Identifying Liquid Rheological Properties From Acoustic Signals

Abstract

The disclosure relates to methods and apparatus for identifying rheological properties of liquids from acoustic signals generated by liquid flow through a pipe. Example embodiments include a method of identifying a rheological property of a liquid flowing in a pipe (101), the method comprising: detecting an acoustic signal generated by the liquid flowing in the pipe using a sensor (105) attached to a rod (104) extending from a wall of the pipe (101) into the liquid; sampling the acoustic signal to provide a sampled acoustic signal; transforming the sampled acoustic signal to generate a sampled frequency spectrum; correlating the sampled frequency spectrum with a stored frequency spectrum from a database of stored frequency spectra of liquids having predetermined rheological properties; and identifying a rheological property of the liquid based on the stored frequency spectrum.

Claims

1. A method of identifying a rheological property of a liquid flowing in a pipe, the method comprising: detecting an acoustic signal generated by the liquid flowing in the pipe using a sensor attached to a rod extending from a wall of the pipe into the liquid; sampling the acoustic signal to provide a sampled acoustic signal; transforming the sampled acoustic signal to generate a sampled frequency spectrum; correlating the sampled frequency spectrum with a stored frequency spectrum from a database of stored frequency spectra of liquids having predetermined rheological properties; and identifying a rheological property of the liquid based on the stored frequency spectrum.

2. The method of claim 1, wherein the rod extends to a centre of an interior volume of the pipe.

3. The method of claim 1, wherein the pipe comprises an obstruction upstream of the rod, the obstruction configured to increase a pressure drop along the pipe by more than 10%.

4. The method of claim 1, wherein an internal cross-section of the pipe varies one of an upstream direction and a downstream direction of the acoustic sensor.

5. The method of claim 1, wherein the rheological property is at least one of (a) a yield shear stress τ.sub.0, (b) a flow index n and (c) a consistency k of the liquid, based on a rheological model of τ=τ.sub.0+k{dot over (γ)}.sup.n, where τ is a shear stress and {dot over (γ)} is a shear rate.

6. The method of claim 1, wherein the step of correlating the sampled frequency spectrum with a stored frequency spectrum is performed using a machine learning algorithm.

7. The method of claim 1, wherein the liquid flowing in the pipe is a single phase liquid.

8. The method of claim 1, wherein the pipe is fully flooded with the liquid flowing in the pipe.

9. The method of claim 1, wherein the sampled frequency spectrum comprises a plurality of sections defining a portion of the sampled frequency spectrum, each section being defined by a parameter representing an amplitude of the acoustic signal within the portion of the sampled frequency spectrum, the database comprising stored frequency spectra having a corresponding plurality of sections and parameters.

10. The method of claim 9, wherein each of the sampled and stored frequency spectra is defined by between 10 and 100 parameters.

11. The method of claim 1, performed as part of monitoring a manufacturing process of a liquid, the method comprising: performing a mixing process on the liquid; passing the liquid through a pipe; and performing the method of claim 1 to identify a stage of the manufacturing process.

12. A computer program comprising instructions to cause a computer to perform the method according to claim 1.

13. An apparatus for identifying a rheological property of a liquid flowing in a pipe, the apparatus comprising: a pipe through which the liquid is arranged to flow, the pipe comprising an acoustic sensor attached to a rod extending from a wall of the pipe into an internal volume of the pipe, the acoustic sensor arranged to detect an acoustic signal generated by the liquid flowing in the pipe; a computer connected to the acoustic sensor and configured to: sample the acoustic signal to provide a sampled acoustic signal; transform the sampled acoustic signal to generate a sampled frequency spectrum; correlate the sampled frequency spectrum with a stored frequency spectrum from a database of stored frequency spectra of liquids having predetermined rheological properties; and identify a rheological property of the liquid based on the stored frequency spectrum.

14. The apparatus of claim 13, wherein the rod extends to a centre of an interior volume of the pipe.

15. The apparatus of claim 13, wherein the pipe comprises an obstruction upstream of the rod, the obstruction configured to increase a pressure drop along the pipe by more than 10%.

16. The apparatus of claim 13, wherein an internal cross-section of the pipe varies upstream and/or downstream of the acoustic sensor.

17. The apparatus of claim 13, wherein the rheological property is at least one of (a) a yield shear stress τ.sub.0, (b) a flow index n and (c) a consistency k of the liquid, based on a rheological model of τ=τ.sub.n+k{dot over (γ)}.sup.n, where T is a shear stress and {dot over (γ)} is a shear rate.

18. The apparatus of claim 13, wherein the computer is configured to correlate the sampled frequency spectrum with the stored frequency spectrum using a machine learning algorithm.

19. The apparatus of claim 13, wherein the sampled frequency spectrum comprises a plurality of sections defining a portion of the sampled frequency spectrum, each section being defined by a parameter representing an amplitude of the acoustic signal within the portion of the sampled frequency spectrum, the database comprising stored frequency spectra having a corresponding plurality of sections and parameters.

20. The apparatus of claim 19, wherein each of the sampled and stored frequency spectra is defined by between 10 and 100 parameters.

21. The apparatus according to claim 13, comprised in a system for processing a liquid, the system further comprising: a mixing tank for containing the liquid; and a measurement loop arranged to divert liquid to and from the mixing tank; wherein the pipe of the apparatus forms part of the measurement loop, the apparatus being configured to measure a rheological property of the liquid passing through the measurement loop.

Description

DETAILED DESCRIPTION

[0034] The invention is described in further detail below by way of example and with reference to the accompanying drawings, in which:

[0035] FIG. 1 is a schematic diagram of fluid flow in a pipe with an acoustic sensor;

[0036] FIG. 2 illustrate a range of shapes of example obstructions for use with the apparatus of FIG. 1;

[0037] FIG. 3 is an example sampled acoustic signal in the absence of an obstruction;

[0038] FIG. 4 is an example sampled acoustic signal in the presence of an obstruction;

[0039] FIGS. 5 and 6 are example frequency spectra for Newtonian liquids;

[0040] FIGS. 7, 8 and 9 are example frequency spectra for liquids exhibiting power law rheological behaviour;

[0041] FIGS. 10, 11 and 12 are example frequency spectra for liquids exhibiting Herschel-Bulkley rheological behaviour;

[0042] FIG. 13 is a plot of accuracy as a function of number of features for various machine learning algorithms;

[0043] FIG. 14 is a flow diagram illustrating an example method of identifying a rheological property of a liquid flowing in a pipe;

[0044] FIGS. 15a-c illustrate an acoustic sensor with alternative pipe configurations FIG. 16 is a schematic cross-sectional diagram of an acoustic sensor attached to a pipe in an alternative configuration;

[0045] FIG. 17 is a schematic diagram of a liquid processing system incorporating an apparatus with an acoustic sensor;

[0046] FIG. 18 is a confusion matrix illustrating true class versus predicted class for trained data over a range of stages in a liquid mixing process;

[0047] FIG. 19 is a confusion matrix illustrating true class versus predicted class for unseen data over a range of stages in a liquid mixing process; and

[0048] FIG. 20 is an example plot of sets of rheological parameters derived from sampled acoustic spectra for multiple stages in a liquid mixing process.

[0049] FIG. 1 illustrates schematically an apparatus 100 for measuring rheological properties of a liquid flowing through a pipe 101. The flow direction through the pipe 101 is indicated by arrows 102, 103. A rod, or pin, 104 is inserted through a wall of the pipe 101 and extends into an internal volume of the pipe 101, i.e. into the liquid flow path. An acoustic sensor 105 is attached to an end of the rod 104 on the outer wall of the pipe 101. As the liquid flows along the pipe 101, acoustic events are set up within the pipe that depend on rheological properties of the liquid and on changes in fluid flow patterns. The rod 104 itself affects the fluid flow patterns in the flow path, generating acoustic events that are detected by the acoustic sensor 105 by being transmitted along the rod 104. The acoustic events do not require the flow to be multi-phase, but are detectable for single phase liquid flow, and in the case of a fully flooded system. The liquid may, however, contain more than one component, for example in the case of a colloidal suspension of solid, liquid or gas in liquid.

[0050] To further enhance the acoustic events generated in the fluid flow, an obstruction 106 may be provided in the pipe 101, the obstruction 106 being positioned upstream of the rod 104. A computer 107 is connected to the acoustic sensor 105 to obtain and sample acoustic signals from the sensor 105 and to perform analysis of the signals as described below.

[0051] The obstruction 106 may be a simple narrowing of the internal bore of the pipe 101 or may be a more complex shape. Some examples of possible shapes of obstruction are illustrated in FIG. 2, including a conical shape (FIG. 2a), multiple passages (FIG. 2b), a semicircular section (FIG. 2c) and a cross shape (FIG. 2d). The obstruction may comprise a flange portion 201 arranged to seal against an end of the pipe 101 and an obstruction portion 202 configured to provide a restriction of the internal volume of the pipe 101. Each of the example obstructions shown in FIG. 2 will tend to increase the pressure drop along the pipe 101 to some extent. The degree to which the obstruction 106 increases the pressure drop along the pipe 101 relates to the degree to which the acoustic signals are enhanced such that rheological properties of the liquid can be determined. As an example, in the absence of an obstruction the pressure drop along the pipe 101 may be as small as around 2%, but with an obstruction the pressure drop may increase by at least 10%, and may be 20% or more. An upper limit on the pressure drop is dependent on the liquid and fittings, but may be less than around 3 bar.

[0052] The rod 104 may be solid or may be hollow, for example including an internal cavity that is not open to the liquid flowing through the pipe 101. A hollow rod may allow for enhancement of acoustic signals detected by the sensor 105.

[0053] In an example experimental apparatus, a stainless steel pipe of 120 mm in length with a 25.4 mm diameter internal bore was used, into which a circular section rod of around 10 mm in diameter was inserted, the rod extending into the middle of the internal bore. Acoustic emission signals were captured with a piezoelectric VS375-M sensor (Vallen Systeme GmbH, Germany), linked to a 2.5 kHz to 2.4 MHz (10 Vpp) AEP5H preamplifier (Vallen Systeme GmbH, Germany) along with a DCPL2 decoupling unit (Vallen Systeme GmbH, Germany), a PicoScope 5000 Series oscilloscope (Pico Technology Ltd, UK) and a personal computer using PicoScope version 6.13.15 software (Pico Technology Ltd, UK). Liquid was pumped through the pipe from a tank, and recirculated back into the tank. Flow rates were adjustable to allow for measurements to be taken in laminar, transitional and turbulent flow conditions.

[0054] The effect of introducing an obstruction in the pipe can be seen to greatly increase the magnitude of the acoustic output from the sensor. FIG. 3 shows an example output of magnitude (arbitrary units) as a function of time over a 500 ms sample for liquid flow in the absence of an obstruction. FIG. 4 shows a corresponding output with an obstruction, in this case on the form of the obstruction shown in FIG. 2b, i.e. with three holes passing along an otherwise solid insert.

[0055] For acoustic sampling, multiple samples were taken, each of a length of 500 ms, a 16 bit resolution and an amplitude of maximum ±1 V. The sampling number was set to 600 kS to ensure that the sampling frequency is at least twice the resonance frequency of the sensor. The choice of 500 ms was chosen as the time required to obtain stable Fast Fourier Transform (FFT) spectra over multiple samples. Three different types of liquids were selected for acoustic measurements, a summary of which is shown in Table 1 below. Distilled water was chosen as an example Newtonian liquid, the addition of glycerol to which changes the consistency but not the flow index or yield shear stress. Solutions of carboxymethylcellulose (MW 70,000) and Carbopol (Lubrizol 940 Non Food Grade) were used as examples of liquids having power law and Herschel-Bulkley rheological properties. A liquid exhibiting power law behaviour will have a zero yield shear stress, while a liquid exhibiting Herschel-Bulkley behaviour will have a yield shear stress. Both types of liquids exhibited shear thinning behaviour, i.e. with a flow index of less than 1. To determine the rheological properties of each liquid, flow curves were obtained and fitted to constitutive models using a Discovery HR-1 rheometer (TA Instruments, USA). The rheometer was equipped with a 60 mm 20 cone-and-plate-geometry and linked to TRIOS software (TA Instruments, USA).

TABLE-US-00001 TABLE 1 Summary of example liquid characteristics. Identifier Constituents τ.sub.y [Pa] κ [Pa s] n Newtonian 1 distilled water 0 0.001 1 (FIG. 5) Newtonian II 70% glycerol, 30% distilled 0 0.03 1 (FIG. 6) water Power Law 1 0.1% 0 0.05 0.76 (FIG. 7) carboxymethylcellulose in distilled water Power Law II 0.2% 0 0.10 0.77 (FIG. 8) carboxymethylcellulose in distilled water Power Law III 0.3% 0 0.16 0.74 (FIG. 9) carboxymethylcellulose in distilled water Herschel-Bulkley 0.10% Carbopol in distilled 0.06 0.02 0.76 I (FIG. 10) water, pH 4.5 Herschel-Bulkley 0.15% Carbopol in distilled 1.62 0.78 0.32 II FIG. 11) water, pH 4.5 Herschel-Bulkley 0.20% Carbopol in distilled 8.24 0.66 0.28 III (FIG. 12) water, pH 4.5

[0056] Frequency spectra from each type of liquid were obtained, examples of which are shown in FIGS. 5 to 12. Each spectrum shows the magnitude, in arbitrary values, of variations in acoustic intensity as a function of frequency.

[0057] Comparing the spectrum for distilled water (FIG. 5) with that for a 70:30 mixture of glycerol and water (FIG. 6), a difference can be seen predominantly in the presence of a peak 501 at around 50 kHz for water, which is largely absent for the 70:30 mixture. For the liquids exhibiting power law behaviour, the results of which are shown in FIGS. 7, 8 and 9, a peak 701 at around 80 kHz is present, the magnitude of which varies with concentration. The presence of other peaks relative to this also varies with concentration. For the liquids exhibiting Herschel-Bulkley behaviour, the frequency spectra shown in FIGS. 10, 11 and 12 show different features, the main difference with the other spectra being a much reduced low frequency peak 1001 relative to the other peaks. Higher frequency components are reduced as the concentration of Carbopol increases, which corresponds with an increase in yield shear stress.

[0058] To determine whether such frequency spectra could be used to identify the rheological properties of a particular liquid, comparisons between unknown and known spectra were made using a machine learning algorithm employing supervised machine learning. In a first step, the spectra were band limited to above 4 kHz, as any signals below this were considered to be environmental noise. Any positive and negative infinite values, i.e. those out of the amplitude range of the measurement equipment, were filtered and replaced by ±1. For each spectrum, the frequency resolution was reduced to 5,000 selected frequencies, and for each selected frequency a relative variance was determined. The relative variance was chosen over a simple variance because in this way the absolute values have been weighted on the mean values. If only absolute values were taken this would have neglected small values of magnitude, even if their relative change was high. Finally, for each sample the 5,000 FFT values with the largest relative variance were selected, resulting in a standardised spectrum suitable for comparison.

[0059] Once the frequency domain matrices were scaled to make them comparable to each other, they were divided into three matrices, representing Training (60%), Optimisation (20%) and Model Validation (20%). Machine learning algorithms were implemented using MATLAB (MathWorks).

[0060] FIG. 13 shows the results of using a variety of standard machine learning algorithms to match frequency spectra using principle component analysis (PCA) with 15 components, showing accuracy as a function of the number of features. It can be seen that increasing the number of features above 5000 does not assist in improving accuracy, and that although there is a range of accuracy between different algorithms there are many algorithms that are suitable with an accuracy of over 85%. The SVM (support vector machine) quadratic algorithm delivered the highest level of accuracy overall.

[0061] An advantage of using PCA in the frequency domain is to choose a set of weights by exploiting the cross-correlations between the signals at particular cycles. For example, the behaviour of the variables under study can be different in the short, medium and long run. Using PCA in the frequency domain thereby allows weights to be chosen depending on the frequency. The difference between PCA in the time domain and frequency domain can be understood in terms of how the eigenvalues are computed. In the time-domain, the correlation matrix is used. In the frequency-domain, the fast Fourier transform of the correlation matrix or the spectral density matrix is used to compute the eigenvalues. However, the disadvantage using this method is that the original time-spectrum cannot be recovered, although this is not of particular importance for application of the invention, given that the aim is to match spectra to identify rheological properties.

[0062] With the different types of liquids as described above, prediction accuracies of generally 95% or greater was possible, indicating that an unknown liquid could be identified with high certainty if a spectrum of a liquid having similar rheological properties has been stored.

[0063] FIG. 14 illustrates an example flow diagram of a method of identifying a rheological property of a liquid flowing in a pipe, which may for example be implemented in a continuous way as part of a process measurement system. In a first step 1401, an acoustic signal is detected, the acoustic signal being generated by the liquid flowing in the pipe and detected using a sensor attached to a rod extending from a wall of the pipe into the liquid. In a second step 1402, the acoustic signal is sampled to provide a sampled acoustic signal. This acoustic signal is then transformed into the frequency domain (step 1403) to generate a sampled frequency spectrum. The sampled frequency spectrum is then correlated with a stored frequency spectrum from a database of stored frequency spectra of liquids having predetermined rheological properties (step 1404).

[0064] A rheological property can then be identified of the liquid flowing in the pipe based on the stored frequency spectrum (1405). The method may be performed continuously as part of an industrial process measurement system to continuously monitor the rheological behaviour of a liquid flowing through a part of the industrial process. A change in rheological behaviour can thereby be automatically identified and, if necessary, notified or otherwise monitored and recorded over time.

[0065] FIGS. 15a, 15b and 15c illustrate alternative configurations to the arrangement of FIG. 1, in which the cross-sectional shape of the pipe 1501a-c varies along its length to provide an equivalent feature to the obstruction 106 of FIG. 1 so as to enhance the acoustic signal received by the acoustic sensor 105. The pipe 1501a, b may for example broaden in diameter, as in FIGS. 15a and 15b, or narrow in diameter, as in FIG. 15c, in a region encompassing the acoustic sensor 105. In a general aspect, an internal cross-section of the pipe 1501a-c may vary upstream and/or downstream of the acoustic sensor 105, for example such that the internal cross-section broadens or narrows upstream of the acoustic sensor 105 and narrows or broadens downstream of the acoustic sensor 105.

[0066] FIG. 16 illustrates a detailed sectional view of an alternative example of an acoustic sensor 105 mounted to a pipe 1601. In this example the sensor 105 is secured to a rod or plug 1604 that is screwed into the pipe wall 1605. The sensor 105 is attached to the plug 1604 with an acoustic coupling 1606 such as a silicone cushion or acoustic glue. The sensor 105 is held in place against the plug 1604 by a strap 1607 extending around the sensor 105 and the outer diameter of the pipe 1601. Other features of the apparatus and method of sensing may be similar to the other embodiments described herein.

[0067] The methods and apparatus described herein may be used as part of an in-line rheological measurement system to monitor the rheology of a liquid within an industrial process. FIG. 17 illustrates schematically a system 1700 incorporating an apparatus 1701 for measuring the rheological properties of liquid in a mixing tank 1702. The liquid in the tank 1702 flows through a measurement loop 1703 comprising a pump 1704, a flow meter 1705 and a measurement apparatus 1701. The apparatus 1701 may be in the form as described elsewhere herein, i.e. incorporating a sensor attached to a section of pipe and a computer 1706.

[0068] As the liquid in the mixing tank 1702 is processed, for example by shear mixing and addition of ingredients, the rheology of the liquid will change. The apparatus 1701 is configured to perform a series of measurements on the liquid flowing through the measurement loop 1703 and determine when the rheology has changed. This can be used to determine when to transition between steps in a manufacturing process. As an example, a manufacturing process for a formulated liquid personal care product was monitored over a series of processing stages involving emulsification followed by additions of water and other ingredients, with a final high shear mixing stage. This process was divided up into 14 classes, as shown in Table 2 below. Each class is associated with a difference in rheological properties. A machine learning algorithm was trained over the processing stages and the training data was then used to predict each stage from other unknown data.

TABLE-US-00002 TABLE 2 Classification of stages during manufacturing of an example liquid product. Class Number Description Class 1 5 Minutes into Emulsification Class 2 10 Minutes into Emulsification Class 3 15 Minutes into Emulsification Class 4 20 Minutes into Emulsification Class 5 25 Minutes into Emulsification Class 6 30 Minutes into Emulsification Class 7 During Addition of Water Class 8 During Addition of Water Class 9 Total Addition of Water Class 10 5 Minutes After Water Addition Class 11 5 Minutes After Addition of Ingredient 1 Class 12 5 Minutes After Addition of Ingredient 2 Class 13 5 Minutes After Addition of Ingredient 3 Class 14 After High Shearing for 5 Minutes

[0069] FIG. 18 illustrates results in the form of a confusion matrix, where the predicted class is plotted against the true class for each stage of the manufacturing process for the trained model. In this example a linear support vector machine (SVM) was used, which showed the highest accuracy at around 99%, with other models such as quadratic SVM showing a slightly lower accuracy at 97.6%. Further accuracy improvements could be obtained through using more iterations.

[0070] FIG. 19 shows a confusion matrix of true class against predicted class for unseen test data, demonstrating an overall accuracy of 94.4%. The largest errors in this case involve distinguishing between different stages of emulsification between 20 and 30 minutes, suggesting that the difference in rheological properties of the liquid over these stages is small. Similarly, a reduction to 90% for class 8 indicates that the change in rheology after addition of water is also small.

[0071] Based on the above example, a trained machine learning algorithm may be used to monitor acoustic signals from an acoustic sensor during a production process to determine when a particular manufacturing process stage of a liquid is complete. In a general aspect therefore, a method of monitoring a manufacturing process of a liquid may involve performing a mixing process on the liquid, passing the liquid through a pipe and performing a method as described herein to identify a stage of the manufacturing process. The mixing process may for example include addition of an ingredient to the liquid and mixing of the liquid, for example by shearing the liquid.

[0072] Acoustic signals measured and processed according to the above examples will tend to contain large amounts of measurement data, typically in the region of thousands to hundreds of thousands of data points per measurement. In particular for online monitoring of rheological measurements it can be challenging to process the measurement data quickly enough. In alternative examples, the measurement data may be simplified prior to a determination of rheological properties without losing the key information provided by the raw signal. An example illustration of a simplified series of measurements is shown in FIG. 20. In this, acoustic signals are measured and processed for stages 1 to 14 of the process as described above. For each stage, 40 seconds of acoustic signals were recorded and processed, the processing taking around 1 minute in each case. The frequency spectrum for each recording was divided into 10 sections, identified by parameters P1 to P10 in FIG. 20. The relative magnitude of each section is plotted in FIG. 20, all relative to the maximum cumulative magnitude of stage 14. The relative contributions of each part of the frequency spectrum represented by the magnitudes of parameters P1 to P10 can be matched with a stored set of measurements representing different rheological properties. In a process such as the one described above, the changes in relative magnitudes of the sections of the frequency spectrum can be used to indicate a transition from one stage to the next. Each set of parameters may be considered to represent a rheological factor for a particular liquid having defined rheological properties.

[0073] The sections of the frequency spectrum for each measurement can be chosen based on the expected key portions of the frequency spectrum and may for example be selected to avoid known regions of unrepresentative noise or unchanging background and/or to select portions that are particularly representative of certain rheological properties. A sampled frequency spectrum may be divided into a plurality of sections, for example 10 or more sections, and an amplitude of each section determined. The resulting set of parameters, which may be arranged in the form of a matrix, is then correlated with a stored set of parameters to identify a rheological property of the liquid. Typical numbers of parameters may be 10 or 20, or in a general aspect may be between around 10 and around 100. A smaller number of parameters will result in faster processing but reduced accuracy, while a larger number of parameters will result in longer processing but greater accuracy. It has been found that 10 parameters is generally sufficient to identify the required rheological properties in the examples described, although more may be needed in other cases where finer distinctions between rheological properties may be required.

[0074] Another factor in determining the accuracy and processing speed is the length of time each acoustic signal is sampled. In the example shown in FIG. 20 samples were taken over 40 seconds. Shorter samples may be sufficient. Samples as short as 0.5 seconds were also taken, which required only 5 seconds of processing, and provided sufficiently accurate representations to determine rheological differences. In a general aspect therefore, each sampled acoustic signal may represent a sampling time of 0.5 seconds or more of the acoustic signal generated by the liquid flowing in the pipe. The sampling time may for example be between around 0.5 second and 60 seconds, with longer sampling times tending to produce more accurate data but with longer processing times.

[0075] Other embodiments are intentionally within the scope of the invention as defined by the appended claims.

REFERENCES

[0076] Rahman M, Hikansson U and Wiklund J 2015 In-line rheological measurements of cement grouts: Effects of water/cement ratio and hydration Tunn. Undergr. Sp. Technol. 45 34-42 [0077] Kotzé R, Ricci S, Birkhofer B and Wiklund J 2016 Performance tests of a new non-invasive sensor unit and ultrasound electronics Flow Meas. Instrum. 48 104-11 [0078] Khulief Y A and Khalifa A 2012 On the In-Pipe Measurements of Acoustic Signature of Leaks in Water Pipelines Volume 12: Vibration, Acoustics and Wave Propagation (ASME) p 429 [0079] Li S, Song Y and Zhou G 2018 Leak detection of water distribution pipeline subject to failure of socket joint based on acoustic emission and pattern recognition Measurement 115 39-44 [0080] Hou R, Hunt A and Williams R. 1999 Acoustic monitoring of pipeline flows: particulate slurries Powder Technol. 106 30-6 [0081] O'Keefe C V, Maron R, Felix J, van der Spek A M and Rothman P 2010 Non-invasive passive array technology for improved flow measurements of slurries and entrained air The 4th International Platinum Conference: Platinum in transition ‘Boom or Bust’ ed CiDRA Holdings (Johannesburg: The Southern African Institute of Mining and Metallurgy) pp 21-30 [0082] Finfer D, Parker T R, Mahue V, Amir M, Farhadiroushan M and Shatalin S 2015 Non-intrusive Multiple Zone Distributed Acoustic Sensor Flow Metering SPE Annual Technical Conference and Exhibition (Houston: Society of Petroleum Engineers) pp 1-9