WO2009118542A1 - Acoustic level determination of a material in a vessel - Google Patents

Abstract

An apparatus for the determination of level of a medium in a vessel comprising an impact device adapted to be disposed to effect mechanical impact upon a vessel at an impact site; a detector remotely spaced therefrom, and thereby attachable to a site on the vessel remote from the impact site; a data acquisition module to acquire frequency response data comprising at least frequency related amplitude data concerning vibration of the vessel at the detector; a data processing module to identify at least one characteristic peak value in the frequency response data; a data library of reference frequency response data for a plurality of vessel fill levels including at least data regarding the variation in frequency of at least one characteristic peak value with fill level.

Description

ACOUSTIC LEVEL DETERMINATION OF A MATERIAL IN A VESSEL

The invention relates to an apparatus and method for the determination of level of a material in a vessel, e.g. a flowable process medium such as a liquid, powder, slurry or the like. The invention especially relates to the in-situ non-invasive determination of a material such as a liquid level in a large storage vessel.

Process medium level measurement is an important diagnostic process that is widely used in the chemical industry and elsewhere to determine levels of a liquid or other process medium in vessels such as drums, reactor vessels, transportation or holding tanks, and the like. There are numerous level measuring devices currently available. However, most of these devices are invasive, requiring sensors to be placed in the vessel or the provision of sight-glasses through which observations may be made for example.

Practical limitations can arise when this process is required to be non-invasive. This has led industry to apply non-invasive solutions such as nucleonic techniques. Using this approach liquid level measurement is estimated from the signal strength attenuation that is observed using radioactive sources and detectors that are located around the vessel. Acoustic techniques have been explored to a limited extent as an alternative. Previous acoustic techniques have, largely, failed to remain non-invasive due to the acoustic impedance mismatch between the media. More specifically the few, non-invasive, ultrasound-based systems that are dedicated to level detection have generally been shown to able only to detect the presence of liquid at fixed points. Research on active acoustics has shown that sound waves are able to travel through walls when high voltage impulses are used. However these achievements are only useful under controlled laboratory conditions and, in contrast to typical process vessels, in pipes with a relatively small radius.

Non-invasive inspection using ultrasonic excitation is limited to small vessels due to acoustic impedance mismatch as shown by M. S. Greenwood and J. A. Bamberger, "Self-calibrating sensor for measuring density through stainless steel pipeline wall", Transactions of the ASME. Journal of Fluids Engineering, vol. 126, pp. 189, 2004. Greenwood and Bamberger's research was aimed at characterising fluids where liquid density was obtained from the acoustic impedance calculated from the echo reflections of a single, externally-injected, ultrasonic pulse within the wall of a steel pipe. Active transmission of sound waves was also explored where speed of sound and acoustic attenuation were evaluated from the multiple reflections within two diametrically opposed sensors for a single high-frequency pulse.

One of the first attempts to mix active acoustics and modal analysis to determine liquid level in scaled cylindrical pipes non-invasively is described in B. W. Northway, N. H. Hancock. and T. Tran- Cong, "Liquid level sensors using thin walled cylinders vibrating in circumferential modes", Measurement Science & Technology, vol. 6, pp. 85-93, 1995. Northway used a PVDF transducer to excite a 31.75mm diameter steel pipe with a frequency-swept sinusoidal wave whilst a second sensor detected the vibrations of the pipe. Northway analysed the first two circumferential modes of the pipe and explored the influence of the level of liquid on the natural frequency suggesting that further circumferential modes would result in a low sensitivity system. In this and other prior art sources the effective transmission of an acoustic wave through a vessel is constrained by limitations defined by the acoustic impedance mismatch and the acoustic attenuation of the media. Ultrasonic techniques are, therefore, limited to small diameter steel pipes and non-metallic vessels.

US 4480468 describes an apparatus for measuring the fluid level in a vessel using a pulse device to cause the vessel wall to vibrate, scanning the oscillation frequency of the wall and processing the output signal from a transducer abutting the wall for indicating the level. US 2003/0015036 describes an acoustic volume indicator comprising a resonator for causing a vessel to vibrate, a detector for receiving vibration data, a frequency detector for converting the vibration data to frequency and comparing the frequency to stored frequency v volume data to obtain the liquid volume which is output using an indicator.

It is generally desirable that apparatus and methods for the determination of process medium level in industrial process vessels using acoustic interrogation should be further developed. The resulting system should be non-invasive; non-hazardous; able to accommodate metal vessels; robust and practical for on-site application; able to accommodate large diameter and/ or tall vessels and a variety of vessel shapes, lagged vessels, and vessels at elevated temperature.

In accordance with the invention in a first aspect, an apparatus for the determination of level of a material in a vessel comprises: a) an impact device adapted to be disposed to effect mechanical impact upon the vessel at an impact site; b) a vibration detector adapted to detect vibration at a detection site on the vessel remote from the impact site; c) a data acquisition module for acquiring measured frequency response data comprising at least frequency related amplitude data concerning vibration of the vessel at the detection site; d) a data library of reference frequency response curves of resonant frequency versus material level for at least two identified characteristic peaks in the vibrational frequency spectrum of the vessel-material system; e) a data processing module for identifying at least two characteristic peaks in the measured frequency response data, said characteristic peaks being at least some of the characteristic peaks for which reference data has been provided; f) processing means for determining a vessel fill level by mapping the frequency of at least one of the measured characteristic peaks to the corresponding reference frequency response curve.

The apparatus of the invention thus acts in use to apply a mechanical percussive strike on a vessel, and to monitor the acoustic behaviour of the vessel in response to this strike. Vibration response data, for example in the form of a voltage response, is derived at the detector and analysed for frequency response. This enables a user, by identifying at least one and preferably a plurality of characteristic peaks in the frequency response data for example associated with particular vibration resonance modes, and by comparing these with reference data, to extrapolate a fill level in a non-invasive manner. Conveniently to facilitate this comparison the apparatus further includes means to enable determination of a frequency shift from reference data in the said at least one peak value, and thereby to extrapolate a fill level and for example includes a comparison module to perform this step.

Notably, it is found that frequency response data for a given vessel follows the same general form as fill level varies, but with a functional frequency shift. This can be identified particularly by identifying shifts in the frequency of identifiable characteristic peaks, for example those produced by particular structural resonance modes. The invention identifies one or more characteristic peaks in the frequency response, and references these to the reference data to determine the extent to which these have been shifted in frequency, and to correlate the shift in frequency to a fill level by comparison with the reference data.

The mechanical impact approach of the present invention can be contrasted to prior art such as that referenced by Northway and Bamberger where efforts to transmit an acoustic wave through a vessel are still limited by the acoustic impedance mismatch and the acoustic attenuation of the media and the practical application therefore limited to small diameter steel pipes and non-metallic vessels. In contrast the apparatus of the present invention uses relatively high-energy mechanical impact excitation, in a non-destructive manner, ensuring that a measurable proportion of the energy that is applied to the vessel is received by the detector.

The impact device preferably comprises a striker adapted to strike at an impact site on the vessel when the apparatus is in position in association with a vessel, for example mounted upon a vessel. The striker conveniently comprises a reciprocating mass and suitable actuator which acts to cause the mass to reciprocate. For example the reciprocating striker is a reciprocating armature. In one possible example the reciprocating striker is the armature of a linear solenoid. Other actuators, such as other electromechanical, mechanical, pneumatic or hydraulic actuators, could be envisaged. In one form, the impact device is adapted to effect a repeated successive impact upon a vessel in use. For example the impact device comprises a reciprocating mass and a suitable actuator which acts to cause the mass to reciprocate cyclically to produce a train of impacts, for example about ten, to generate a signal. This may be beneficial as it allows data to be collected over a longer time interval and may increase the level of excitation.

The reciprocating mass may be provided with a striking formation at the point intended to impact with the vessel. Preferably, the striking formation is of a resilient material of different, for example softer, composition from the material of the body of the reciprocating mass. For example the striking formation may comprise a polymeric material on a metal reciprocating mass body. Softer materials can be useful for preferentially exciting the lower frequencies that correspond to particular significant resonances in typical vessels.

Preferably, the impact device is adapted for mounting on a vessel at an impact site, and for example comprises a housing having suitable mounting means to effect this. A suitable impact site may be found on a vessel structure which is at least acoustically coupled to a vessel body containing material at a level to be determined, and will for example be on a vessel wall.

Preferably, the apparatus is adapted for mounting on a vessel in such manner that the impact device impacts a vessel structure such as a vessel wall in a direction normal to a vessel structure.

Preferably, the detector comprises a vibration sensor, and is for example a mechanical to electrical transducer. In use, the detector is attached in acoustically coupled engagement with a surface of a vessel, and is for example positioned in mechanical engagement thereupon. Conveniently therefore, the vibration sensor comprises a thin film vibration sensor. Thin film sensors are ideal due to their flexibility which allows ready mounting, with good contact area, on the curved wall of a container such as a cylindrical container. The vibration sensor is preferably composed of material that acts inherently as a mechanical to electrical transducer, for example comprising a piezoelectric material. In a particularly preferred embodiment, a sensor for use in accordance with the invention comprises a polyvinylidene fluoride (PVDF) thin film sensor.

In a particularly convenient embodiment, a secondary sensor is provided which, in use, is located at the impact site, with the impact device so disposed as to impact against or near this sensor at the impact site. The resulting response on the secondary sensor can be measured to act as a reference for further analysis. The primary detector sensor is located remote from the impact site, and serves as the principal data acquisition sensor.

In a preferred embodiment, a comparison module is provided to compare the measured signal with the data library and determine a frequency shift and hence extrapolate a fill level as above described. Alternatively the comparison step may be performed manually, for example by comparison against a hard reference library. The apparatus conveniently comprises a display to display a measured result, and may serve as a means to enable a comparison to be made with the data library.

Each of the modules in accordance with the invention, that is, the data acquisition module, the data processing module, the data library, and where applicable a comparison module, is conveniently an electronic data processing module, for example comprising an electronic data register and appropriate processing means. Suitable data links are provided between each module to enable communication between the modules. A central processor may be provided. One or more of the data processing modules may for example be disposed on or composed of a suitably programmed computer.

The data processing model module preferably includes a frequency response function analyser to derive a frequency response function from the acquired data. The frequency response function analyser preferably includes a fast Fourier transformer. The frequency response function analyser preferably includes means to calculate an auto-power spectrum (APS) and/or a cross-power spectrum (XPS). These minimise baseline effects and facilitate the recognition of characteristic resonant peaks, and hence the frequency shift in such peaks with varying fill level.

In a preferred embodiment an apparatus as above described is provided in combination form such that the combination is enabled in situ to process all data at the vessel and provide level data directly to a plant control system. Alternatively, some or all of the data processing and control functionality may be provided remotely; for example via a suitable remote data link.

In accordance with the invention in a further aspect, an apparatus as above described is provided fitted to a vessel in order to enable determination of material fill level in the vessel. In particular, the apparatus is disposed upon an external surface of the vessel in such a manner that an impact device is enabled to effect a mechanical impact at an impact site on a vessel and especially a vessel side wall; a detector is spaced remotely therefrom, and in particular on an opposite side, for example to be diametrically opposed; and fill level may therefore be determined in non-invasive manner. The selected impact site may depend on the vessel structure, and on constraints such as the presence of lagging, or physical structures such as pipe-work, walkways and access ports etc. A suitable impact site is located on a vessel structure which is acoustically coupled to the vessel body, for example a vessel wall. Preferably, the apparatus is fitted to effect impact generally normal to such a vessel structure or wall. Impact sites on structures welded to the vessel wall such as angle iron sections or blanked flanges on short pipe outlets, may be of use, for example on lagged vessels. In such a situation the impact is preferably normal to the vessel wall. It is also within the scope of the invention to impact onto plates welded flush to a vessel wall and this may be of use if a wear plate is required. It may be preferable that the impact site is located generally towards a midpoint of the height, for example at least no closer than 1/3 linear distance to the top or the bottom of a vessel.

The invention confers particular advantages where the vessel is a large static storage vessel, for example having a wall to wall distance across, and in the case of a cylindrical vessel a diameter, of at least 50 cm and/or a capacity of at least 100 litres. Particularly preferably, the vessel is a cylindrical storage vessel. However, although the invention offers particular advantages with larger vessels, it should not be considered as limited in application to such vessels. Prior art acoustic techniques typically only make point level measurements, and the present invention offers advantageous continuous monitoring of level for all vessel sizes.

In accordance with the invention in a further aspect, a method for the determination of level of a medium in a vessel comprises the use of the above apparatus to determine the level of a material such as a liquid at an unknown fill level in a vessel.

In particular, the method comprises the steps of: a) providing reference information comprising reference frequency response curves of resonant frequency versus material level for at least two identified characteristic peaks in the vibrational frequency spectrum of the vessel-material system; b) disposing a vibration detector in acoustically coupled engagement with a surface of said vessel; c) effecting mechanical impact on the vessel at an impact site remotely spaced from said detector; d) acquiring a vibration signal at the detector to produce measured frequency response data comprising frequency related amplitude data regarding the vibration of the vessel; e) processing said measured frequency response data to identify at least two characteristic peaks in the measured frequency response data, said characteristic peaks being at least some of the characteristic peaks for which reference data has been provided; f) determining a vessel fill level by mapping the frequency of at least one of the measured characteristic peaks to the corresponding reference frequency response curve.

The characteristic peaks in the spectrum are preferably resonant frequencies for the vessel- material system. The preferred method calculates the level within a vessel by tracking the frequency of one or more characteristic resonant frequency peaks.

The apparatus should be calibrated based on a current measurement and a set of frequency response curves for each natural frequency of interest. The frequency response curves used for the determination of liquid and peak tracking may be and are preferably obtained by the design and solution of mathematical models. A shell model comprising the vessel body and its physical constraints is designed based on Love's elasticity theory. The natural modes of the vessel are solved using finite element modelling tools. The natural vibration modes of the fluid contained are modelled similarly based on Helmholtz equation. The free frequencies of the vessel-fluid system may then be then obtained by coupling the response of the vessel and the fluid at various levels of fluid. The calibration process is a user-assisted operation that aims at correcting the inherent errors generated during the model. This calibration sets the number of peaks to be tracked and the frequency range (or "width") of the window in the frequency spectrum within which each peak is expected to be found.

When calibrated, the apparatus is ready to acquire and process data. The data processing operations comprise a signal processing module and a peak tracking module. The acquisition process consists of sampling data generated by the response of the system to a single impact or a set of multiple impact pulses. The recorded signal is then preferably processed digitally in a data processing unit which may be a dedicated unit or part of a personal computer for example. The processing step comprises resolving a frequency response function to facilitate the identification of a characteristic peak. This processing includes windowing (exponential for single pulse or Hanning for continuous pulses), filtering to reduce noise and calculation of the frequency content of the signal. Additional processing may include the determination of the cross power spectrum or auto- power spectrum along with further filtering to reduce baseline effects and facilitate the recognition of peaks representing resonance frequencies. Both acquisition and digital processing can be performed iteratively, i.e. engaged in a loop (Loop 2 in Fig 11) and repeated as required and the resulting data averaged to reduce noise and enhance peaks. This processing reduces or removes unwanted components on the spectra of the signals, enhancing the quality and facilitating the detection of resonant frequencies.

The resulting data set is then submitted to the peak tracking module to identify peaks and evaluate the corresponding level of liquid. During these two operations liquid level, location of peaks, windows and signal fingerprints are stored for further reference. The preferred system may run in continuous mode (Loop 1 of Fig 11) at intervals inversely proportional to the maximum rate of fluid flowing in/out of the vessel to allow proper operation of the peak tracking module.

The peak tracking module (shown schematically in Figure 12) is a set of processes that allow automatic peak recognition and tracking. This methodology is based on the detection of each characteristic peak within a predetermined set of points (window) within the frequency content of a signal. The window is a range of frequencies within which a characteristic peak is expected to be located. By operating on frequencies within a predetermined "window" or range of frequencies, it is not necessary to process the whole frequency spectrum, but rather only those parts of the spectrum within which the selected specific characteristic peaks are to be found . Initially, the frequency range of the window is calculated from the frequency of resonant peaks calculated from the initial system data, i.e. the modal or calibration data. In preferred operation, if a characteristic peak is found at higher or lower frequency than the central frequency of the window applied, then the frequency range of the window is recalculated and the resulting new, or corrected, window is used for the subsequent measurement. That is to say, each window is preferably selected to include the range of frequencies where the particular characteristic peak was found in a previous measured frequency spectrum. In this way, specific characteristic peaks of the frequency spectrum may be tracked and their shift in frequency used to determine the volume of material in the vessel- material system. The frequency range of each window and/or corrected window is stored so that it may be used to process a subsequent data set.

The module also employs a database of spectral fingerprints that serves as a reference for the window(s) to be applied. In a preferred embodiment, the first operation in the peak tracking module is determining the status of the fingerprints database. When this is empty the default window width specified in the calibration is applied. When the database is not empty the system then identifies if a spectral correlation is required. Spectral correlation is particularly useful when the system is started up after an elapsed time from the last liquid level measurement and is performed by correlating the current measurement to each of the datasets within the spectra database. The best match gives an approximation of the current level and the width and location of the window(s) within which peaks are to be searched for. If spectral correlation is not required then the corrected window from the last measurement is loaded and applied to the current measurement before being submitted to a peak detection algorithm.

Spectral peaks are identified from the spectral data within the applicable window. Spectral peak detection techniques are known, and any suitable technique can be used. One method which may be used to identify the relevant frequency peaks has the following steps. First, the frequency at which the gradient on the frequency spectrum changes from positive to negative is established. Once this peak has been identified a dynamic amplitude threshold is set a prescribed number of measurement points before and after the peak position. If all measurements below the peak and above the threshold amplitude are on a positive gradient and if all measurements above the peak and above the threshold amplitude are on a negative gradient the peak can be confirmed as not being a spike / noise. If an appropriate frequency window is then applied either side of the frequency peak of interest it may be tracked by numbering and then periodically counting the number of peaks up from zero Hertz. An appropriate frequency window may be established through prior modelling work. This technique has been used for automated peak tracking.

A peak detection algorithm is shown in Fig 6. It has been designed to identify peaks based on the general shape of a peak. The data points of the window being processed are evaluated to determine a change of the polarity of the slope indicating the presence of a peak. To confirm that the point in question is a peak a second rule should be applied: the amplitude of a predetermined number of the data points preceding or succeeding the point evaluated must be of lower value than that of the point in question. When a peak has been identified using the rules of a peak detection algorithm, the peak is marked and a new threshold level is devised based on the amplitude of the peak relative to the baseline. Similarly the width and location of the window containing the peak is saved.

The resulting peaks are used to determine liquid level. The frequency of each peak is mapped to the frequency response curves correlating liquid level with resonant frequency and the liquid level value is calculated using linear interpolation.

In a preferred embodiment of the method, more than one resonant frequency peak is tracked and a frequency response curve is assembled for each of the tracked resonances, as shown in Fig 9.

The movement of the resonant frequency peaks is not linear with vessel fill level and in some regions, particularly at low and high vessel fill, the frequency response may be relatively flat.

Different resonant frequencies respond in differing manners to vessel fill changes. This eliminates the possibility of using linear interpolation to approximate this volume. Furthermore, the range where the volume of liquid can be approximated using a single resonance curve may be limited to about 20%-80% of the tank volume. It is therefore preferable to track the movement of multiple resonant frequency peaks.

To increase accuracy in the determination of liquid level a curve with low sensitivity regions (where the curve has a flat or nearly flat slope) may be ignored when calculating liquid volume from a particular resonant frequency peak. This operation is performed by a curve discriminator. This process locates the frequency of each peak on their corresponding frequency response curve and discards those curves which have a slope less than a predefined value. When more than one peak is found within a particular window, the curve discriminator may be programmed to discard these peaks and switch to another curve/peak. Alternatively the curve discriminator may calculate the trend of the peaks (using information in the spectra database) and use a Kalman filter to predict each peak location.

Four algorithms that may be used for tracking the movement, i.e. the shift in frequency, of multiple resonant frequency peaks between spectra measured at different fill levels are further described below.

1) A windowing and peak counting technique, as described above can be used to identify the positions of several peaks. The average vessel fill value obtained with these peaks could then be used to improve the accuracy of level measurements over the full range of vessel fill. This technique will allow a measure of compensation for regions of flat response for a particular resonant frequency. 2) A windowing and peak counting technique, as described previously, can be used to identify the positions of several peaks. An algorithm may be implemented that will allow the system to track a particular resonant frequency peak over a particular vessel fill range. When this particular resonant frequency approaches a region of flat response the system is programmed to switch to monitoring a different particular resonant frequency peak that is producing a greater response to changes in vessel fill.

3) Frequency correlation. An algorithm may be implemented to determine the overall frequency shift of peaks within a specified frequency range. This algorithm would operate similarly to the cross-correlation method to determine lag of non-periodical signals; the shift of the frequency range of interest is determined by evaluating the convolution of this with the spectrum obtained for a known vessel fill level.

4) Peak shape / neural network. Similarly to frequency correlation, this technique would determine frequency lag of a particular peak by identifying the overall shape of one or more consecutive, non- overlapping peaks using a neural network. The neural network must be trained using information (spectra) obtained at known levels.

These multiple peak assessment algorithms may be employed individually or in combination.

It is preferred to select low frequency resonant frequencies as higher frequency peaks have a tendency to move over large frequency ranges relative to each other and may in some cases cross over. Peak cross over would render peak tracking impossible with most techniques.

The invention will now be described by way of example only with reference to the accompanying drawings in which:

Figure 1 is an illustration of a possible algorithm used to obtain the frequency response function;

Figure 2 is a sensor suitable for incorporation into a device in accordance with the inventions;

Figure 3 illustrates the frequency response of the sensor of figure 2; Figure 4 illustrates super imposed traces of signals received at the detector sensor in use;

Figure 5 illustrates variation in traces with fill level;

Figure 6 is an example of an operations chart for a peak detection algorithm;

Figure 7 illustrates peaks detected by the algorithm of figure 6;

Figure 8 illustrates characteristic resonant frequency shift with fill level; Figure 9 illustrates frequency response curves with fill level for various resonant states;

Figure 10 is a schematic diagram of one embodiment of the level detection apparatus;

Figure 11 is a system operation chart for the method used in the invention;

Figure 12 is^' an operations chart for the peak tracking module. In the example the use of an apparatus and method comprising an embodiment of the invention with a cylindrical vessel for containing a liquid medium is considered. Such vessels are commonly found in industry in the form of water tanks and stirred vessels. Other vessel shapes, including without limitation cylinders with funnelled ends, horizontal cylinders and rectangular vessels, and other contained media, including without limitation solids such as flowable solids for example in powder form, slurries and the like, can be envisaged without departing from the scope of the invention.

As an example, a vessel, 10 is shown in Figure 10. The vessel is partially filled with a liquid 12 and has in-flow 14 and out-flow 16 pipe-work for filling and emptying the vessel. An impact device 18 is mounted on the vessel and is arranged to strike the vessel in response to a signal from a control module 24. A vibration sensor 22 is mounted on an opposed part of the vessel and is acoustically coupled to the vessel wall. Signals from the sensor are processed by the module 24 and output to an output module 26 which is a computer. For the presented embodiment a vertical, cylindrical steel vessel 10 with diameter 60cm, height 110cm and wall thickness 5.6mm has been used. In order to allow analysis cylindrical containers can be regarded as cylindrical shells. The vibration of thin cylindrical shells has been widely studied and any suitable model therefor may be applied.

The system developed uses polyvinylidene fluoride (PVDF) thin film sensors. PVDF is a polymer that exhibits a strong piezoelectric activity that makes it ideal for implementing acoustic sensors. Although electrical to mechanical conversion using PVDF is not as effective as that of ceramic substrates such as PZT it is an excellent material for some ultrasonic transmission applications due to its great bandwidth. Thin film sensors are ideal in the present application due to their flexibility which allows ready mounting, with good contact area, on the wall of a cylindrical container.

The transducers that are used in the embodiment are shown in schematic in Figure 2. These sensors measure 41 x 16mm, consist of rectangular elements of 52μm thick piezoelectric film with silver ink, screen printed, electrodes, and have a wideband response as shown in Figure 3. This makes them ideal for obtaining the frequency response of a system with minimal disturbance due to the sensor. This frequency response was obtained for the specified sensors by observing the relationship between emission and reception of a sinusoidal chirp signal that is transmitted between two PVDF sensors. The low cut-off frequency is determined by the input impedance of the acquisition system.

A conditioning circuit for the signals that are received from the sensors has been designed to improve the quality of the data that are to be processed. The signals that are sensed by the PVDF are passed, initially, through an instrumentation voltage amplifier. A second stage, comprising a series of programmable amplifiers and filters, sets a cut-off frequency of 10 kHz and ensures that aliasing is reduced. For impact excitation the maximum frequency that is processed in the present system is 10 kHz. The last stage of the circuit conditions the signal such that it can be digitised by a NI-6143 acquisition card. A sampling frequency of 125 kHz has been used. The electronic system has been designed to acquire a maximum of 8 simultaneous channels but expandability to create circular arrays of receivers is feasible. Digital triggering creates a time delay in the data that are acquired as the mechanical response of the impact device is not constant. This problem is corrected in the digital signal processing stage on the computer.

The impact device consists of a 24VDC, push acting, linear solenoid 28 with a maximum stroke of 5mm. This solenoid, mounted on an aluminium support, is located at mid height (55cm) on the outer wall of the tank and it is actuated according to parameters that are set by the user. This solenoid generates a sharp impulse on the outer wall of the tank by hitting the plunger 18 against one of the PVDF sensors 20. The resulting response on this sensor acts as a reference for further analysis. A second PVDF sensor 22 is located diametrically opposite the impact point and serves as the principal data input.

In an example method of use, a short 24V pulse (15ms) is applied to the impact device to create a sharp impulse directly on the surface of the sensor at the impact site mounted on the external wall of the tank. The range of interest, in the audible region, is sampled at a frequency of 100 kHz to obtain good quality measurements and anti-aliasing filters are set to a cut-off frequency of 10 kHz corresponding to the maximum frequency excited by a steel-tip impact. Time traces from the sensor placed on the opposite side of the vessel to the impact (the principal data input sensor) show that the response is highly repeatable as shown on the representative traces in Figure 4 where 10 acquired signals are overlaid. This degree of repeatability has been observed in traces lasting up to 800ms.

Signal processing of the acquired data is performed initially on a custom-designed conditioning circuitry. The processing includes, among other operations, triggering delay correction, filtering and calculation of the spectra of the signals from which resonant frequencies are extracted. This feature extraction is based on a peak detection algorithm that is specifically designed to determine the frequency response curves from which the volume of the vessel is estimated. A process to obtain a frequency response function from the signals acquired by the method of, or by the acquisition module of, the invention is presented in Figure 1. This algorithm, as seen on the diagram, includes the calculation of the auto-power spectrum (APS) and the cross-power spectrum (XPS) of the signals, as these facilitate the recognition of resonant peaks due to the minimisation of baseline effects.

Alternative signal processing to obtain the frequency response of the tank have been considered. Wavelet analysis and spectrograms might, for instance, yield information regarding the free decay of the signal if performed in its continuous form while discrete wavelet analysis or the use of wavelet packets in combination with neural networks could facilitate a fully automatic system. This latter form of processing would, however, require significant computational resource and the reliability of the system would depend on a comprehensive calibration that in many industrial processes is, practically, impossible.

Data have been acquired for different levels of water starting from H/L=O up to H/L=1 where "H" is the level of water and "L" is the height of the vessel. For instance in Figure 5 a significant change in the time traces is noticeable for 0 litres and 315 litres of water and this illustrates the richness of data that are provided using this technique. The change in response is the result of the interaction of the fluid with the structure including hydrostatic pressure on the inner walls and added mass to the coupled system. Measurements such as these have been processed to deduce the frequency response of the tank.

Spectral peak detection techniques are already known. A module containing a basic algorithm is described in Figure 6. The module features the identification of potential resonant frequencies by identifying peaks in the spectral content of a signal. The suggested algorithm to detect these peaks follows a threshold detection with a slight modification to increase peak-shape sensitivity. The example algorithm referred to herein as the peak detection algorithm (PDA), is presented in Figure 6. The basic criterion for the detection of a possible resonant frequency peak is based on a comparison of the amplitude of a point with respect to the other data points in its vicinity. The algorithm presented here detects peaks with high accuracy in relatively noiseless signals as expected. However, some disadvantages are readily evident: discontinuities due to noise can be interpreted as possible resonant peaks. Moreover, closely located resonant frequencies can be discarded in error due to the shape conservation criterion.

Figure 7 presents an example of peaks detected and discarded using the PDA module where two resonant frequencies are present. The threshold level for this example is five units and the number of points in the forward vicinity is four. Using these conditions, the two main resonant frequencies, indicated as P1 and P3, are detected successfully. Peaks due to noise, as indicated by P2, are excluded because, for example, the amplitude of the following four points after P2 do not decay below the threshold level. However, this criterion detects noise peaks such as the one indicated as P4 when they are located in close proximity to a resonance peak.

Windows of 2.5s duration were processed as these included 99% of the energy of the signals acquired. This yielded frequency resolution of 0.4 Hz. The resulting spectral fingerprints can be analysed for different levels of water in the vessel. For instance the resonance at 326 Hz (H/L=0) representing the 5th circumferential mode in Figure 8 can be observed to move as the liquid level increases as shown for the case with 315 litres of water (H/L=1) in Figure 8 in which the peak has moved to 213 Hz. Similar behaviour affects the other natural frequencies. The relationship between the resonant frequencies of three resonances and the known volume of liquid is shown in Figure 9 and, from this, the liquid level can be deduced. A 9th degree polynomial was fitted to the lowest natural frequency curve from Figure 9 to evaluate the repeatability of the system. A volume of 133.5 litres of water was added to the tank and a new data set was acquired. The predicted volume, 136 litres, was found to be within +2% of the real volume. These results validated the potential of the technique as a way of determining the volume-frequency relation. -

In order to emulate a "real" situation and to determine the effect on the measurements a stirrer was mounted on the tank. Acoustic emission signals showed that broadband noise was added to the system. Nevertheless when data were acquired using impact excitation the resonant components were still extracted successfully. The estimated volume of water was correctly determined to be 225 litres.

Finally, the resonant frequencies that have been derived from the measurements have been compared to those obtained from the preliminary finite element model of the dynamic structure for an empty vessel. An identical model has been solved using aluminium as the material for the vessel. The results confirm that frequencies associated with the same vibration modes as those with a steel-based model depend on the properties of the material. In practice the bending shape of the vessel at resonant frequencies must be obtained, when required, from the frequency response function of multiple traces obtained in circular arrays of sensors located at different planes.

Claims

Claims

1. A method for the determination of level of a material in a vessel comprising the steps of: a) providing reference information comprising at least two reference frequency response curves of resonant frequency versus material level for at least two selected characteristic peaks in the vibrational frequency spectrum of the vessel-material system; b) disposing a vibration detector in acoustically coupled engagement with a surface of said vessel; c) effecting mechanical impact on the vessel at an impact site remotely spaced from said detector; d) acquiring a vibration signal at the detector to produce measured frequency response data comprising frequency related amplitude data regarding the vibration of the vessel; e) processing said measured frequency response data to identify at least two characteristic peaks in the measured frequency response data, said characteristic peaks being at least some of the selected characteristic peaks for which reference data has been provided; f) determining a vessel fill level by mapping the frequency of at least one of the measured characteristic peaks to the corresponding reference frequency response curve.

2. A method according to claim 1 , wherein said reference information is derived by the solution of mathematical models for the frequency response of the vessel and material.

3. A method according to claim 1 or claim 2, wherein said reference information is derived from a calibration of the vessel-material system.

4. A method according to claim 3, wherein said calibration is used to modify reference information derived from mathematical models.

5. A method according to any one of the preceding claims, wherein, when the frequency of a measured characteristic peak corresponds with a flat portion of the reference frequency response curve, the fill level is determined by mapping another one of the measured characteristic peaks to its respective reference frequency response curve.

6. A method according to any one of the preceding claims wherein the vessel fill level is determined by mapping the frequency of at least two of the measured characteristic peaks to the corresponding reference frequency response curves.

7. A method according to any one of the preceding claims, wherein frequency response data and optionally also corresponding calculated fill level information is stored in a spectrum database.

8. A method according to claim 7, wherein the measured frequency response data is compared with stored frequency response data to identify parameters for identification of the characteristic peaks.

9. A method according to any one of the preceding claims, wherein the measured frequency response data is processed by a signal processing module and a peak tracking module.

10. A method according to claim 9, wherein the peak tracking module identifies the frequency of a characteristic peak by a method including the steps of: a) evaluating the data points within a predetermined frequency window to determine the frequency at which the polarity of the slope changes, indicating the presence of a possible peak; and b) determining whether the amplitude of a predetermined number of the data points preceding or succeeding the possible peak identified in step a) are of lower value than that of the possible peak.

11. A method according to claim 10, wherein the frequency range of a new frequency window is calculated from the frequency of said characteristic peak, stored and used to process a subsequent data set.

12. An apparatus for estimating the level of a material in a vessel comprising: a) an impact device adapted to be disposed to effect mechanical impact upon the vessel at an impact site; b) a vibration detector adapted to detect vibration at a detection site on the vessel remote from the impact site; c) a data acquisition module for acquiring measured frequency response data comprising at least frequency related amplitude data concerning vibration of the vessel at the detection site; d) a data library of reference frequency response curves of resonant frequency versus material level for at least two identified characteristic peaks in the vibrational frequency spectrum of the vessel-material system; e) a data processing module for identifying at least two characteristic peaks in the measured frequency response data, said characteristic peaks being at least some of the characteristic peaks for which reference data has been provided; f) processing means for determining a vessel fill level by mapping the frequency of at least one of the measured characteristic peaks to the corresponding reference frequency response curve.

13. An apparatus according to claim 12, wherein the detector comprises a vibration sensor comprising a mechanical to electrical transducer.

14. An apparatus according to any one of claims 12 or 13, wherein the impact device comprises a reciprocating mass and a suitable actuator wherein the reciprocating mass is provided with a striking formation composed of a resilient material of different composition from the material of a body of the reciprocating mass.

15. An apparatus according to any one of claims 12 to 14, wherein a secondary vibration detector is provided, said secondary vibration detector being located, in use, at the impact site.