An Experiment to Investigate the Feasibility of Using Piezoelectric Transducers to Monitor Corrosion Damage in Storage Tanks
Stephen M Williams
Paper presented at II Eccomas thematic conference on smart structures and materials, C.A. Mota Soares et al. (Eds.), Lisbon, Portugal, July 18-21, 2005
This experiment was an investigation into the feasibility of detecting structural change comparable with that of typical corrosion in storage tanks by generating and detecting low amplitude structural vibrations with piezoelectric actuators and sensors. The work was part funded by the EC within the GROWTH initiative (fifth framework) under the acronym 'Piezodiagnostics', from which derives the term for the technological concept, 'PD technology'.
Testing at long range (several metres) can be accomplished by using Lamb waves. The ultimate testing range achievable is limited by the attenuation of the waves due to beam spread and scattering. The concept behind the approach in this research is that by reducing the testing frequency to that of global structural vibration, scattering can be reduced leading to much longer testing ranges and permitting the use of modal analysis for condition monitoring. There is no attenuation due to beam spread as modal testing effectively employs a standing wave. The approach used in this project differs from conventional modal testing because piezoelectric actuators rather than electromechanical actuators generate the vibrations. Consequently, the amplitude of vibrations generated and detected is much smaller.
The main finding was that the generation and measurement of the modal response using low amplitude structural vibrations was shown to be possible on a scaled model tank using piezoelectric transducers. Furthermore, prominent indications appeared on the FRF of the tank wall when masses, representative of comparable corrosion, were attached.
The work described in this report was conducted at TWI and jointly funded by the TWI corporate research programme (CRP), and partly by the EC through a 5th framework project called 'Piezodiagnostics'  . The project consortium had nine partner companies from the UK, France, Poland and Spain. Piezodiagnostics was part of the EC GROWTH programme, its duration was 3 years and it lasted between 2002 and 2005 (visit www.cordis.lu/ for details).
The identification of corrosion in large structures is generally difficult without full access, which is not always possible. Structural degradation due to corrosion tends to be slow and progressive. Once damage grows beyond acritical size catastrophic failure can occur suddenly, hence there is a need to identify and monitor its course. Lamb waves, which exist at frequencies between tens and hundreds of kilohertz in steel, are increasingly being used for long range monitoring. Their main advantage is that they can be propagated over substantial distances with minimal attenuation. The success of Lamb waves for pipe-work inspection owes much to the prismatic properties of this type of structure. For non-prismatic structures such as storage tanks, however, attenuation due to beam spread and scattering is a serious drawback and consequently no commercial systems based on this technology are currently in existence. The Piezodiagnostics project was dedicated to the development of a novel concept. The principle is that by reducing the testing frequency to that of structural vibrations, scattering can be reduced still further leading to much longer testing ranges than those achievable with Lamb waves. Furthermore, the approach used in this project was to excite with piezoelectric rather than electromechanical devices. Consequently, the amplitude of vibrations generated and detected is much smaller. Such low amplitude microscopic vibrations require much less energy to induce in the structure, are intrinsically safe and represent a realistic and practical approach to the condition monitoring of large industrial structures. The project consortium has named this concept ''PD' (PiezoDiagnostic) technology'.
This paper describes research to develop PD technology for monitoring corrosion in storage tanks. These structures do not have a prismatic symmetry so beam spread would be a major source of attenuation in a pulse propagated from one point and directly detected at another; therefore a different approach is required. Conventional modal testing involves the harmonic vibration of the whole structure at frequencies that are dependent on its size and geometry. The reported sensitivity of these methods has generally been poor owing to the comparatively long wavelengths involved.
At present the most widely used commercial monitoring systems applicable to oil storage tanks are based on acoustic emission (AE). These systems are used for corrosion mapping of the tank floor  . AE usually has to be followed up with another method, such as magnetic flux leakage (MFL), which can detect and size individual defects. However, the correlation between AE data and flux leakage results suggests that 'AE testing always should be considered as a global test and that decision taking is not black and white 'yes or no''  . Therefore, there exists a clear need for a more reliable method of global monitoring with fewer false calls.
Low frequency modal testing would appear to offer the benefit of being able to acquire a fingerprint of the whole structure, or a large part of it, in a single relatively short test (short compared with the time required to cover a whole tank with a technique such as MFL). In conventional modal testing macroscopic structural vibrations are induced by gross disturbances. Continuous excitation is usually induced by electromechanical means which would be difficult to implement on a real tank. This experiment seeks to determine whether by employing microscopic structural vibrations PD technology offers a practical solution that can be implemented on storage tanks. Furthermore it seeks to determine whether using such vibrations for modal testing results in improved sensitivity compared with conventional electromechanical excitation. Better sensitivity would allow less substantial and more typical corrosion or structural modification to be identified.
- To investigate by experimentation the possibility of obtaining modal properties from the wall of a model storage tank by generating and detecting microscopic structural vibrations with piezoelectric transducers (PD technology).
- To assess the potential of PD technology for detecting structural modifications typical of localized corrosion in storage tanks.
4. Experimental approach
The main approach was to obtain the modal response of the tank wall by conducting a stepped frequency scan and comparing the fingerprint generated under different configurations (locations of transducers) and in different tank states (location of simulated corrosion, if present). In a stepped frequency scan the excitation frequency is changed in steps of a fixed size. At each frequency, a steady state sinusoidal structural response, of the same frequency as the excitation, is established and measured simultaneously at a number of locations on the structure. At a natural frequency (otherwise known as a mode) of the structure, the structural vibrations reinforce (constructive interference)and a large response is measured.
The resolution of the stepped scan was 0.1Hz. Corrosion was simulated by added mass (see section 5.2). The frequency response function (FRF) was used for fingerprint analysis. It is defined as  :
FRF = Response / Force
The force was measured by a sensor located in an impedance head. The measured force varies according to the instantaneous acceleration, its peak magnitude being dependent on the loading. The response was measured by five sensors; one of these sensors was a Bruel and Kjaer type 8001 accelerometer  , housed in the impedance head. Its response was proportional to the axial acceleration. The other four sensors were prototype devices provided by CEGELEC CNDT of France. The response of these sensors was proportional to local deformation. A conventional FRF is derived from displacements, velocities or accelerations and therefore, strictly speaking, the CEGELEC sensors do not measure the true FRF. However, it is reasonable to assume that some proportionality exists between radial displacement and local distortion in the region of the sensor. This being the case then the transfer FRF derived from the sensors will exhibit the characteristics of an FRF acquired by conventional measurements. The CEGELEC sensors, being separate from the force sensor, each generated a transfer-FRF (sensor response divided by force measured at impedance head).
This paper describes experiments to evaluate the capability of PD technology for monitoring the tank wall. Techniques for monitoring the tank floor were explored but are not reported here as the data were inconclusive.
5. Equipment and apparatus
5.1 Model tank
The model tank was designed and built at TWI. It consisted of a 4m diameter tank with a bottom plate comprising two 6mm thick semi-circular plates of steel. The plates were butt-welded together and a 1 metre high shell consisting of three sections was fillet welded to the bottom plate. The three sections were themselves butt-welded together. The bottom plate was trimmed to leave a rim of 100mm. The whole tank was treated with a bituminous coating. While not being an exact scaled replica this design still represented some basic tank features such as the shell, fillet weld, rim and butt welds. The tank was mounted on sand to simulate the foundation materials used in practice.
5.2 Added masses
In this experiment added mass rather than metal loss represented structural change due to corrosion. At the long wavelengths used in these experiments, loss of section due to corrosion primarily reduces the local mass and stiffness causing changes in the dynamic properties. An equivalent increase in mass will cause local stiffness increase and comparable changes in the dynamic properties. Therefore added mass was considered to be equally valid as a way of demonstrating the potential to detect section loss caused by corrosion and is more convenient as it can be made reversible. There were 2 attached masses and these are illustrated in Figure 1(a) and Figure 1(b). One was a circular steel disc of 2.5 mm thickness and 26 cm diameter. This was attached to the wall with Araldite rapid adhesive. The other was a steel block that was attached with a clamp.
The sensors supplied by CEGELEC are prototype piezo-ceramic devices whose output voltage is proportional to the local surface deformation. The basic structure of this type of sensor is shown in Figure 1(f). The sensor is attached with a layer of fast-acting HBM X60 epoxy, the main characteristic of which is its hardness. The actuators supplied by CEDRAT are amplified piezo-actuators (APA). These are employed to excite the tank's structural vibration response. The basic structure of an APA is shown in Figure 1(e). The actuator works by pulling its two poles together in response to an applied voltage within the range -20V to 150V, which is generated by the conditioning unit in response to an order signal between -1V and7.5V. The voltage causes the bar of piezo-ceramic material between the poles to extend. Once the voltage is removed the actuator returns to its original shape.
5.4 Testing apparatus
The tank wall was excited with the active-tendon configuration. Each end of the actuator is connected to an adapter that interfaces it with a Dyneema cable (made by Eurocord BV). This material was chosen because it was lightweight and therefore the effects of self-weight would be negligible. The cable is attached to the tank wall with a cable clamp. A tension of approximately 100N was applied by pulling the cable through the clamp with a hanging weight gauge 1 . The actuator suspended in the centre of the tank is shown in Figure 1(c), and the actuator wall mounting is shown in Figure 1(d). The two positions of the impedance heads allowed verification that the force produced by the actuator was the undistorted order signal transmitted through the cable to tank wall.
Fig.1. Tank monitoring apparatus
Figure 2 illustrates the different testing configurations and tank states. 'Active tendon' was used for excitation of the tank wall. In 'active tendon' there were two possible locations of the cable. These were either configuration AT1 or AT2 depending on whether the actuator's cable was connected to location 1 or location 2. Both locations were across diameters of the tank and were at right angles to the each other. The results from AT1 were omitted for brevity. They were generally similar to those presented for AT2.
To test the PD concept on tanks a range of stepped frequency scans were conducted exclusively using piezoelectric actuators and sensors. The experiments are summarized in Table 1. In this table lower case letters indicates location on wall (W) and 'As new' means no attached mass.
In the first 4 experiments scans were performed between 1Hz and 200 Hz at a frequency resolution of 0.1Hz. The scan of the unmodified tank was repeated in experiment 5. Each scan lasted approximately 1 hour.
Follow-up stepped frequency scans were performed in the range 85Hz to 115 Hz with a step size of 1Hz. This range exhibited consistent sensitivity to structural modification. Two experiments were conducted, 5 repeat experiments were performed with the tank in the unmodified state (experiment 4a) and eight repeat experiments were performed with the smaller mass attached at location d'. Location d' was at the same circumferential location as d except the wall height was 1m instead of 0.5m (experiment 4b).
Table 1: Summary of tank wall experiments with actuator in configuration AT2.
|Experiment No||Tank state||Scan resolution (Hz)|
6. Results and discussion
6.1 Repeatability of stepped frequency scans
Experiment 1 and experiment 5 were repeat scans of the tank wall with no modification. Examples of the FRFs for sensors 2 and 4 for each experiment are overlaid in Figure 3 (a) (sensor 1), Figure 4 (a) (sensor 2), Figure 5(a) (sensor 3) and Figure 6(a) (sensor 4). Note that in these figures the 100Hz to 200Hz range is stacked above the 1Hz to 100Hz range with an arbitrary offset. Note also that the repeat scans were taken two weeks apart. Between the first and second experiment, the actuator cable was disassembled and then reassembled and re-tensioned. These scans all indicate good qualitative agreement between repeat experiments. There does not appear to be any difference in the repeatability between the upper and lower frequency ranges. The closest agreement appears to be in the frequencies of the modal peaks.
Fig. 6. Tank wall FRFs at Sensor 4 by stepped frequency scan (0.1 Hz step)
The amplitudes of the peaks are noticeably more variable, although not excessively so. However it is worth pointing out an exception to this. A prominent mode appears at 97 Hz in the FRF of sensor 2 for one experiment but not the other. This is discussed below in relation to results from the 1Hz resolution scans.
At 1Hz resolution the scans appear smoother than the preceding scans at 0.1Hz but no important information is lost. This means that future applications of the PD technology to tanks could be completed 10 times faster (the 200Hzrange could be scanned in 6 minutes instead of 1 hour).
Figure 7(a) shows the low resolution FRFs of the unmodified tank. These show good agreement with the scans taken at 0.1Hz resolution 120 days earlier. The repeatability, illustrated by the proximity of the overlaid scans, is excellent for these scans which were all taken on the same day. However, the repeatability is noticeably poorer for two modes either side of a prominent mode at 103Hz at sensor 2. These modes exhibit frequency variability which is visibly higher than in the other modes (frequency variability is practically absent in other modes). One of the modes corresponds with the mode that appeared on only one of the repeat scans taken at 0.1Hz resolution, as mentioned above(see Figure 4(b)). The different characteristics of these modes suggest that they may be physically different to the other modes. However, further experimentation is required to establish with certainty the nature of these modes.
Fig. 7. Tank wall FRF taken by stepped frequency scan (1Hz step size)
The significance of the observed repeatability will depend on how it compares with the sensitivity of the method to added masses. This is the subject of the next section.
6.1.1 Sensitivity to added masses
Experiment 3, experiment 5 and experiment 8 were scans of the tank wall with modification 'W(a)', 'W(b)' and 'W(d)' respectively. The FRFs for these experiments are overlaid in Figure 3(b) (sensor 1), Figure 4(b) (sensor 2), Figure 5(b) (sensor 3) and Figure 6(b) (sensor 4). Note that in these figures the 100Hz to 200Hz range is stacked above the 1Hz to 100Hz range with an arbitrary offset. Note also that all three scans were taken over a period one week.
The first point to note is that, below 100Hz, there is very little difference between the scans for the modified and unmodified tank (with the single exception of the 90Hz mode at sensor 2 when the tank had modification W(a)). In other words there is generally no sensitivity to structural change below 100Hz. However, between 100Hz and 200Hz there is clearly more difference between the modified and unmodified states, i.e. the higher frequency range has better sensitivity to structural change. The most important differences are discussed below.
Between 196Hz and 200Hz there are dramatic indications on the spectra of the tank modified with 'W(b)' and 'W(d)' on all sensors but particularly sensor 2. A prominent growth in modal amplitude, and a slight frequency shift, occurs at sensor 1 (tank wall states W(b) and W(d)), sensor 2 (wall states W(b) and W(d) and sensor3 (wall states W(a) and W(b)). Only sensor 4 is not sensitive to any of the added masses in this region of the spectrum. In some cases, most notably sensor 2, the change in amplitude is very pronounced. The modal amplitude at this region of the spectrum for states W(b) and W(d) are 12 times the amplitude in the unmodified state.
Inspection of the 100Hz to 200Hz frequency band reveals that, while the spectra of the tank wall in states W(b) and W(d) are similar to each other, that of state W(a) differs noticeably from both of them. In fact the spectrum of W(a) is very similar to that of the unmodified tank, suggesting that there was less sensitivity to this modification than for states W(b) and W(d). There is no obvious explanation for this since all modifications are in the same position relative to the actuator cable attachment points (see Figure 2). However state W(a) does show some notable indications. For example, between the modes at approximately 83Hz and 103Hz there is a subtle difference between the spectra of the unmodified tank and the tank with the modification 'W(a)' (see Figure 4). Apparently a pair of modal peaks detected on sensor 2 slightly above 90 Hz on the modified tank have slightly shifted frequency and widened (the second of these peaks is interrupted at the break between the two graphs). Indications also appear at about 110Hz in the sensor 2 spectrum of the tank with modifications 'W(b)' and 'W(d)' (see Figure 4). Despite being less dramatic than the modal amplitude increases between 196Hz and 200Hz, these spectral shifts are nonetheless, consistent in that all of the tank modifications produced a change in the this region of the FRF.
The lower resolution (1Hz) scan between 85Hz and 115Hz was conducted to confirm this sensitivity to structural change (using a smaller 0.89kG attached mass) in this region of the spectrum. These scans cannot be directly compared with the 0.1Hz scans because the mass, location and method of attachment of the modification was different. However, the mass and location were similar and therefore the damage sensitivity of observed with the larger mass would be expected to be comparable with that obtained with the smaller mass.
Figure 7(b) shows that the FRFs of sensor 1, sensor 3 and sensor 4 are most prominently changed by the presence of the added mass. The most obvious changes are in the modal amplitudes which appeared to be related to the torque applied to the clamp attaching the mass to the tank wall. The greater the applied torque the greater was the modal amplitudes. The effect on sensor 2 was much prominent. This may have been influenced by its close proximity to the actuator attachment point suggesting that damage sensitivity is reduced as the sensor is brought closer to the actuator attachment point. Less obvious but equally important is the slight reduction in the frequency of the mode at 115Hz.Sensor 3 on Figure 7 shows this mode is downshifted by about 1Hz by the added mass. Although only a small shift it is nevertheless has significance because of the good frequency repeatability observed at this mode. A similar frequency shift occurs on sensor 1 at the 115Hz mode and at the 93Hz mode, both of which are downshifted by approximately Hz. In fact it does appear that other modes are downshifted slightly, although some by less than 1Hz. The reason that this was not observed with the larger mass was probably because it was attached at mid-wall height rather than at the top of the tank as was the smaller mass. Finally, there are two spectral peaks that occur at 86Hz on sensor 1 and sensor 3of the modified tank. These were observed on only one of the eight repeat scans. This peak might therefore be considered anomalous, the one possible explanation is that this is a repeatability outlier from the mode at 82Hz although this would not seem likely given that frequency repeatability on this sensor does not appear to be more than 1Hz.
The method of using active-tendon coupled piezoelectric excitation to generate and detect the modal response of a tank wall (so called PD technology) has been shown to produce repeatable and consistent FRFs, the modal frequencies in particular. There does not appear to be any frequency dependence of repeatability below 200Hz. The sensitivity to structural change, unlike the repeatability, did appear to be frequency dependent. The upper 100Hz to 200Hz frequency range was more sensitive than the lower 1Hz to 100Hz range. In most cases changes in the state of the tank wall were represented by differences the spectral amplitude (most notably amplitude increases of more than an order of magnitude in some cases). However, The sensitivity was dependent on attached mass location and it was found that the test was more sensitive the nearer the structural modification was to the top of the tank. In this position modal frequency shifts were also detected in spectra of the modified tank wall. Despite the design symmetry, different circumferential locations of the modification produced different modal properties suggesting the possibility of damage location.
- It has been shown to be possible to obtain the modal properties of a model storage tank using piezoelectric transducers (PD technology).
- PD technology has detected structural modifications on the wall of a model storage tank comparable with that of localised corrosion, demonstrating the potential of this method for corrosion monitoring on full sized storage tanks.
- EC project G1RD-2001-00659 (PIEZODIAGNOSTICS) Project report (23/03/04). Technical appendices.
- Sokolkin A V, Ievlev I Yu and Cholakh S. O: 'Use of Acoustic Emission in Testing Bottoms of Welded Vertical Tanks for Oil and Oil Derivatives'. Russian Journal of Nondestructive Testing, volume 38, number 12, pages 902-908, December 2002.
- P.J. van de Loo - Shell, The Netherlands. B. Herrmann - Dow, Germany: 'How Reliable is Acoustic Emission? The Quantified Results of an AE User group Correlation Study!'. NDT.net - February 1999, Vol.4 No.2.
- Ewins D J: 'Model testing: Theory and practice'. Publ: Research Studies Press Ltd, Taunton, Somerset, England, 1984.
- Bruel and Kjaer type 8001 impedance head'. Publ: Bruel and Kjaer, DK-2850, Naerum, Denmark. Revision September 1982. English BE 0500-12.