WO2024194641A1

WO2024194641A1 - Ultrasonic non-destructive testing

Info

Publication number: WO2024194641A1
Application number: PCT/GB2024/050760
Authority: WO
Inventors: Jonathan Mark Allin; Attila GAJDACSI; Jacob Owen DAVIES
Original assignee: Permasense Limited
Priority date: 2023-03-22
Filing date: 2024-03-21
Publication date: 2024-09-26
Also published as: GB2628384A

Abstract

A method for ultrasonic non-destructive testing of an object under test comprises receiving sets of measurement data representing respective sets of output ultrasonic vibrations detected at a surface of the object in response to transmission of respective pulses of input ultrasonic vibrations into the surface of the object. The sets of measurement data include at least two sets of measurement data representing output ultrasonic vibrations detected in response to transmission of pulses of input ultrasonic vibrations having different frequency content. The method comprises processing the plurality of sets of measurement data to determine at least one property of the object.

Description

ULTRASONIC NON-DESTRUCTIVE TESTING

The present technique relates to the field of ultrasonic non-destructive testing.

Ultrasonic non-destructive testing may be used to detect a property of an object under test. An ultrasonic detector located at the surface of a solid object can detect output vibrations received from within the object in response to transmission of a pulse of input vibrations into the surface of the object, so that properties of the object which may not be visible at the surface can be detected. For example, where the object is a wall of a pipe, vessel, tank or other conduit or container, the ultrasonic non-destructive testing can be used to detect changes in the inner surface of the conduit/container, such as monitoring changes in the thickness or surface roughness of the wall, for example caused by corrosion or erosion. As an example, using monitoring techniques to track the corrosion or erosion of the inner surface of a pipe in a refinery may permit the safe refining of oil which would otherwise be regarded as too difficult due to the way in which it corrodes or erodes the pipes of the refinery.

At least some examples provide a method for reducing influence of frequency dependent factors on a determination of at least one property of an object under test using ultrasonic nondestructive testing, the method comprising: receiving sets of measurement data representing respective sets of output ultrasonic vibrations detected at a surface of the object in response to transmission of respective pulses of input ultrasonic vibrations into the surface of the object, the plurality of sets of measurement data including at least two sets of measurement data representing output ultrasonic vibrations detected in response to transmission of pulses of input ultrasonic vibrations having different frequency content; and processing sets of measurement data to determine at least one property of the object, where the processing comprises reducing influence on determination of the at least one property of the object of at least one set of measurement data having a feature indicating insufficient confidence in that set of measurement data providing a correct estimate for the at least one property of the object.

At least some examples provide a computer program comprising instructions which, when executed by a computer, control the computer to perform the method described above. The computer program may be stored on a storage medium. The storage medium may be a transitory or non-transitory storage medium.

At least some examples provide a system for performing ultrasonic non-destructive testing of an object under test, the system comprising: at least one measurement device for coupling to the object under test, to capture a set of measurement data representing output ultrasonic vibrations detected at a surface of the object in response to transmission of a pulse of input ultrasonic vibrations into the surface of the object; and processing circuitry to process at least two sets of measurement data captured by the at least one measurement device in response to transmission of pulses of input ultrasonic vibrations having different frequency content, to reduce influence of frequency dependent factors on a determination of at least one property of the object, where the processing circuitry is configured to reduce influence on determination of the at least one property of the object of at least one set of measurement data having a feature indicating insufficient confidence in that set of measurement data providing a correct estimate for the at least one property of the object.

Further aspects, features and advantages of the present technique will be apparent from the following description of examples, which is to be read in conjunction with the accompanying drawings, in which:

Figure 1 schematically illustrates an example of a system for performing ultrasonic nondestructive testing of an object under test, where in this particular example the object is a pipe wall;

Figure 2 illustrates propagation of pulses of ultrasonic vibrations through the wall;

Figure 3 illustrates the receipt of a reference pulse, and a detected pulse reflected from a distal surface of the wall, following transmission of an input pulse of ultrasonic vibrations transmitted into a proximal surface of the wall;

Figure 4 illustrate confounding factors which may make reliable measurement of pipe wall thickness more complex;

Figure 5 illustrates an example of build-up of scale on inner surface of the pipe wall;

Figure 6 illustrates an example of pulses of different frequency content;

Figure 7 illustrates a waveform with backwall echo which has been distorted by a combination of a rough internal surface or layers of scale on the internal surface,;

Figure 8 illustrates a waveform where the backwall echo has low distortion;

Figure 9 illustrates simulation results illustrating errors in thickness measurement for different thicknesses of scale layer and different transducer frequencies where the thickness is determined using a time of flight methodology taking the timing from the time where the amplitude of the echo envelope is maximum.;

Figure 10 illustrates a method of capturing sets of measurement data based on ultrasonic non-destructive testing using pulses of input ultrasonic vibrations with different frequency content;

Figure 1 1 illustrates a method for ultrasonic non-destructive testing comprising processing the sets of measurement data captured based on pulses of input ultrasonic vibrations with different frequency content, to determine at least one property of the object under test;

Figure 12 illustrates an example where a given property of the object is determined based on an aggregate function of two or more estimated property values calculated from the respective sets of measurement data;

Figure 13 illustrates an example of selecting between the sets of measurement data based on a comparison of the sets of measurement data;

Figure 14 illustrates an example where at least one set of measurement data that gives a result inconsistent with at least one previous measurement of a given property is excluded from the determination of at least one property of the object; Figure 15 illustrates an example where a comparison with at least one reference set of measurement data is used to select which set of measurement data to use for determining at least one property of the object; and

Figure 16 illustrates an example where a similarity metric determined between a set of measurement data and a reference set of measurement data is used to select which set of measurement data to use for determining at least one property of the object.

Ultrasonic non-destructive testing can be used to determine properties of an object which may not be visible from the surface. A pulse of input ultrasonic vibrations is transmitted into the surface of the object, and a detector (which can be the same device as the transducer which generated the pulse of input ultrasonic vibrations, or could be separate from the transducer) detects corresponding output ultrasonic vibrations at the surface. A set of measurement data is generated based on the output ultrasonic vibrations detected at surface. The measurement data can be used to determine a property of the object such as the thickness of a wall, whether a crack is present in the object, or a change in surface roughness of a remote surface of the object hidden from visual inspection.

However, the inventors have recognised that a number of confounding factors may sometimes make reliable measurement of a given property of the object more difficult. For example, a solid object may have internal structures such as voids (empty spaces within the solid body) or grains (a region of the object with a different crystal structure to other regions) which may cause internal reflections of the pulse of input ultrasonic vibrations which do not represent the property being tested (such as the distance to, or condition of, a remote surface on the other side of a wall from the surface at which the input ultrasonic vibrations are transmitted into the object). Also, there may be additional layers at either the proximal surface at which the input ultrasonic vibrations are input to the object or at the distal surface from which the vibrations are reflected, which may make measurements of properties of the distal surface unreliable. For example such layers may include paint layers or other coatings applied to the object, a scale layer formed based on material deposited from, or formed in a reaction with, the product contained within the object, and/or a corroded surface layer caused by corrosion of the material forming the body of the object. Therefore, it has been found that sometimes a measurement taken based on the ultrasonic non-destructive testing can be unreliable. This does not affect all measurements, but it can be difficult to determine whether, for a given set of measurement data, the property determined with that set of measurement data is reliable or not.

The inventors have recognised that the influence of such confounding factors may be highly frequency dependent, so that a measurement made at one frequency may be strongly affected by these confounding factors while a measurement at another frequency may be much less affected and give a more reliable result. However, prediction of which frequency will be the best to use for a particular measurement may be extremely difficult because the distortion in output vibrations caused by reflections from features such as the internal voids or grains or scale/corrosion/paint layers may vary depending on the position and thickness of such features and on other properties such as the temperature at which the measurement is taken. For example, as the thickness of the scale layer on the inner surface of the container or conduit grows, there is a change in the frequencies at which reflections from the scale layer distort the main reflections from the container/conduit wall. As the thickness of the scale layer will not be known in advance because the scale layer is hidden inside the container or conduit, it is not possible to manually inspect the object to select the correct frequency to use. In any case, even for a given thickness of scale layer, the frequency-dependent effects of interference/distortion may change between measurements taken at different temperatures, because speed of sound in a solid may vary with temperature.

In the examples discussed below, at least two sets of measurement data are received, representing output ultrasonic vibrations detected in response to transmission of pulses of input ultrasonic vibrations having different frequency content. The sets of measurement data are processed to determine at least one property of the object. As explained further below, a variety of techniques may be used to select which measurement to use or to combine the measurements to give an output result for the determine property of the object, but in general, by using two or more measurements taken based on pulses of input ultrasonic vibrations with different frequency content, then even if one measurement gives an unreliable result because the measured waveform was distorted due to undesired reflections from features such as those discussed above, one or more of the other measurements can be less affected and can give a better result. Hence, more reliable detection of the property can be performed.

The pulses of input ultrasonic vibrations used for the at least two sets of measurement data have at least one of: different centre frequencies; different bandwidths; and a different burst pattern of frequency content (e.g. for the latter case, the pulse may be a multi-burst pulse comprising a number of bursts of ultrasonic vibrations in different parts of the frequency range, and the burst pattern for the different pulses may vary in terms of the number of bursts and/or the frequency content of each burst). Hence, with all of these examples, it is possible to introduce variation in frequency content covered by respective pulses used for generating the at least two sets of measurement data, to increase the likelihood that at least one of the measurements can give a good result for at least one property of the object.

The pulses of input ultrasonic vibrations used for the measurement data may span a frequency range from a minimum frequency Fmin to a maximum frequency Fmax where Fmax/Fmin is greater than or equal to 1 .3. By covering a relatively wide range of frequencies, this increases the likelihood that at least one of the pulses of input ultrasonic vibrations will yield a corresponding set of measurement data which can provide a reliable measurement for at least one property of the object.

The pulse which includes vibrations at the minimum frequency Fmin may be a different pulse to the pulse which includes vibrations at the maximum frequency Fmax. Probing the lowest frequency and the highest frequency using different pulses can give more effective results than a single pulse with a very wide bandwidth spanning both the minimum and maximum frequencies.

Other examples may use pulses of input ultrasonic vibrations collectively spanning a wider frequency range, for example with the ratio Fmax/Fmin having a value of at least 2, or at least 2.2, or at least 2.4, or at least 2.6, or at least 2.8, or at least 3, or at least 3.2, or at least 3.4, or at least 3.6, or at least 3.8, or at least 4. By spanning over an octave range between the minimum and maximum frequency spanned by the two or more pulses as a whole, there can be increased likelihood that at least one of the measurements gives a reliable result.

In some examples, each pulse spans a range from frequency F1 to frequency F2, where F1 is the lowest frequency at which the amplitude is at least 10% of the amplitude of the peak in the frequency domain that has the greatest amplitude, and F2 is the highest frequency at which the amplitude is at least 10% of the amplitude of the peak in the frequency domain that has the greatest amplitude. In some examples, the minimum frequency Fmin mentioned above corresponds to the minimum value of F1 for any of the two or more peaks used to generate the sets of measurement data, and the value Fmax mentioned above corresponds to the maximum value of F2 for any of the two or more peaks used to generate the sets of measurement data.

In other examples, the range from Fmin to Fmax may be defined in terms of the centre frequencies of the pulses, so that the pulses of input ultrasonic vibrations may have centre frequencies which span a frequency range from Fmin to Fmax greater than or equal to 1 .3, or greater than any of the other thresholds listed above.

It is not essential for every set of measurement data to be based on a pulse of input ultrasonic vibrations with different frequency content from the pulses used for any other set of measurement data. It is possible to capture two or more sets of measurement data based on pulses with the same frequency content (e.g. to make repeated readings using the same form of input pulse so that an average or other aggregate function can be applied to those readings to obtain a result). However, even if there is some repeat usage of the same form of input pulse, at least two of the sets of measurement data are based on pulses of input ultrasonic vibrations with different frequency content, to give variety in the frequencies used for different sets of measurement data so that there is an increased probability that at least one of these measurements is less affected by undesired reflections from features other than the main feature being probed.

In some examples, only two sets of measurement data may be captured in response to transmission of pulses of input ultrasonic vibrations having different centre frequencies.

However, in some examples, at least three sets of measurement data may be captured in response to transmission of pulses of input ultrasonic vibrations having different centre frequencies. Analysis of simulation results has shown that the effects of distortion caused by confounding features such as a scale layer may be such that, as the frequency of input ultrasonic vibrations is increased for a scale layer of a given thickness, there may be a region at lower frequency where the measurements are more distorted by reflections from the scale layer of a given thickness, followed by a region at intermediate frequency where measurements are more reliable, followed by a region at higher frequency where measurements again become more distorted. Taking at least three measurements based on pulses with different frequency content can reduce the chance that all the measurements correspond to regions highly affected by distortion (e.g. because one of two measurements happened to be in the lower-frequency distorted range and the other of the two measurements occurred in the higher-frequency distorted range). In practice, more than three measurements could be performed for the same test object using input pulses with different frequency-content. Taking a larger number measurements can also be helpful to allow techniques such as majority voting to be used to exclude outlying sets of measurement data which give a very different result from the majority of other sets.

The technique of taking multiple measurements based on input pulses with different frequency content can be particularly useful when the object is a multi-layer structure comprising two or more layers, where the property being detected may be a property of at least one of those layers. For example, the layers may include a core layer which is the principal layer being inspected (e.g. the core layer may be the wall of a conduit or container). The core layer could be metallic, plastic or ceramic, for example. The layers may also include at least one additional layer, such as a paint layer, corroded surface layer, scale layer and/or lining layer (e.g. a continuous weld lining or an explosion bonded lining), which may generate additional reflections causing distortion in the echoes of the waveform of the output ultrasonic vibrations which represents reflections from the metallic core layer. It is also possible that there may be internal voids or grains within the core layer which may confuse the measurement taken. The effects of the additional layers and internal structures such as voids or grains made be frequency dependent and so taking multiple measurements based on pulses of different frequency content allows for more reliable measurements of at least one property of the object by enabling selection of a measurement set which is less distorted, even if it is not known in advance which frequency will be best for a given scenario.

In one particular example, the object may be a wall. For example, the wall may be a wall of a container or conduit. More particularly, the wall may be a wall of the pipe. The technique can be particularly useful in use cases where the container or conduit is intended to carry an abrasive or corrosive substance, such as crude oil or other products of the oil refinery industry. If the fluid or other substance contained in the container or conduit is abrasive or corrosive, deterioration of an inner surface of the wall may be expected and it can be difficult for the condition of the inner surface of the wall to be detected by manual inspection without emptying the container or conduit of the product, which would disrupt processing of the product. Inspecting the condition of the inner surface by a non-destructive, non-invasive method using ultrasonic vibrations can therefore be very useful in such scenarios. However, such products may often cause accumulation of scale layers or presence of surface roughness on the surface due to abrasion or corrosion, which may make measurements of properties based on a single pulse having a given frequency content potentially unreliable in some conditions. The use of multiple frequencies can therefore greatly help to increase the reliability of the results determined based on the ultrasonic vibrations.

Various properties of the test object can be determined using the method. For example, the at least one property may comprise at least one of: a distance to a feature of the object (e.g. the feature could be a crack in the object or a distal surface of the object), a thickness of at least one layer of the object; and a measure of surface roughness of a distal surface of the object (the distal surface is a surface from which the pulses of input ultrasonic vibrations are reflected after transmission into a proximal surface of the object).

The sets of measurement data (captured based on the at least two pulses of input ultrasonic vibrations having different frequency content) may be captured at different times. A first pulse of input ultrasonic vibrations may be transmitted into the surface of the test object and a first set of measurement data may be captured representing the corresponding output ultrasonic vibrations detected at the surface of the object in response to transmission of the first pulse. Subsequently, at a time when the echoes caused by multiple reflections caused by the input of the first pulse have attenuated (died down) to such an extent that they would no longer be detectable, a second pulse of input ultrasonic vibrations having different frequency content to the first pulse may be transmitted into the surface of the test object, and a second set of measurement data may be captured representing the corresponding output ultrasonic vibrations detected at the surface of the object in response to transmission of the second pulse. Similarly, third, fourth or further sets of measurement data may be captured based on further pulses of input ultrasonic vibrations being transmitted. Each of these sets of measurement data may be independent of the previous set, captured with sufficient time in between the transmission of the respective pulses of input vibrations so that reflections caused by a pulse of input ultrasonic vibrations transmitted for one set of vibrations are not detectable within the next set of measurement data captured in response to transmission of a subsequent pulse of input ultrasonic vibrations.

A variety of techniques can be used to process the sets of measurement data captured based on at least two pulses of input ultrasonic vibrations having different frequency content. Hence, the principle of determining the property of the object based on measurements taken with input pulses of different frequency content can be of general application, irrespective of the particular way in which those sets of measurement data are processed. No single processing algorithm is essential for determining the property based on the measurements taken at the different frequencies. There are a wide range of alternative processing techniques which can be applied. Some examples are discussed below.

In some examples, the processing comprises determining a given property of the object as a function of two or more feature values for the given property, the two or more feature values derived from two or more of the at least two sets of measurement data respectively. The feature values could be estimates for the property itself (e.g. a wall thickness, or indication of surface roughness), or could be waveform features of the output ultrasonic vibrations derived from the measurement data (e.g. the position or width, in either the time domain or the frequency domain, of a peak in the ultrasonic vibrations, for example). Hence, multiple values for a given feature may be derived from the measurements made with pulses of different frequency content, and then processed to select the final result value determined for the given property.

For example, the processing may comprise determining the given property of the object as an aggregate function of the two or more feature values. An aggregate function is a function which receives two or more input values and which outputs a single value determined as a function of the input values. For example, the aggregate function could be any of:

• a mean of the two or more feature values. This function could be particularly useful if there is also a step of filtering out outliers or nonsensical/inconsistent results, such excluding an estimate based on comparisons with results from other measurements, or based on a comparison with previous historic measurements of the object, as discussed further below.

• a median of the two or more feature values. This can be useful as a way of implementing a “majority voting” scheme, where selecting the median result from among the estimates of the given property has the effect of excluding outlying estimates for the given property which are most likely to be the unreliable results based on distorted output ultrasonic vibrations.

• a mode (most common value) of the two or more feature values. This can select the most likely result based on the one that occurs most often (if the feature value is a continuous variable then the variable may be “binned” into one of a number of ranges to give discrete options for which the relative frequency of occurrence can be analysed in order to determine the mode value).

• a minimum of the two or more feature values. One might find it surprising that selecting the minimum of the two or more feature values may give a reliable result, as it might be imagined that this would cause an unreliable outlier to be selected. However, for some use cases the minimum value among the feature values may be considered the most reliable. For example, if the feature value is the thickness of a wall and the confounding factor affecting measurements is the accumulation of a scale layer or other layer on the distal surface of the wall, in practice the additional layer will tend to cause additional reflections at a later time than the reflections from the wall surface itself, so that any distortion caused by these reflections will tend to increase the measured time of arrival of the output pulse of vibrations at the detector (rather than decrease the measured time of arrival). Therefore, for thickness measurements it may be that the earliest arriving pulse of output vibrations seen for any of the measurements taken at different frequencies may be most reliable as one may expect it has not been subject to as much distortion based on the later arriving reflections from the scale/corrosion layer. Therefore, for thickness measurements, determining the final result as the minimum or the two or more feature values can be effective.

• a maximum of the two or more feature values. This may be useful, for example, where the feature values are waveform features, e.g. height-to-width ratio of a peak in the output vibration waveform, where the peak with the highest height-to-width ratio may be the one that is least distorted and so the best candidate for a reliable measurement of a property of the object.

The determination of a feature value based on a given set of measurement data may also comprise applying a discrimination function, to exclude from determination of the feature value an echo of the output ultrasonic vibrations which fails to meet a discrimination criterion. The discrimination criterion could be a threshold criterion, for example, which excludes echoes of a size less than a threshold. The aggregate function mentioned above can be applied to the resulting waveforms of the measurement data after the discrimination function has been applied. Applying a discrimination function can be useful to deal with the possibility of internal voids or grains, or pits in a surface caused by corrosion or abrasion, which may cause some early reflections compared to the reflections from the feature of interest (e.g. a distal surface of a wall). The discrimination function may for example exclude weaker echoes of vibration from subsequent analysis. As voids, grains, or small pits in the wall surface may cause relatively weak early reflections compared to the larger peak caused by the reflection from the main wall surface, the reflections from such confounding features may be more likely to be excluded by the discrimination function, so that the result of applying an aggregate function (such as the minimum function) to the results of applying the discrimination function to each set of measurement data can give a more accurate measurement of the given property.

In some examples, the processing of the sets of measurement data may comprise excluding, from determination of the at least one property of the object, at least one set of measurement data determined to have a feature indicating insufficient confidence in that set of measurement data providing a correct estimate for the at least one property of the object. The reason for insufficient confidence could vary, for example this could be due to the set of measurement data being an outlier that is very different from other measurements, or because it has other properties (not derived from a comparison with other measurements) that indicate it is statistically unlikely to be a reliable measurement, such as an indication of how distorted an echo is, an indication of whether there is a repeating pattern of such echoes, or an indication of how much the degree of distortion changes with successive echoes.

In some examples, the processing may comprise selecting at least one of the sets of measurement data to use to determine a given property of the object, based on a comparison of the sets of measurement data. Hence, the sets of measurement data may be compared with each other to identify at least one set of measurement data which might be unreliable because it differs from others. Hence, the method may comprise excluding, from determination of the property of the object, at least one outlying set of measurement data which is dissimilar to a majority of the other sets of measurement data. For example, a similarity metric (such as maximum value of a cross-correlation or cross-covariance of a pair of sets of measurement data) may be detected for respective pairs of sets of measurement data and used to identify a set of measurement data which differs from a majority of other sets of measurement data, so that the different set of measurement data can be excluded from calculation of a given property of the object.

In some examples, the processing may comprise excluding, from the determination of at least one property of the object, a set of measurement data which would yield a result for a given property of the object which is inconsistent with at least one previous measurement of the given property of the object, and determining the at least one property based on at least one other nonexcluded set of measurement data. For example, if historical measurements of the property of the object are available (e.g. wall thickness measurements from previous days), these may give a general impression of the condition of the wall. If a given set of measurement data gives a nonsensical result (e.g. that the wall has suddenly grown in thickness by a significant amount), that set of measurement data can be excluded from the sets of measurement data used to calculate the given property of the object, as it can be assumed that the excluded set was erroneous because of the influence of distortion based on the confounding features mentioned earlier. The previous measurement of the given property of the object may be a historical measurement made at an earlier time (e.g. an earlier hour, day, month or year) - i.e. not merely an earlier measurement in a series of measurement data being captured at different frequencies in the current instance of applying the testing. For example, if the testing method is performed at regular or irregular intervals separated by a number of hours, days or weeks, the previous measurement could be measurements made on one or more previous instances of applying the testing method.

In some examples, the processing comprises selecting at least one of the sets of measurement data to use to determine the property, based on a comparison of the sets of measurement data with at least one reference set of measurement data. Hence, rather than comparing the property of the object derived from the measurement data against previous measurements of the property, the measurement data can be compared with a reference set of measurement data, to identify whether a given set of measurement data (captured based on an input pulse of ultrasonic vibrations with a given frequency content) is significantly different from the expected pattern represented by the reference set of measurement data. The reference set may be a set of measurement data which is relatively little affected by distortion or other adverse influence of reflections from features such as scale, corrosion, voids or grains.

In some examples, the reference set of measurement data represents an ideal set of output vibrations or a simulated set of output vibrations. For example, a model pulse of output vibrations which is free of distortion may be used as the reference set of measurement data. In some examples, the reference set of measurement data may comprise at least one previously captured set of measurement data captured previously for the object under test in response to transmission of a pulse of output vibrations into the surface of the object. That previously captured set of measurement data could, for example, be captured at initial installation of the object (prior to being used to contain a potentially corrosive or abrasive product, say), so as to provide a baseline measurement before the confounding factors expected to arise later have occurred. Alternatively, the previously captured set of measurement data could be captured later in the lifetime of the object, but may have been manually inspected by a person to check that the previously captured set of measurement data is free of distortion and so can serve as a reliable reference set of measurement data.

For example, the at least one previously captured set of measurement data may comprise a set of measurement data captured at an earlier date. Again, as mentioned above, in practice the non-destructive testing method may be performed periodically at regular or irregular intervals, e.g. one or more times per day, or on certain days at intervals of time. Previous measurements may be representative of a pattern of output vibrations expected to be detected in response to the input vibrations (e.g. a pattern of output vibrations with a relatively undistorted main peak). If a latest set of measurement data deviates significantly from such previous sets of measurement data then it can be excluded from further processing.

When one or more sets of measurement data are excluded from calculation of the at least one property (e.g. because they have statistically low confidence in providing an accurate estimate of the property of the object, because the comparison with the reference set of measurement data shows a relatively low similarity with the reference set, or because it gives a result for a given property which is inconsistent with at least one previous measurement as discussed above), any remaining sets of measurement data can be used in different ways to determine at least one property. In some cases, an aggregate function (e.g. median, mean, mode, minimum and/or maximum) as discussed above can be applied to feature values derived from the remaining sets, and used to estimate the property of the object. Alternatively, a single one of the remaining sets of measurement data can be selected and used to determine the at least one property.

In some examples, the processing comprises determining, for each set of measurement data, a similarity metric based on a comparison with a reference set of measurement data; and selecting the at least one of the sets of measurement data based on the similarity metric determined for each set of measurement data. The similarity metric can be any indication of relative similarity between a set of measurement data and the reference set of measurement data. In some examples, the similarity metric can be obtained by at least one of: cross-correlation (sliding dot product) of a set of measurement data with the reference set of measurement data; and cross-covariance of a set of measurement data with the reference set of measurement data. For example, the similarity metric could be the maximum value of a waveform obtained by cross- correlation or cross-covariance of the set of measurement data and the reference set of measurement data. The selection of the at least one of the sets of measurement data may prioritise selection of a set of measurement data with a similarity metric indicating greater similarity to the at least one reference set of measurement data in preference to selection of a set of measurement data with a similarity metric indicating less similarity to the at least one reference set of measurement data. This tends to reduce the likelihood that a measurement set heavily distorted by reflections from artefacts is used to determine the property of the object.

In some examples, the ultrasonic non-destructive testing may comprise a calibration phase, when a test system is calibrated for use with the specific object under test, and an inspection phase, when the calibrated test system is used for monitoring of the condition of the object under test.

In some examples, the frequency dependent factors that may cause unreliable measurements may include a time-varying factor which is variable with operating conditions of the object under test and/or variable over the lifetime of the object under test. For example, if the frequency dependent factors include influence of a scale layer, the scale layer may grow over time, changing the frequencies at which unreliable measurements of the at least one property would be made. Also, some types of noise may be highly temperature-dependent, so the frequencies at which the unreliable measurements would occur may vary depending on current operating temperature.

Hence, in some examples, the inspection phase of ultrasonic non-destructive testing may comprise determination of the at least one property of the object based on processing of the sets of measurement data associated with the two or more pulses of ultrasonic vibrations having different frequency content. Hence, rather than merely using multiple pulses with different frequency content during the calibration phase when the test system is being set up but the object under test is not yet being monitored, two or more pulses with different frequency content can be used to make measurements of the at least one property even during the inspection phase. This can be helpful for use cases dealing with time-varying factors such as growth of scale layers with time or temperature-dependent effects.

The method of receiving and processing the sets of measurement data can be performed by a computer (e.g. a server) in response to a computer program comprising instructions which, when executed by the computer control the computer to perform the method as discussed above. In some examples, the receiving and processing steps could be performed on a computer which also instructs the capture of the sets of measurement data. In other examples, previously captured sets of measurement data (e.g. received from a storage device or via a network) may be received by the computer and processed to determine the property.

A system for performing ultrasonic non-destructive testing of an object under test may comprise at least one measurement device for coupling to the object under test, to capture a set of measurement data representing output ultrasonic vibrations detected at a surface of the object in response to transmission of a pulse of input ultrasonic vibrations into the surface of the object; and processing circuitry to process at least two sets of measurement data captured by the at least one measurement device in response to transmission of pulses of input ultrasonic vibrations having different frequency content, to determine at least one property of the object. The processing circuitry could be part of a measurement device, or separate from the measurement device(s). The processing circuitry could be at the same site as the measurement device measuring the test object, or could be at a different site remote from the measurement device.

Figure 1 schematically illustrates an example of a system 2 for ultrasonic non-destructive testing of an object. In this example, the object is a pipe wall and the testing is for monitoring pipe wall thickness, to detect thinning of the pipe wall due to corrosion or abrasion by the contents of the pipe (e.g. the contents could be a product of the oil refinery industry). It will be appreciated that the pipe is just one example of a test object and similar methods can be used for testing other objects. Wall thickness is just one example of a property which can be measured for the object using ultrasonic non-destructive testing. Other example properties which can be detected include the detection of presence of cracks (and if detected, the position of the crack), and detection of a measure of surface roughness. However, in the examples below the detection of thickness of a pipe wall serves as a specific example to illustrate the approach described above.

The system 2 comprises a number of sensors 4, 6, 8 each attached to a respective pipe 10, 12, 14. Each pipe has an outer surface corresponding to a proximal surface to which one of the sensors 4, 6, 8, is attached and an inner surface corresponding to a distal surface from which reflections of pulses of ultrasonic vibrations are detected. The pipe may carry a corrosive fluid or a mixed phase fluid which subjects the inner surface of the pipe to corrosion and/or erosion (e.g. sand within crude oil may erode the inner surface of pipe). In this particular example, each of the sensors 4, 6, 8 communicates wirelessly with a gateway 16 either directly or via a mesh network formed of the sensors. The gateway 16 in turn communicates with a server 18. However, in other examples, a wired data collection and transmission mechanism could be used. Also, it is possible for the processing of the sensor data to be performed locally within the sensor device itself, e.g. with a device which has an onboard data logger and display. Therefore, it will be appreciated that both the particular location at which the sensor data is processed and the way in which the data is transmitted from the sensor to the location at which it is processed can vary.

The sensors 4, 6, 8 illustrated in Figure 1 are waveguide sensors well suited to high temperature applications, but other sensor types are possible such as pulse echo mode sensors (same transducer sends and receives) that may be suited to lower temperature environments. Examples of transducers/sensors that can be used for transmitting and detecting input and output pulses of ultrasonic vibrations are described in WO 2007/051959 A1 and WO 2016/066997 A1 , the contents of which are incorporated herein by reference in their entirety.

At periodic intervals, such as daily or every 12 hours (or less if more frequent monitoring is required), or at more irregular intervals such as once a day or multiple times per week not necessarily at the same time each day, each of the sensors 4, 6, 8 may perform a test to capture data for determining the pipe wall thickness of the pipe 10, 12, 14 to which it is attached. This test may be performed by transmitting a pulse of input ultrasonic vibrations into a proximal surface of the pipe wall (either directly or indirectly via a wedge or coupling fluid) and then detecting reflected ultrasonic vibrations returned back to the proximal surface. The received vibrations may be sampled with a high rate analogue-to-digital converter and then wirelessly transmitted via the gateway 16 to the server 18. The server 18 may then perform signal processing upon these signals representing the received ultrasonic vibrations at the proximal surface in order to identify a propagation delay of the ultrasonic pulses through the pipe walls and accordingly the pipe wall thicknesses. As discussed in WO 2014/167285 A1 , the contents of which are incorporated herein by reference in their entirety, this signal processing can use a comparison of the received ultrasonic vibrations with a previously detected pulse of output ultrasonic vibrations that was received at the proximal surface (e.g. in a measurement made on a previous instance of doing the inspection) in order to identify a time of arrival of a current pulse of output ultrasonic vibrations. This comparison may use cross-correlation, cross-covariance, a similarity function or other forms of comparison seeking to match received ultrasonic vibrations with a previously detected pulse of output ultrasonic vibrations. The analysis performed may determine the pipe wall thickness, but may also or alternatively be used to detect other changes in the distal (inner) surface of the pipe, such as changes in the inner surface profile of the pipe due to different types of corrosion/erosion.

The results of the analysis by the server 18 may be sent to a user terminal 20 where they can be displayed and interpreted by a user of the system. It will be appreciated that the pipes 10, 12, 14, the sensors 4, 6, 8 and the gateway 16 may be at a different physical location (such as in a completely different country) from the server 18 and the user terminal 20, and the user terminal 20 may be in a different location from the server 18. The present techniques are well suited to remote monitoring of large scale plant, such as oil refineries or chemical processing plants.

The signal processing of the vibrations detected at the sensor 4, 6, 8 to generate the set of measurement data could be performed either at the server 18 or within circuitry local to the measurement sensors 4, 6, 8 themselves. In general, a set of measurement data can be determined which represents the waveform of the output vibrations detected in response to the pulse of input ultrasonic vibrations being transmitted into the surface of the pipe wall.

Figure 2 schematically illustrates the propagation of a pulse of ultrasonic vibrations through a pipe wall. The pulse of ultrasonic vibrations may be transmitted along a transmitting waveguide 22 to a proximal surface 24 of the pipe wall. The coupling to the proximal surface 24 may be direct or indirect. Received ultrasonic vibrations pass into a receiving waveguide 26 from the proximal surface 24 some time after the input pulse was sent into the pipe wall.

Illustrated in Figure 2 is a direct path 28 between the transmitting waveguide 22 and the receiving waveguide 26. This direct path gives rise to a reference pulse of ultrasonic vibrations that may be used to compensate for the transmission times along the waveguides 22, 26 as well as other effects, such as delays in the triggering and transmission of the pulse (other transducers operating in a pulse echo mode may use reflection from the proximal surface as the timing trigger). A first-order reflecting path 30 through the wall is illustrated showing the input ultrasonic vibrations propagating through the thickness of the wall, reflecting from a distal surface 32 of the wall and then returning through the thickness of the wall back to the proximal surface 24 where they form the current pulse of output ultrasonic vibrations for which the arrival time can be detected (e.g. using the cross-correlation, cross-co-variance, similarity functions or other forms of comparison as previously discussed). Determining the arrival time of this current pulse of output ultrasonic vibrations relative to the arrival time of the reference pulse permits a propagation delay time to be calculated corresponding to the propagation through the thickness of the pipe wall using standard trigonometry. This propagation delay may in return be used to determine a wall thickness and monitor factors such as the rate of corrosion or the rate of erosion of the wall.

While Figure 2 shows separate transmitting 22 and receiving 26 waveguides, other examples may detect the reflected output pulse of ultrasonic vibrations using the same device that transmitted the corresponding input pulse of ultrasonic vibrations. Hence, it is not essential to have a separate transmitter and receiver. A single waveguide could act as both transmitting waveguide 22 and receiving waveguide 26.

Figure 3 schematically illustrates received output ultrasonic vibrations at a proximal surface. A short time after the pulse of input ultrasonic vibrations is transmitted into the wall, a reference pulse of ultrasonic vibrations will be received corresponding to propagation of those ultrasonic vibrations along the direct path 28 parallel to the proximal surface of the wall. At a later time the detected pulse reflected from the distal surface of the wall along path 30 is received. The first-order reflected pulse will be the first to be received. In practice, the second-order and higher- order pulses may be too dispersed/confused to be accurately detected when the backwall is not smooth, but for smooth backwalls may provide useful additional information. Figure 3 schematically illustrates that the detected pulse reflected from the distal surface of the wall has been subject to phase changes and distortion due to reflection from the non-uniform distal surface in a manner which lengthens the pulse in time and changes the detailed phasing of the signal. The result of these distortions introduced by reflection from the non-uniform distal surface is that the time of arrival determined from the output waveform alone may sometimes be unreliable. The techniques described in WO 2014/167285 A1 referenced above may be used to improve reliability of measurements based on the reflection of a single pulse of input vibrations, by comparing (e.g. by cross-correlation, cross-covariance or another similarity metric) the waveform representing the corresponding pulse of output vibrations with a reference waveform such as an earlier measurement of the same test object or an idealised model waveform.

While Figure 3 shows an example using a reference pulse, it is not essential to use such a reference pulse, and other examples could use an internal timing reference. For example, the reference could be the time when the transmit signal was generated. The inventors have recognised that, although many measurements based on a single pulse of a given frequency content are successful and give a useful measure of thickness or another property of the pipe, sometimes such a measurement provides an unreliable result. This is because the test objects may have present any of a number of confounding features which can generate reflections which may distort the main peak caused by reflection from future of interest (e.g. reflection from the distal wall surface to measure thickness).

For example, for monitoring of a pipe wall, as shown in Figure 4 there may be a coating layer 40 (such as a metallic paint layer), which can generate reflections which distort the main pipe wall first order reflection. Also, there can be a layer of scale build-up 42 on the inside of the pipe caused by deposits from the products contained within the pipe (or caused by reaction between the product and the pipe wall material). As shown in the image in Figure 5, the scale layer thickness may vary around the circumference of the pipe and vary over time as the scale accumulates. The scale layer may be of non-uniform internal structure and so can generate highly variable reflections depending on the particular size and internal layer structure of the scale material. The presence or absence of scale, and if present its relative thickness, is not detectable from the outside the pipe and it would be undesirable to use a more intrusive technique to establish the scale thickness - this would defeat the whole purpose of non-invasively testing the pipe using ultrasonic vibrations. Hence, reflections from within the scale layer can cause distortion in the measured waveform, and it is not practical to determine by visual/manual inspection of the pipe the particular details of the scale layer to allow for quantification of the error in the estimate of the pipe wall thickness caused by the presence of the scale layer.

As shown in Figure 6, the transducer which generates the pulse of input ultrasonic vibrations may be capable of generating pulses at different frequencies. It can be useful to use an ultrasonic transducer which has a bandwidth (frequency range able to be generated by the transducer) that is greater than 100% of the centre frequency of its frequency range (that is, the difference between minimum and maximum frequencies supported by the transducer is greater than the centre frequency of the transducer’s frequency range), to increase the range of frequencies supported. It can also be useful to use a broadband pulse of input ultrasonic vibrations (which has frequency content spanning a larger range of frequencies in comparison to a narrowband pulse of relatively pure frequency content) so that the broadband pulse of input ultrasonic vibrations can be relatively short-lived in the time domain as shown in Figure 6, to reduce interference between the tail of the pulse of input vibrations and the incoming peaks of output vibrations detected following reflection of the input vibrations from the wall surface or other feature of interest (a narrowband pulse would have a much larger number of peaks extending for a greater period of time, so would make it harder to separate the output vibrations from the input vibrations in the measurement result). To reduce the number of peaks in the input pulse, it can be useful for the transducer to operate in a non-resonant mode (off-resonance mode) which is not at the resonant frequency of the transducer (e.g. if the transducer comprises a piezoelectric block, the block is excited at a frequency other than the natural resonance frequency of the block). Damping can be applied to the transducer to control its resonance frequency. As mentioned above, suitable transducers are described in WO 2007/051959 A1 and WO 2016/066997 A1 , but other transducer types can also be used.

The transducer can be controllable to generates different pulses of input ultrasonic vibrations with different frequency content. While Figure 6 shows an example of pulses of different centre frequency, pulses of input ultrasonic vibrations can also have different bandwidths (frequency range between the minimum and maximum frequencies included in the pulse), so two pulses with the same centre frequency could nevertheless include different frequency content. Also, it would be possible to use multi-burst pulses which include a number of bursts of ultrasonic vibrations having different frequency content, and provide for different burst patterns of frequency content in the respective bursts of a given multi-burst pulse.

The inventors have found that the effects of features such as scale or paint layers and internal voids or grains within the test object can be highly frequency dependent, so that while measurements taken with a first pulse of input ultrasonic vibrations having a first frequency content may produce a distorted waveform as shown in Figure 7, a measurement of the same test object using a second pulse of input ultrasonic vibrations having a second frequency content (e.g. a different centre frequency and/or bandwidth to the first pulse) can produce waveform which has a much less distorted main echo as shown in Figure 8, allowing for more precise determination of the time of arrival of the output ultrasonic vibrations and hence more accurate measurement of thickness or other properties of the test object.

Figure 9 illustrates an analysis of the error in thickness measurement predicted to occur when a single pulse of input vibrations is used to measure thickness, for different combinations of pulse centre frequency and the thickness of the scale layer 42 inside the pipe. The values plotted on the heat map are obtained by simulation of the waveforms modelled to occur for reflections from the inner pipe wall surface and a scale layer on the inner surface of the pipe wall, following transmission of a pulse of input vibrations at the outer surface of the pipe wall. While the legend on the right-hand side of the graph shows heat map representations for negative values for the thickness error, in practice the thickness error is zero or positive for all values of transducer frequency and scale thickness because the scale layer reflections will arrive later than the reflections from the distal wall of the pipe (given that the scale layer is further than the distal wall from the proximal wall at which the input vibrations are transmitted). Therefore, the effect of the scale layer in the event of distortion is to cause the thickness measurement to be an overestimate in the thickness rather than an under estimate. Hence, all values represented in Figure 9 are from the upper part of the thickness error range representing positive values corresponding to percentage error greater than or equal to 0%.

As can be seen in Figure 9, the effects of the distortion vary considerably for a given scale layer thickness as transducer frequency increases, and there can be regions where thickness error is greater for one frequency than another. It can be also seen that for a given scale layer thickness, there can be a region at a lower frequency and a region at a higher frequency which are both subject to appreciable thickness errors, with an intervening region at an intermediate frequency less affected by thickness error. This is because the distortion of an echo in the time domain caused by interference between reflections from confounding factors such as a metalscale interface and scale-fluid interface can change different features of the echo, such as the first crossing past a threshold, or time where the amplitude of the echo is maximum. Where thickness measurements are based on such properties of an echo, the measured thickness will have an error.

It will be appreciated that the simulation results shown in Figure 9 are simulating one particular setup for the pipe monitoring system and so the absolute values for the thickness error at different frequencies and scale thicknesses may be highly dependent on the particular setup of that system. Similar analysis could be done using a frequency domain representation of the captured signal where distortion can be seen as different frequency features in the frequency spectrum. Nevertheless, the principle that the error caused by distortion from reflections from the scale layer depends on both scale layer thickness and frequency is generally applicable to other setups.

The frequencies at which the thickness error is greatest may depend not only on the thickness of scale layer, but also on other properties such as the temperature at which the measurements are performed. Therefore, it can be difficult to predict in advance which frequencies would be most appropriate for measuring the pipe wall thickness for a particular system at a particular time.

Nevertheless, as shown by the crosses shown in Figure 9, if measurements are taken at multiple different frequencies, by supplying pulses of input ultrasonic vibrations with different frequency content, it is possible to apply an automated (computer-implemented) processing algorithm to the resulting sets of measurement data, to obtain a better estimate for the thickness or other property of the test object which is less affected by errors caused by scale layers, paint layers or other artefacts.

As shown in Figure 9, the variation of thickness error with transducer frequency for a given scale layer thickness can span frequency ranges wider than the typical range covered by a pulse used for ultrasonic testing. Hence, it can be useful for the pulses to have frequency content spanning a range of frequencies for which the maximum frequency Fmax included in any of the pulses used for the sets of measurement data and the minimum frequency Fmin included in any of the pulses have a ratio Fmax/Fmin which is greater than 1 .3. More particular, it can be useful to span over an octave frequency range (an octave being the interval between a frequency F and a frequency 2F), i.e. to have the ratio Fmax/Fmin being 2 or more. As shown in Figure 9, it can be further preferable for the ratio Fmax/Fmin to be even greater than two, e.g. to be greater than 2.5, greater than 3 or greater than 3.5, say. Although it would be possible to implement the technique with only two measurements at different frequency regions, to avoid incorrect results because both measurements occur in regions of increased distortion, it can be very useful to take more than two readings with different pulses of different frequency content, e.g. to take 3, 4, 5, or a greater number of measurement sets based on different pulses of input ultrasonic vibrations with different frequency content. Taking a larger number of sets of measurement data can also help to allow use of majority voting, comparison or aggregate functions to eliminate outlying data sets based on dissimilarity with a majority of other sets of measurement data.

Hence, by taking measurements at different frequencies, a more reliable result for the property can be obtained.

Figure 10 illustrates a method of capturing sets of measurement data based on ultrasonic non-destructive testing of a test object. At step 100, a pulse of input ultrasonic vibrations is transmitted into a surface of the object under test. At step 102, a set of output ultrasonic vibrations is detected at the surface of the object in response to transmission of the pulse of input ultrasonic vibrations. The output ultrasonic vibrations can be detected using a sensor at the same position at which the input ultrasonic vibrations were input (e.g. using a combined transmitter/receiver, such as a piezoelectric element which can be excited by applying an electronic signal to generate the input pulse and which outputs an electronic signal in response to the output ultrasonic vibrations causing vibration of the piezoelectric element). Alternatively, separate transmitter and receiver devices may be used to generate the input pulse and detect the output vibrations respectively.

At step 104, a set of measurement data is generated representing the detected set of output ultrasonic vibrations. For example, the electronic signal generated by the sensor based on the output ultrasonic vibrations may be sampled using an analogue to digital converter to generate a digital representation of the waveform of the output ultrasonic vibrations. Further signal processing (e.g. the cross-correlation with a reference waveform as described in WO 2014/167285 A1 ) can also be applied to generate the measurement data. The signal processing for generating the measurement data could be performed within the sensor 4, 6, 8 itself or at a separate location such as the gateway 16 or the server 18.

At step 106, it is determined whether all sets of measurement data required for the current instance of performing the test have been captured. The test requires at least two sets of measurement data to be captured in total, based on pulses of input ultrasonic vibrations with different frequency content. Hence, if there is still at least one remaining set of measurement data to be captured, then at step 108 the sensor 4, 6, 8 is controlled to change the parameters of the pulse of input ultrasonic vibrations to be used to capture the next set of measurement data. For example, the input waveform of the electronic signals provided to the transducer which will excite the pulse of input ultrasonic vibrations can be varied, by changing parameters which influence the centre frequency, bandwidth and/or burst pattern of frequency content of the pulse of input ultrasonic vibrations. The control of the input waveform can be performed by circuitry local to the sensor or can be defined by a remote device such as the gateway 16 on the server 18, which may provide commands instructing the sensor 4, 6, 8 to generate the corresponding input waveform.

Having defined the input pulse to be used for the next set of measurement data (so that the input pulse has different frequency content to pulses used for previous sets of measurement data), the method returns to step 100 to transmit the (updated) pulse of input ultrasonic vibrations and capture a corresponding set of measurement data. The pulse of input ultrasonic vibrations for the next set of measurement data is input into the object at a time when any reflections of the previous pulse of input ultrasonic vibrations used for the previous set of measurement data have died down so that the effects of the previous pulse of input ultrasonic vibrations are not detectable in the next set of measurement data. Hence, each set of measurement data represents an independent measurement, although in practice the sets of measurement data may be captured relatively closely together in time (e.g. at intervals of several seconds), with an interval between the transmission of the pulses of input ultrasonic vibrations for respective sets of measurement data that is longer than the expected delay between the transmission of a pulse of input ultrasonic vibrations and the receipt of the last detectable reflection in the output ultrasonic vibrations caused by that pulse of input ultrasonic vibrations.

Once all required sets of measurement data have been captured, then at step 1 10 the capture of measurement data is ended.

While Figure 10 shows an example where each set of measurement data is based on a pulse of input ultrasonic vibrations of different frequency content, it is also possible to capture more than one set of measurement data at each frequency content, while still having at least two sets of measurement data with respective input pulses of different frequency content.

Figure 1 1 shows a method of processing the sets of measurement data captured using the method of Figure 10. The method of Figure 11 could be performed immediately after capturing the measurement data, or could be performed at a later time. The method of Figure 11 could be performed on the same computing device which generates the measurement data at step 104 of Figure 10, or could be performed at a different computing device, e.g. a remote server 18 or user terminal 20 in the system of Figure 1 .

At step 150, a number of sets of measurement data are received, e.g. by receiving the measurement data direct from the computing device which calculates the measurement data, by receiving the measurement data over a network, or by reading the measurement data from a memory or other data storage device. There are at least two sets of measurement data. The sets of measurement data represent respective sets of output ultrasonic vibrations detected at a surface of the object under test, in response to transmission of respective pulses of input ultrasonic vibrations into a surface of the object. The pulses of input ultrasonic vibrations used to capture at least two of the sets of measurement data have different frequency content (e.g. different centre frequencies and/or different bandwidths).

At step 152 the sets of measurement data are processed to determine at least one property of the object. As discussed in the examples of Figures 12 to 16 below, there are a wide variety of ways in which this processing can be done, but in general the processing is aiming to identify which of the sets of measurement data may be less affected by errors caused by reflections from artefacts other than the feature of interest, so that the at least one property of the object can be determined based on those sets of measurement data deemed less affected by such errors. The measurement data could be processed in the time domain or in the frequency domain. For example, while Figure 3 shows an example of determining the propagation delay between pulses in the time domain (which can be used to determine wall thickness), the time domain signal could be transformed to the frequency domain, and in the frequency domain data spacing between interference minima can be related to the thickness. Similar concepts of looking for features and discrimination apply as in the time domain processing. Processing in the frequency domain can be particularly useful when the echoes received in response to the transmitted pulse overlap in time.

Figure 12 shows a first example of carrying out the processing step 152 of Figure 1 1 . At step 200, an optional discrimination step is performed, to apply a discrimination function to echoes in each set of measurement data. The discrimination function excludes at least one echo not meeting the discrimination criterion from subsequent analysis. For example, the discrimination criterion could be whether the maximum value of the echo is greater than a threshold value, or another thresholding function for eliminating weaker echoes in the waveform of the measurement data. This can be useful if the test object is such that it could include some internal structures such as voids or grains, which may produce weaker reflections at an earlier timing than the reflection of the main feature of interest (such as the proximal surface of the wall or position of a crack). By excluding such weaker reflections from subsequent analysis, the subsequent analysis can be more reliable.

At step 202, a feature value is determined for each of the measurement sets (if step 200 is performed, the feature value is determined based on the result of applying the discrimination function to each measurement set). For example, a thickness value or other property of the object is determined from each measurement set, or the feature value could be a waveform feature derived from the waveform represented by the measurement set (e.g. a measure of peak position, width and/or distortion).

At step 204, a given property of the object is determined as the aggregate function of two or more of the feature values determined at step 202. In some cases the aggregate function could be applied to all of the feature values determined at step 202. For example, the given property can be determined as the minimum, maximum, mean, mode or median of the property values determined at step 202. In other examples, the aggregate function could be applied to a subset of the feature values determined at step 202, for example those values which are not excluded due to representing nonsensical results (such as a wall thickness which is much greater than any previous wall thickness measured for the test object). In another example, a certain fraction of outlying results could be excluded from the subset of feature values used to apply the aggregate function.

Using the median of the feature values as the aggregate function can be particularly useful, because this is a simple way of implementing a majority voting scheme where the median result is likely to be representative of the majority of the results. However, other approaches may use more complex majority voting functions to evaluate whether a given feature value is in the majority of similar measurements that can be used to product the final result for the given property, or is in a minority of dissimilar measurements which should be excluded.

For the specific case of thickness measurement in the presence of scale layers, using the minimum function as the aggregate function (which returns as the given property the minimum of the two or more thickness values) can be useful as, especially if a discrimination function is applied at step 200 to exclude the effects of early reflections from voids, grains or pits in the wall caused by surface roughness, one can expect that the scale layer will cause late arriving reflections tending to distort the peak caused by reflection from the proximal wall surface so as to increase the estimated time of arrival of that peak, causing an overestimate in the thickness measurement. Therefore, the sets of measurement data which have a reflection peak arriving earlier can be estimated to be the sets of measurement data that are least affected by distortion caused by the presence of the scale layer, so that the most reliable thickness measurement to use may be the minimum thickness measured for the sets of measurement data that passed the discrimination function 200.

Figure 13 illustrates another example of the processing step 152 of Figure 1 1. At step 250 the sets of measurement data captured for the different frequency-content input pulses are compared, e.g. by cross-correlation or cross-covariance of their waveforms. At step 252 at least one measurement set to use for determining the property of the object under test is selected based on the comparison. For example, an outlying measurement that is dissimilar to the majority of other sets can be excluded from the subsequent calculation of the property of the object. For example, less similar waveforms tend to have a lower peak value of the cross-correlation or crosscovariance function, and so a measurement set which has low peak cross-correlation or low peak cross-covariance with other sets of measurement data can be excluded. It will be appreciated that other parameters could also be used to characterise the degree of similarity between waveforms of the sets of measurement data, and could be used to decide which sets of measurement data to exclude from subsequent calculation of the property of the object. Step 252 could select either a single set of measurement data (deemed representative of the majority of similar measurements) to use for subsequent calculation of the property of the object, or could select more than one set of measurement data which are then used in combination to calculate the property of the object (e.g. by deriving separate estimates of the object property from each of the selected sets of measurement data and then processing those estimates in an aggregate function such as determining the mean or median of the estimates).

Figure 13 shows an example where a measurement set is determined to be excluded due to having a feature indicative of low confidence that the measurement set can be used to provide a reliable estimate for the property of the test object. In Figure 13, the low confidence determination is based on the excluded measurement set being dissimilar to other measurement sets. However, in other examples, a measurement set could be determined to have low confidence in providing an accurate result for the property of the test object, based on other measures of confidence, such as how distorted an echo is, whether there is a repeating pattern of echoes, and/or how much the degree of distortion of the successive echoes changes.

Figure 14 illustrates another example of the processing step 152 of Figure 1 1. At step 300, for each measurement set, it is determined whether a result for a given property derived from that measurement set would be consistent with previous measurements of the given property for the object under test. For example, it can be determined whether a thickness measurement derived from a given measurement set is consistent with previous measurements of the thickness made at a previous time (e.g. on an earlier day). At step 302, a set of measurement data that gives a result inconsistent with previous measurements (e.g. one that gives a nonsensical result much greater than previous measurements) is excluded. One or more non-excluded measurement sets are then used to calculate at least one property (this can include properties other than the property that was compared with the previous measurement to determine whether the result was nonsensical - e.g. while comparison of thickness with a previous thickness measurement may be used to decide reliability of measurement sets to decide which sets to exclude, remaining measurement sets that were not excluded can then be used to determine other properties of the test object such as evaluating surface roughness).

Figure 15 illustrates another example of the processing step 152 of Figure 1 1. At step 350 the sets of measurement data captured using the input pulses of different frequency content are compared with at least one reference set of measurement data. This could be a previous set of measurement data captured at a previous time for the test object (e.g. captured on a previous day), or could be an idealised model waveform or a set of measurement data obtained by simulation. In general, the reference set of measurement data can be chosen to be relatively free of distortion or other causes of error and may have a relatively clear main peak corresponding to the arrival of the reflection of the input pulse from the feature of interest. At step 352, based on the comparison, at least one set of measurement data to use for determining at least one property of the object is selected based on the comparison. For example, a set of measurement data considered dissimilar from the reference set of measurement data may be excluded.

Figure 16 illustrates an example of Figure 15 in more detail. At step 400 a similarity metric (e.g. a metric based on cross-correlation or cross-covariance) is determined between each set of measurement data and a reference measurement set (which again could be a model/simulated waveform or a previously captured set of measurement data measured for the test object itself). At step 402, a selection is made between the sets of measurement data, to prioritise selection of a set of measurement data whose similarity metric indicates greater similarity to the reference measurement set, in preference to a less similar set of measurement data with a similarity metric indicating smaller similarity to the reference measurement set.

In some examples discussed above, a set of measurement data determined to have a feature indicating insufficient confidence in providing a correct estimate is excluded from the determination of the at least one property. It is also possible to reduce influence of the set of measurement data on the determination of the at least one property in other ways, e.g. by generating the measurement of the at least one property by a weighted combination of values derived from each set of measurement data, and assigning the erroneous set of measurement data a smaller weight than other non-erroneous sets of measurement data so that the influence of the erroneous set of measurement data can be reduced.

Further examples are set out in the following clauses:

1. A method for ultrasonic non-destructive testing of an object under test, the method comprising: receiving sets of measurement data representing respective sets of output ultrasonic vibrations detected at a surface of the object in response to transmission of respective pulses of input ultrasonic vibrations into the surface of the object, the sets of measurement data including at least two sets of measurement data representing output ultrasonic vibrations detected in response to transmission of pulses of input ultrasonic vibrations having different frequency content; and processing the sets of measurement data to determine at least one property of the object.

2. The method of clause 1 , in which the pulses of input ultrasonic vibrations used for said at least two sets of measurement data have at least one of: different centre frequencies; and different bandwidths; a different burst pattern of frequency content.

3. The method of any of clauses 1 and 2, in which the pulses of input ultrasonic vibrations used for said at least two sets of measurement data span a frequency range from a minimum frequency Fmin to a maximum frequency Fmax where Fmax/Fmin is greater than or equal to 1 .3.

4. The method of any preceding clause, in which the sets of measurement data comprise at least three sets of measurement data captured in response to transmission of pulses of input ultrasonic vibrations having different centre frequencies.

5. The method of any preceding clause, in which the object is a multi-layer structure comprising a plurality of layers.

6. The method of clause 4, in which the plurality of layers include a core layer and at least one of: a paint layer; a corroded surface layer; a scale layer; and a lining layer.

7. The method of any preceding clause, in which the object is a wall.

8. The method of clause 7, in which the wall is a wall of a pipe. 9. The method of any preceding clause, in which the at least one property comprises at least one of: a distance to a feature of the object; a thickness of at least one layer of the object; and a measure of surface roughness of a distal surface of the object from which the pulses of input ultrasonic vibrations are reflected after transmission into a proximal surface of the object.

10. The method of any preceding clause, in which the processing comprises determining a given property of the object as a function of two or more feature values derived from two or more of the at least two sets of measurement data respectively.

11. The method of clause 10, in which the processing comprises determining the given property of the object as an aggregate function of the two or more feature values.

12. The method of clause 1 1 , in which the aggregate function is one of: a mean of the two or more feature values; a median of the two or more feature values; a mode of the two or more feature values; and a minimum of the two or more feature values; and a maximum of the two or more feature values.

13. The method of any of clauses 10 to 12, in which determination of a feature value based on a given set of measurement data comprises applying a discrimination function to exclude from determination of the feature value a peak of the output ultrasonic vibrations which fails to meet a discrimination criterion.

14. The method of any preceding clause, in which the processing comprises excluding, from determination of the at least one property of the object, at least one set of measurement data determined to have a feature indicating insufficient confidence in that set of measurement data providing a correct estimate for the at least one property of the object.

15. The method of any preceding clause, in which the processing comprises selecting at least one of the sets of measurement data to use to determine a given property of the object, based on a comparison of the sets of measurement data.

16. The method of clause 15, comprising excluding, from determination of the property of the object, at least one outlying set of measurement data which is dissimilar to a majority of the other sets of measurement data.

17. The method of any preceding clause, in which the processing comprises: excluding, from the determination of at least one property of the object, a set of measurement data which would yield a result for a given property of the object which is inconsistent with previous measurements of the given property of the object, and determining the at least one property based on at least one other non-excluded set of measurement data.

18. The method of any preceding clause, in which the processing comprises selecting at least one of the sets of measurement data to use to determine the property, based on a comparison of the sets of measurement data with at least one reference set of measurement data. 19. The method of claim 18, in which each reference set of measurement data represents one of: an ideal set of output vibrations or a simulated set of output vibrations; and at least one previously captured set of measurement data captured previously for the object under test in response to transmission of a pulse of output vibrations into the surface of the object.

20. The method of any preceding clause, in which the processing comprises: determining, for each set of measurement data, a similarity metric based on a comparison with a reference set of measurement data; and selecting at least one of the sets of measurement data to use to determine the property, based on the similarity metric determined for each set of measurement data.

21. The method of clause 20, in which the similarity metric is obtained by at least one of: cross-correlation of a set of measurement data with the reference set of measurement data; and cross-covariance of a set of measurement data with the reference set of measurement data.

22. The method of any of clauses 20 and 21 , in which the selection of the at least one of the sets of measurement data prioritises selection of a set of measurement data with a similarity metric indicating greater similarity to the at least one reference set of measurement data in preference to selection of a set of measurement data with a similarity metric indicating less similarity to the at least one reference set of measurement data.

23. A computer program comprising instructions which, when executed by a computer, control the computer to perform the method of any preceding clause.

24. A storage medium storing the computer program of clause 23.

25. A system for performing ultrasonic non-destructive testing of an object under test, the system comprising: at least one measurement device for coupling to the object under test, to capture a set of measurement data representing output ultrasonic vibrations detected at a surface of the object in response to transmission of a pulse of input ultrasonic vibrations into the surface of the object; and processing circuitry to process at least two sets of measurement data captured by the at least one measurement device in response to transmission of pulses of input ultrasonic vibrations having different frequency content, to determine at least one property of the object.

In the present application, lists of features preceded with the phrase “at least one of” means that any one or more of those features can be provided either individually or in combination. For example, “at least one of: [A], [B] and [C]” encompasses any of the following options: A alone (without B or C), B alone (without A or C), C alone (without A or B), A and B in combination (without C), A and C in combination (without B), B and C in combination (without A), or A, B and C in combination.

Although illustrative embodiments of the invention have been described in detail herein with reference to the accompanying drawings, it is to be understood that the invention is not limited to those precise embodiments, and that various changes and modifications can be effected therein by one skilled in the art without departing from the scope of the invention as defined by the appended claims.

Claims

1 . A method for reducing influence of frequency dependent factors on a determination of at least one property of an object under test using ultrasonic non-destructive testing, the method comprising: receiving sets of measurement data representing respective sets of output ultrasonic vibrations detected at a surface of the object in response to transmission of respective pulses of input ultrasonic vibrations into the surface of the object, the sets of measurement data including at least two sets of measurement data representing output ultrasonic vibrations detected in response to transmission of pulses of input ultrasonic vibrations having different frequency content; and processing the sets of measurement data to determine at least one property of the object, where the processing comprises reducing influence on determination of the at least one property of the object of at least one set of measurement data having a feature indicating insufficient confidence in that set of measurement data providing a correct estimate for the at least one property of the object.

2. The method of claim 1 , in which the pulses of input ultrasonic vibrations used for said at least two sets of measurement data have at least one of: different centre frequencies; different bandwidths; and a different burst pattern of frequency content.

3. The method of any of claims 1 and 2, in which the pulses of input ultrasonic vibrations used for said at least two sets of measurement data span a frequency range from a minimum frequency Fmin to a maximum frequency Fmax where Fmax/Fmin is greater than or equal to 1 .3.

4. The method of any preceding claim, in which the sets of measurement data comprise at least three sets of measurement data captured in response to transmission of pulses of input ultrasonic vibrations having different centre frequencies.

5. The method of any preceding claim, in which the object is a multi-layer structure comprising a plurality of layers.

6. The method of claim 4, in which the plurality of layers include a core layer and at least one of: a paint layer; a corroded surface layer; a scale layer; and a lining layer.

7. The method of any preceding claim, in which the object is a wall.

8. The method of claim 7, in which the wall is a wall of a pipe.

9. The method of any preceding claim, in which the at least one property comprises at least one of: a distance to a feature of the object; a thickness of at least one layer of the object; and a measure of surface roughness of a distal surface of the object from which the pulses of input ultrasonic vibrations are reflected after transmission into a proximal surface of the object.

10. The method of any preceding claim, in which the processing comprises determining a given property of the object as a function of two or more feature values derived from two or more of the at least two sets of measurement data respectively.

11 . The method of claim 10, in which the processing comprises determining the given property of the object as an aggregate function of the two or more feature values.

12. The method of claim 1 1 , in which the aggregate function is one of: a mean of the two or more feature values; a median of the two or more feature values; a mode of the two or more feature values; and a minimum of the two or more feature values; and a maximum of the two or more feature values.

13. The method of any of claims 10 to 12, in which determination of a feature value based on a given set of measurement data comprises applying a discrimination function to exclude from determination of the feature value a peak of the output ultrasonic vibrations which fails to meet a discrimination criterion.

14. The method of any preceding claim, in which the processing comprises excluding, from determination of the at least one property of the object, said at least one set of measurement data determined to have the feature indicating insufficient confidence in that set of measurement data providing a correct estimate for the at least one property of the object.

15. The method of any preceding claim, in which the processing comprises selecting at least one of the sets of measurement data to use to determine a given property of the object, based on a comparison of the sets of measurement data.

16. The method of claim 15, comprising excluding, from determination of the property of the object, at least one outlying set of measurement data which is dissimilar to a majority of the other sets of measurement data.

17. The method of any preceding claim, in which the processing comprises: excluding, from the determination of at least one property of the object, a set of measurement data which would yield a result for a given property of the object which is inconsistent with previous measurements of the given property of the object, and determining the at least one property based on at least one other non-excluded set of measurement data.

18. The method of any preceding claim, in which the processing comprises selecting at least one of the sets of measurement data to use to determine the property, based on a comparison of the sets of measurement data with at least one reference set of measurement data.

19. The method of claim 17, in which each reference set of measurement data represents one of: an ideal set of output vibrations or a simulated set of output vibrations; and at least one previously captured set of measurement data captured previously for the object under test in response to transmission of a pulse of output vibrations into the surface of the object.

20. The method of any preceding claim, in which the processing comprises: determining, for each set of measurement data, a similarity metric based on a comparison with a reference set of measurement data; and selecting at least one of the sets of measurement data to use to determine the property, based on the similarity metric determined for each set of measurement data.

21 . The method of claim 20, in which the similarity metric is obtained by at least one of: cross-correlation of a set of measurement data with the reference set of measurement data; and cross-covariance of a set of measurement data with the reference set of measurement data.

22. The method of any of claims 20 and 21 , in which the selection of the at least one of the sets of measurement data prioritises selection of a set of measurement data with a similarity metric indicating greater similarity to the at least one reference set of measurement data in preference to selection of a set of measurement data with a similarity metric indicating less similarity to the at least one reference set of measurement data.

23. A computer program comprising instructions which, when executed by a computer, control the computer to perform the method of any preceding claim.

24. A storage medium storing the computer program of claim 23.

25. A system for performing ultrasonic non-destructive testing of an object under test, the system comprising: at least one measurement device for coupling to the object under test, to capture a set of measurement data representing output ultrasonic vibrations detected at a surface of the object in response to transmission of a pulse of input ultrasonic vibrations into the surface of the object; and processing circuitry to process at least two sets of measurement data captured by the at least one measurement device in response to transmission of pulses of input ultrasonic vibrations having different frequency content, to reduce influence of frequency dependent factors on a determination of at least one property of the object, where the processing circuitry is configured to reduce influence on determination of the at least one property of the object of at least one set of measurement data having a feature indicating insufficient confidence in that set of measurement data providing a correct estimate for the at least one property of the object.