 6-pole, switched capacito low-pass filter, with a cutoff frequency set by a clock input The frequency of the clock signal may be 100 times that of th cutoff frequency. This frequency is set by the programmabl square wave generator 4.12. The anti-aliasing filter cloc feedthrough filter 4.7 removes clock pulses in the output of th programmable anti-aliasing filter 4.6. These pulses are at frequency of 100 times the cutoff frequency of the anti-aliasin filter 4.6. The signal then passes through a high pass filte 4.8, which can be a single-pole RC filter with a cutoff frequenc of 0.5 Hz.

The level shifter 4.9 takes a bipolar signal and converts i to a unipolar signal for processing by the A/D converter 4.14 The input range of the level shifter used in this embodiment i -2 to +2 volts, with an output range of 0 to 4 volts.

A peak detector 4.10 is used to sense the level of t amplified signal before it is filtered by the anti-aliasing filt 4.7. This step is desirable to prevent saturation of t anti-aliasing filter 4.7. The peak detector output is sample and is used for setting the gain of the adjustable gain amplifi 4.5 under firmware control.

The analog switch 4.11 is used to select, under control the firmware, either the filtered signal for sampling, or the pe detector output. A programmable square wave generator 4. provides a square output with a frequency preferably between kHz and 320 kHz, controllable by the firmware. This frequency used directly to control the cutoff frequency of the anti-aliasi low pass filter 4.7, and it is further divided by the sample a interrupt generator 4.13. The output frequency of t programmable square wave generator 4.12 is divided by 40 provide timing for the analog to digital converter 4.14 duri signal sampling. This generates a simultaneous interrupt to t TMS320C10 digital signal processor 4.18. The 10-bit analog digital converter, at a rate determined by the programmable squa wave generator 4.12, converts the amplified and filtered anal signal to digital form, which is then read by the TMS320C digital signal processor 4.18. A user programmable input comprises a set of range switche which may be programmed by the user to select the maximum rang the range of fluid density, and calibration coefficents of t device. A read only program memory comprises 8192 16-bit wor of read only memory containing the program instructions for t TMS320C10 digital signal processor 4.18. These instructions a also referred to as firmware. A read/write data memory compris 8192 16-bit words of read/write data memory and is used f storage of the large sampled data and spectrum arrays. T digital signal processor can be a TMS320C10 digital sign processing microprocessor integrated circuit. The control log 4.19 provides the following functions: a. Input/Output port strobe circuit. b. Anti-aliasing filter clock feedthrough filter data latch. c. Analog switch select latch. d. Programmable square wave generator counter latch. e. Program memory page select latch. f. Read/write data memory address latch, g. Output circuit type sense, h. Output circuit data latch, i. Watchdog timer circuit.

There are two types of output circuits 4.20 - square wave and analog. The unit automatically senses which circuitry is installed, and provides output signals proportional to the flow.

Before referring to Figures 9 and 10, a further overview and definition of terms is provided.

"Digital Signal Processing". The concepts of digital signal processing have been established in the last forty years and are well documented in textbooks. Some of the terms used here are mentioned to give a brief terminology reference.

"Units" is used to denote the dimensionless measure of the frequency domain. The digital signal processing technique is based on sampling, at a selectable sampling rate. Performing a 1024-point complex FFT results in 1024 complex array elements, of which only the first 512 are needed to create a real powe spectrum. These 512 are described as having their "horizontal axis" in dimensionless units. [N units corresponds to an actua frequency at the sampling rate (in Hertz) multiplied by (N/1024) , where N is between 0 and 511.] The signal processing technique used here are independent of actual sampling frequency, so al the references are kept as dimensionless units.

"Cross-correlation" is a mathematical process where the inpu is two real arrays and the output is a single real array. Th output tends to be larger where both input arrays have the sam shape. A. "direct cross-correlation" is used on small array because it is the fastest method for them, and uses all rea arithmetic. "Fast convolution" is faster and thus used i performing a cross-correlation on large arrays. The complex FF of each real array is computed, the two arrays ar complex-multiplied, and then an inverse-FFT of the complex resul array is performed to get the cross-correlation array result.

"Q" is a measure of the 3-dB bandwidth of a signal, i.e. th ratio of the center frequency to the distance between th half-power frequencies (the bandwidth) . Referring next to Figure 9 the process steps utilized i executing the best mode of the present invention are described.

The Start 1.0 indicates the start of execution of the progra when the unit is first turned on. Initialization 1.1 comprise the following steps: a. Set the gain of the Adjustable Gain Amplifier 4.5 t minimum. b. Set values of constants in data memory c. Read the user input 4.-15, which determines maximu flow rate, density range and calibration coefficients of the unit d. Set the Programmable Square Wave Generator 4.1 frequency based upon the maximum flow rate. e. Set the Anti-aliasing Filter Clock Feedthrough Filte 4.7 frequency; f. Determine the number of power spectra to averag during calculation cycle. This number will range from 10 to 18 depending upon the maximum flow rate programmed by the user.

In the Initialize Calculation Cycle, the followin initialization is done: a. Set the Analog Switch 4.11 to select the Pea Detector 4.10 output; b. Set the gain of the Adjustable Gain Amplifier 4. based upon the peak detector output; c. Set the Analog Switch 4.11 to select the filtere signal from the Level Shifter 4.9 output; and d. Initialize all data arrays and set the number o spectra accumulated to zero.

In step 1.3, Sample 1024 Points, at a rate, determined b the output frequency of the Programmable Square Wave Generato

4.12 and the Sample and Interrupt Generator 4.13, the filtere analog signal is converted tα^{^}digital data and stored in th sampled data array.

The power spectrum is calculated at step 1.4 by doing a fas Fourier transform (FFT) of the sampled data, and then summing th squares of the real and imaginary components. Since the sampl data is real, the resultant power spectrum is symmetrical, and t relative power at frequencies from 0 to 511 units are used. The power spectrum is then added into a sum of power spect array in step 1.5 for later averaging. By averaging sever spectra, the signal-to-noise ratio is improved, and spectr jitter is reduced. A typical unaveraged power spectrum is sho in Figure 12. Referring again to Figure 9, the number of spect accumulated is compared at step 1.6 with the number determin during initialization step 1.1, and, if not equal, continues wi steps 1.3 to 1.6.

In step 1.7, Calculate Average Power Spectrum., the spectr values in the sum of power spectra array are divided by the numb of spectra accumulated, resulting in the average power spectr array as illustrated in Figure 13.

After noise filtering in steps 2.1 to 2.11 and sign estimation steps 3.1 to 3.12 in Figure 10, the resul representing the flow rate, is output at 1.8 in Figure 9 eith as a square wave with a frequency proportional to the flow, or a 4-20 mA analog signal proportional to the flow.

Blocks 2.1 through 2.11 represent the noise filtering step

In summary, the noise removal of steps 2.1 through 2.11 is bas on signal templates with a Q of 3, and noise templates with a of 30. The program first removes the low frequency noise from t spectrum. Then it performs a cross-correlation between a noi template and the spectrum. This helps to identify locations possible noise peaks while also performing some spectr smoothing. Peaks in the cross correlation result represe

^possible noise peaks. These peaks are evaluated by checki whether they more closely match the signal template or the noi template. If they appear to be noise peaks, they are removed a replaced by a linearly interpolated set of points. This progr performs this noise removal algorithm twice. ^{~^}-~*Dire cross-correlation is used for all noise filtering functi cross-correlations. Specifically, in step 2.1, Remove Low Frequency Data, t criteria for removal of very low frequency noise is to chec beginning at a frequency of 1 unit, to see whether the next val is smaller than the current value. When the next value is n smaller than the current value, the checking process stops. T current value becomes the top endpoint for interpolation. T bottom endpoint for interpolation always has a value of 0 at units. Any points between 0 and the top endpoint are linear interpolated (straight-line interpolation) . The Noise Template Cross Correlation step 2.2 involv cross-correlating a series of preferably six noise templates (t last five of which have a nominal Q of 30, and are respective centered at 20, 40, 80, 160 and 320 units) with the sign spectrum. In this case, the result is called the noi cross-correlation array. An example of the result is illustrat in Figure 14.

The cross-correlation is computed in six pieces, one pie for each noise template. The start and end points for each pie are: template #1, 0 to 13; #2, 14 to 27; #3, 28 to 56; #4, 57 112; #5, 113 to 225; #6, 226 to 510 units. For templates #2 th #6, their start and end points were determined as follows. T start points are each template's center divided by the square ro of 2, and the end points are its center times the square root 2, with the special case that the template centered at 320 has end point at 510 units.

The template values are stored constants, stored alre normalized, packed together into adjacent array elements. templates are chosen with a size estimate based on the 10 perc power level of the function. The function used to compute values of the template is

1 (3)

where a is the normalized frequency (a=f/f₀, where f₀ is center frequency) , and Q is 30 for noise templates. The templa are meant to be used with direct cross-correlation. The templa are each scaled to unit area by keeping a running sum of th values and then dividing each value by the total sum. A templat extraction method is used that extracts them directly out, one a a time when needed. Template #1 is a single point template, so for simplicity th power spectrum elements 0 thru 13 are copied into th cross-correlation result array.

For the last five templates, the piece of th cross-correlation that is performed is done in the followin manner. Given a template consisting of elements tp[i], and a powe spectrum density consisting of elements psd[i], the 'direc cross-correlation array consisting of elements c[t] is compute for each value of t (the range of t will be the same as the rang of the template tp) as follows: A summation is performed over the range of i from summatio start to summation end. Summation start is at t-center, an summation end is at end+t-center. The summation is of th products of tp[i-t+center] times psd[i].

This procedure is executed twice, and an example of th result of the second pass is illustrated in Figure 15.

Referring again to Figure 9, in the step 2.3 labeled Fin Minimum and Maximum Values, all relative minimum and maximu values in the noise cross correlation array are found. Th frequencies , in units, at which the minima and maxima occur ar stored in the min-max array. Next, at step 2.4, F_c = Frequency a First Maximum., the process of finding and removing noise peak is performed at the location of each maximum, starting with th location of the first maximum value found in the nois cross-correlation array. In Determine Valid Range, step 2.5, the valid range extend as wide as it can, from the larger of: ^~~

1. The location of the minimum at the left side of thi maximum,

2. The location of the left side 10% level of a "nois template" (a template with a Q characteristic of noise {call i

Q_n} ; Q_n = 30 has been used here but other values could be use the template is centered at the current maximum, and has a valu of unity at its center) ; to the smaller of:

1. The location of the minimum at the right side of th maximum, 2. The location of the right side 10% level of the "noi template" centered at the current maximum,

3. 511 units

The location of the left side 10% level of a "noise templat can be accurately estimated for Q_n greater than 2 as the truncat integer value of center frequency, divided by Q_n, times (1 min a factor) , or:

left 10% level location = center /ι-^fag^torj (₄)

The factor is dependent on the Q characteristic of noi in the following manner:

factor (5)

where P-stands for the power level. Since P=0.1 (derived from t percent of unity when the template has a value of unity at i center), then if Q_n = 30, the factor is 0.05.

Similarly the location of the right side 10% level of

"noise template" can be accurately estimated for Q_n greater th 2 as the truncated integer value of center frequency, divided

Q_n, times (1 plus a factor) , plus one, or:

right 10% level location = 1+center (6)

The next step 2.6 is to calculate spectrum fit to template Over the valid range the shape of the actual data is evaluate The signal is assumed to have a nominal Q of 3 whereas the noi is assumed to have a nominal Q of 30 as specified above. If t noise model provides the better match to the data in the val range, then points within the valid range are marked for remova

If the valid range has only 3 points, interpolation of a mo accurate estimate of the location of the peak based on a best f of a noise template to the data is performed. This step necessary for reasonable discrimination between signals and noi when the noise peak lies between adjacent spectral points. T peak is interpolated to an accuracy of 1/4 unit, by comparing i fit to templates between left and center at 1/4 unit interval and then between right and center.

The fit of the power spectrum to the noise template is th compared at the step 2.7, Better Fit to Noise Template?, to t fit of the power spectrum to the signal template. This is direct comparison of two numeric values. If the fit to the noi template is the larger of the two numbers, then the next step performed, i.e., Mark Points for Removal 2.8. In this next st

2.8, if the fit of the power spectrum to the noise template higher than its fit to the signal template, all points in t valid range (determined in step 2.5) are marked for removal. addition, if the left minimum point is surrounded by points mark for removal, it should also be marked for removal. Steps 2 through 2.10 are repeated for all maximum values in the min-m array, as the F_c = Frequency of Next Maximum 2.10 is increment for each cycle.

When the process is done checking all maxima, as shown

2.9, it then proceeds to Remove Marked Points 2.11. The locatio of all of the points to be removed are stored in the remove arra For each sequence of points marked to be removed, the points the power spectrum are replaced by a straight line. If t magnitude of a point to be removed is less than the interpolat straight line value, it is not replaced.

Referring now to Figure 10, Blocks 3.1 through 3.12 represe the signal estimation steps. This routine performs sign estimation for the insertion vortex meter. The average noise-filtered power spectrum density is cross-correlated wi seven signal templates for a first-pass estimate. A sign template is then constructed at that frequency and is aga cross-correlated with the spectrum for up to two iterations. result is then curve-fitted to determine the signal peak w better resolution. A final check is made by verifying that result is consistent with criteria of signal strength and no minimization.

In Signal Template Cross Correlation 3.1, the first pass template method requires extraction of seven templates to cove the full spectrum. For ease of implementation, the templates ar equally spaced on a log scale by a factor of 2. The templates hav center frequencies of 5, 10, 20 ,40 ,80, 160 and 320. Th templates are normalized to unity area (sum) so that they are no biased when used with a white noise background. A valid range ha been established for each template from f λ to f₀(\[2~). Each of the templates is partially cross-correlated with th spectrum. This is done directly because it is efficient for th low frequency templates of 5, 10, 20, 40, and 80 units, and vi fast convolution using the FFT, which is more efficient for th higher frequency 160 and 320 unit templates. The working templat was generated by taking the signal template and circularl shifting it so that its peak is at zero. When the convolution i employed, mirrored versions of templates are used. Mirroring o the template refers to mirroring about the origin.

Within the valid range for each template, the largest valu in the cross-correlation and the corresponding frequency i checked to see if the value is as large as the previous maximum If it is, then the frequency is assumed to be located close to th actual signal and the frequency corresponding to this larges value is chosen as the first signal frequency estimate. In th extremely unlikely condition that multiple frequencies end.up wit the same value the lowest frequency of the two is assumed mos accurate.

The inputs to this routine are the power spectrum density an the complex FFT of it. The complex array is expected to have th format required by the 32010 assembly language FFT routine, wher the real and imaginary parts are interleaved. The power spectru density is intended to have only 512 points used out of it; th other 512 are expected to be zero-filled. The routine uses th ROM template. The integer first-pass signal frequency estimat is returned. This routine also produces an averag cross-correlation average of the cross-correlation over the entir spectrum range it examines, for use in the final check routin 3.10.

In Set ITERATION = 1, step 3.2, the process of finding th signal peak is performed twice if needed, and an indicator fla is needed. Here it is set to indicate that this is the firs 5 iteration.

In Set CENTER=Frequency at Maximum Value of Cro Correlation., step 3.3, the purpose is to find the maximum val in the cross correlation array, and set CENTER equal to t frequency at which this maximum value occurs. Then in step 3.4

10. Cross Correlate Spectrum with Signal Template at CENTER, once first-pass signal frequency estimate "peak" has been made, a n signal template is generated at this frequency. This new templa is then cross-correlated with the spectrum. A" mirrore zero-centered template is used in a fast convoluti

15 implementation of the cross-correlation in order to end up wi a peak near the signal template peak. The array "x" has alrea been FFT'd, so it is ready for convolution with the template th gets generated.

In step 3.5, Is Cross Correlation Peak Close Enough

20 Center?, the convolution result is checked to find the peak a 2:1 range around the signal template peak, from template pe divided by the square root of two to template peak times t square root of two. If the convolved peak is within "one perce or one unit (whichever is greater)" of the signal template pea

25 then the next step is interpolation to the actual peak. If it not within those limits after one iteration, then the convoluti is repeated with a new template at the new estimated sign template peak. If after the second iteration it is still n within those limits, then this is a failure. This routine chec

30 whether bpeak is close enough to apeak. The apeak is consider to be a signal template peak, while bpeak is the result of convolution. The bpeak should be apeak +/- 1 unit, or bpe should be within one percent of apeak, whichever is larger. bpeak is close enough, the routine returns a 1; if it is not clo

35 enough, this routine returns a 0.

Step 3.6, ITERATION = 1?, is provided so that the process finding the signal peak is performed twice, if needed, and indicator flag is needed. Here it is checked to determine if is the first iteration. Step 3.7, Set ITERATION = 2, is provid so that the process of finding the signal peak is performed twic if needed, and an indicator flag is needed. Here it is set indicate that this is the second iteration.

In step 3.8, Is Spectrum Data Satisfactory f Interpolation?, first, a check is made to find the peaks in t convolution result. If two adjacent peaks are found, then t interpolated peak is at the center of them. If more than t adjacent points are equal, or two non-adjacent peak points a equal, then this is a failure, and the output is set as if no fl (zero frequency) occurred. Also, if the failure indicator is s ( second peak was 0) then interpolation is not performed.

Generally, in step 3.9, interpolation is performed by parabolic curve fit to the three points nearest and including t peak "*peak". If only one point is the peak (which will almo always be the case) then a parabolic curve fit is performed usi three points, this point and the point on each side. Th parabolic curve fit uses the derivatives from Cramer's rule f solving the three simultaneous equations of the form

y = ax² + bx + c (7)

with the solution x - -~D,— (8)

⁰ 2D, ' '

where D^y,(x₂-x₃)+y₂(x₃-x.,)+y₃(x.,-x₂) (9)

and D₂=x₁ ²(y₂-y₃)+x₂ ²(y₃-y₁)+x₃ (y₁-y₂) (10)

The solution x₀ is the frequency peak between the points (x,,^

(^ ^■ y ) ' ^and (^χ3'Y3) ^•

The final check at step 3.10 determines whether the res of the entire functional algorithm is reasonable.

In the steps Second-Pass Estimate and Interpolation to Sig Peak there were failure conditions. If after two steps of

Second-Pass Estimate the result is still not within reasona limits, then this was a failure. If there are the wrong type peaks in the Interpolation to Signal Peaks, then this was failure. If either of these conditions occur, the output was se as if no flow (zero frequency) occurred. Those checks were don in those routines and are not done in this routine. The peak from the Second Pass step must be at least frequency-dependent ratio times the cross-correlation averag (from First Pass, over the entire range) value of th cross-correlation function, in order to prevent very smal cross-correlation peaks from being incorrectly recognized a signals. Note that this is not the interpolated peak value, onl because for it extra computation of derivatives would have bee necessary in the interpolation step, and the second pass pea value is expected to be very close to the (not-calculated interpolated peak value. If the peak is less than the ratio time the average, then the output is set as if no flow (zer frequency) occurred.

The final rejection criteria in step 3.10 comprises tw checks. The first is a check of the signal amplitude as function of density and frequency. The second is a check of th ^' the signal frequency as a function of density.

1. First Check

The signal power must be greater than a minimum limit whic is determined as a function of density and frequency. T magnitude of the power spectrum at the estimated signal frequen is compared with a threshold value which is determined from t square- of the density and the fourth power of the estimat frequency. If the magnitude is less than the threshold value, t output is set as if no flow (zero frequency) occurred.

2. Second Check The signal frequency must be greater than a minimum lim which is a table lookup determined as a function of density. T density is shown in pounds per cubic foot. Density Minimum lb/ft³ Frequency, Hz <0.1 150

<0.25 75

<0.5 50 Gas , >0.5 38

Liquid 5

In step 3.11, Result = Interpolated Peak, if the final che criteria are passed, the resultant frequency is assumed valid f outputting. However, if the final check criteria are not pass the output is set as if no flow (zero frequency) occurred, shown at 3.12.

The last step of the process simply calculates the flow rate from Formula (1) , i.e. Flow = K Frequency in the

Digital Signal Processor 4.18 of Figure 11 and outputs a useable signal by means of the output circuit 4.20 of Figure

11.

The foregoing is considered as illustrative only of the principles of the invention. Further, since numerous modifications and changes will readily occur to those skilled in the art, it is not desired to limit the invention to the exact construction and operation shown and described, and accordingly all suitable modifications and equivalents maybe resorted-to falling within the scope of the invention as defined by the claims which follow.

Claims

The embodiments of the invention in which an exclusive property or privilege is claimed are defined as follows:

1. A method of measuring the flow of a fluid in a pipeline, wherein said pipeline has noise, comprising the steps of: a) means for inserting a flow barrier in the pipeline, thereby forming a wake of vortices; b) means for providing a means for creating an electronic signal for each vortex, wherein said means for creating an electronic signal also produces electronic signals for the noise; c) means for transmitting the electronic signals of the vortices and noise to an electronic processor; d) means for converting the electronic signals of the vortices and noise vibrations from the time domain to the frequency domain; e) means for removing the electronic noise signals from the electronic vortex signals by means of cross-correlational computations, thereby computing the vortex frequency; and f) means for computing the flow from the computed vortex frequency.

2. The method of measuring the flow of a fluid in a pipeline of claim 1, further comprising the steps of: g) inserting the flow barrier of step (a) further comprising a bluff body; h) providing a means for creating an electronic signal for each vortex of step (b) further comprising a wing and strain gauge assembly; i) providing the electronic processor of step (c) with amplifiers, filters, an analog to digital converter, and a central processing unit.

3. The method of measuring the flow of a fluid in a pipeline of claim 1, further comprising the steps of: j) filtering out any high frequency content of said transmitted electronic signals of step (c) .

4. The method of measuring the flow of a fluid in a pipeline of claim 1, further comprising the steps of: k) converting the signals of step (c) to digital signals by repetitively sampling the signals 1024 times each four seconds;

5. The method of measuring the flow of a fluid in a pipeline of claim 1, further comprising the steps of:

1)converting the electronic signals of step (d) using fast Fourier transform (FFT) calculations.

6. The method of measuring the flow of a fluid in a pipeline of claim 1, further comprising the steps of: m)storing the results of step (d) ; n)averaging said stored results, thereby computing the average power spectrum density (PSD) of the electronic signals of the vortices and noise vibrations.

7. The method of measuring the flow of a fluid in a pipeline of claim 6, further.comprising the steps of: o) removing very low frequencies from the average PSD.

8. The method of measuring the flow of a fluid in a pipeline of claim 7, further comprising the steps of: p) cross-correlating the average PSD with approximately six narrow bandwidth (high Q) signals representative of the noise signals; q) storing the resultant cross-correlations as noise template cross-correlations having peaks; r) comparing each noise template cross-correlation peak first with a wide bandwidth (low Q) vortex signal and then with a narrow bandwidth (high Q) noise signal; s) eliminating the noise template cross-correlations resulting from (r) which are more similar to narrow bandwidth (high Q) noise signals; t) storing the results of (s) as the filtered PSD, wherein the filtered PSD has at least one peak; u) estimating the vortex frequency from the filtered PSD by means of cross-correlating the peak(s) of the filtered PSD with at least six broad bandwidth (low Q) vortex signal templates, repeating this step, and computing the best estimate of the vortex frequency therefrom; v) comparing the best estimate of the vortex frequency in step (u) with a minimum threshold value of the amplitude of said best estimated vortex frequency; w) repeating steps (p) through (v) if said amplitude of said best estimated vortex frequency is less than said minimum threshold value; x) computing the flow of the fluid in the pipeline from the best estimated vortex frequency if the amplitude of said best estimated vortex frequency is above said minimum threshold value; y) displaying the flow computed in step (x) ; and z) repeating steps (m) through (y) every ten seconds.

9. A method of measuring flow using a vortex shedding flowmeter, comprising the steps of: aa) mounting the vortex shedding flowmeter in a pipeline; bb) creating electronic signals from the vortex shedding flowmeter proportional to the quantity of vortices produced by said vortex shedding flowmeter; cc) transmitting said electronic signals and any other noise signals to an electronic processor; dd) converting the transmitted electronic signals of step (cc) from the time domain to the frequency domain by means of Fourier analysis; ee) removing the electronic noise signals from the electronic vortex signals by means of cross-correlational computations, thereby computing the vortex frequency; and ff) computing the flow from the vortex frequency computed in step (ee) .

10. A method of measuring flow, comprising the steps of: aaa) mounting a vortex shedding flowmeter in a pipeline, wherein said vortex shedding flowmeter further comprises a transducer producing electronic signals proportional to the shedding vortices and pipeline noise; bbb) inputting said transducer signals into an electronic processor; ccc) converting said transducer signals from the time domain to the frequency domain; ddd) removing all but the shedding vortex signals in step (ccc) by means of cross-correlations; eee) computing the flow from the remaining shedding vortex signal of step (ddd) .

11. A method of producing an electronic signal proportional to flow, comprising the steps of: i) mounting a vortex shedding flowmeter in a pipeline, wherein said vortex shedding flowmeter further comprises a transducer producing electronic signals proportional to the shedding vortices and noise; ii) converting said transducer signals from the time domain to the frequency domain; and iii) removing the noise signals from the vortex signals by means of signal signature comparisons.

12. The method of producing an electronic signal proportional to flow of claim 11, further comprising the steps of: iv) averaging and digitizing the frequency domain signals of step (ii) ;

13. The method of producing an electronic signal proportional to flow of claim 11, further comprising the steps of: v) comparing the repetitive low Q signal signature of the vortex signal relative to the random and/or high Q signals of noise as the means of signal signature comparisons of step (iii) .

14. The method of producing an electronic signal proportional to flow of claim 11, further comprising the steps of: vi) using Fourier analysis in step (ii) to convert said transducer signals from the time domain to the frequency domain.

15. The method of producing an electronic signal proportional to flow of claim 11, further comprising the steps of: vii) using cross-correlational computations in step (iii) for removing the noise signals from the vortex signals.

16. An apparatus for measuring the flow of a fluid in a pipeline, wherein said pipeline has noise, comprising: a) means for inserting a flow barrier in the pipeline, thereby forming a wake of vortices; b) means for providing a means for creating an electronic signal for each vortex, wherein said means for creating an electronic signal also produces electronic signals for the noise; c) means for transmitting the electronic signals of the vortices and noise to an electronic processor; d) means for converting the electronic signals of the vortices and noise vibrations from the time domain to the frequency domain; e) means for removing the electronic noise signals from the electronic vortex signals »by means of cross-correlational computations, thereby computing the vortex frequency; and f) means for computing the flow from the computed vortex frequency. g) filtering out any high frequency content of said transmitted electronic signals of step (c) . h) converting the signals of step (c) to digital signals by repetitively sampling the signals 1024 times each four seconds; i) converting the electronic signals of step (d) using fast Fourier transform (FFT) calculations. j) storing the results of step (d) ; k) averaging said stored results, thereby computing the average power spectrum density (PSD) of the electronic signals of the vortices and noise vibrations.

1) removing very low frequencies from the average PSD. m) cross-correlating the average PSD with approximately six narrow bandwidth (high Q) signals representative of the noise signal^'s; n) storing the resultant cross-correlations as noise template cross-correlations having peaks; o) comparing each noise template cross-correlation peak first with a wide bandwidth (low Q) vortex signal and then with a narrow bandwidth (high Q) noise signal;

17. An apparatus for measuring flow using a vortex shedding flowmeter, comprising: aa) means for mounting the vortex shedding flowmeter in a pipeline; % bb) means for creating electronic signals from the vortex shedding flowmeter proportional to the quantity of vortices produced by said vortex shedding flowmeter; cc) means for transmitting said electronic signals and any other noise signals to an electronic processor; dd) means for converting the transmitted electronic signals of (cc) from the time domain to the frequency domain by means of Fourier analysis; ee) means for removing the electronic noise signals from the electronic vortex signals by means of cross-correlational computations, thereby computing the vortex frequency; and ff) means for computing the flow from the vortex frequency computed in (ee) .

18. An apparatus for measuring flow, comprising: aaa) means for mounting a vortex shedding flowmeter in a pipeline, wherein said vortex shedding flowmeter further comprises a transducer producing electronic signals proportional to the shedding vortices and pipeline noise; bbb) means for inputting said transducer signals into an electronic processor; ccc) means for converting said transducer signals from the time domain to the frequency domain; ddd) means for removing all but the shedding vortex signals in (ccc) by means of cross-correlations; eee) means for computing the flow from the remaining shedding vortex signal of (ddd) .

19. An apparatus for producing an electronic signal proportional to flow, comprising: i) means for mounting a vortex shedding flowmeter in a pipeline, wherein said vortex shedding flowmeter further comprises a transducer producing electronic signals proportional to the shedding vortices and noise; ii) means for converting said transducer signals from the time domain to the frequency domain; and iii) means for removing the noise signals from the vortex signals by means of signal signature comparisons. iv) means for averaging and digitizing the frequency domain signals of step (ii) ; v) means for comparing the repetitive low Q signal signature of the vortex signal relative to the random and/or high Q signals of noise as the means of signal signature comparisons of step (iii) . vi) means for using Fourier analysis in step (ii) to convert said transducer signals from the time domain to the frequency domain. vii) means for using cross-correlational computations in step (iii) for removing the noise signals from the vortex signals.

20. In a method of measuring flowrate of a fluid in a pipeline having extraneous noise, wherein a transducer is used to generate a timme domain flow signal having a frequency indicative of the flowrate of the fluid, but where the noise results in noise signals that are superimposed over the flow signal in frequency ranges in the frequency range of the flow signal, the improvement comprising: transforming the time domain signals into frequency domain signals; identifying a stable signature characteristic of the frequency domain flow signal that is dependably different than the frequency domain noise signals; utilizing said signature characteristic of the flow signal to separate the noise signals from the flow signal; identifying the frequency of the flow signal; and determining the flowrate of the fluid as a function of the frequency of the flow signal.

21. The improvement of claim 20, including the step of identifying the Q value of the flow signal and comparing it to the Q value of noise signals in frequency and amplitude ranges comparable to the frequency and amplitude range of the flow signal, and formulating said signature characteristic in terms of a distinctive Q value.

22. The improvement of claim 21, including the step of removing signal components that have Q values substantially different than the Q value of said flow signal.

23. The improvement of claim 22, including the steps of creating a noise template with a Q value similar to the noise signal components, superimposing said noise template on the signal, and using cross correlation techniques to identify and remove the noise components of the signal.

24. The improvement of claim 23, including the steps of creating a flow template with a Q value similar to the flow signal, and using cross correlation techniques to identify the frequency of the flow signal.

25. Flow meter apparatus for measuring the flow rate of a fluid flowing in a pipe, comprising; a monolithic stem body adapted for being positioned in the flowing fluid and having a blunt, bluff body surface portion facing the flowing fluid, a narrowed plate portion trailing said bluff body surface portion, a thin wing portion extending at both its upper and lower ends from upper and lower portions of said stem body, but both the leading and trailing edges of said wing portion being separated from said narrowed plate portion, said upper portion of said stem protion being shaped in a smooth, rounded contour at its edge that faces the fluid flow and contoured smoothly to a converging, slim trailing edge; a cylindrical shroud mounted on and encircling the blbody, narrow plate, and vane portions of said stem b; an elongated probe extending from the upper portiof said stem body and adapted for insertion

5 t roughe wall of said pipe; and transducer means associated with said wing ortionr sensing flexures of said wing portion and creatinlectrical signals indicative of said flextures.

26The flow meter apparatus of claim 25, including signal )cessing means for determining flow rate of the fluid fάng in the pipe from frequency of said wing

!^' flexure; i 27 The flow meter apparatus of claim 26, wherein i

[ said siςl processing means removes noise components from

I the sigr. by identifying signature characteristics of the f flow sigi that are different than the noise signals and i 5 using sa signature characteristic to remove the noise f signals.