Ultrasound is already widely used for inspection in the aircraft industry and is a robust, rugged technology that could be used on the factory floor. However, there are problems associated with the inspection of automotive structures that would make some methods difficult to apply, the main problem being that of variable bondline dimension. Low frequency techniques which track through bond resonances are widely used in the aircraft industry and rely on the fact that the resonance will shift in the presence of a disbond. In an automotive structure, where bondlines can easily vary between 0.5 mm to 3.0 mm over very short spaces, it becomes extremely difficult to separate changes in through bond resonance caused by varying bondline thickness from those caused by disbonding. For this reason the authors have pursued a time domain pulse echo method with the aim of identifying echoes of diagnostic value that can be anlaysed when the structure dimensions are not known. The potential for quantitative NDT is greatly increased since it is possible to assess the bondlines thickness, detect gross disbonding and estimate the acoustic impedance of the adhesive. While the application of pulse echo ultrasound to adhesive joints is not new, there are specific problems that need to be addressed when considering NDT of automotive structures. This paper will describe the method, its potential, and the practical problems of pulse echo time domain NDT of automotive structures.
Figure 1. Principal echoes excited in a three layer adhesively bonded joint.
The main echoes of diagnostic interest are the front face echo (A) which is sensitive to coupling and the presence of paint layers. The period of the top substrate reverberations (B1, B2 etc) are a simple measure of the substrate thickness and the decay rate is sensitive to the acoustic impedance behind the substrate, and hence can be used to detect disbonding. The final component of interest is the bottom substrate echo (C), where the arrival time can be used to measure the bondline thickness. The phase of this component is sensitive to the acoustic impedance behind the adhesive layer and will reverse in the presence of disbonding.
In practice the interpretation of echoes excited in the structure is extremely difficult due to attenuation in the adhesive layer, pulse overlapping and the effects of coupling and paint layers. In order to investigate these issues a multilayer wave propagation model has been employed [1,2,3] to simulate the acoustic response of an adhesive joint. The model uses a system of matrices to describe any number of parallel layers which can support compression and shear wave propagation with complex wavenumbers. The use of a complex wavenumber permits the effects of absorption in the adhesive layer to be included in the simulation. Whilst the model is limited to parallel layers and assumes a plane wave source of infinite extent it has been a invaluable tool in addressing the practical problems of NDT of adhesive bonded automotive components. Figure 2a shows the simulated unit impulse response for 1 mm thick aluminium plates bonded with 1 mm of adhesive. The components A, B1,B2 etc, can be identified and the effect of absorption in the adhesive can clearly be seen in the dispersed shape of the bondline component C. Figure 2b shows the same simulation with the bottom substrate completely disbonded and the change of phase of component C is observed. Figure 2c shows the simple case when the top substrate is completely disbonded, note the absence of any bondline echoes and the change in decay rate of the substrate reverberations indicating loss of adhesive contact with the top substrate.
Figure 2a. Simulated unit impulse for a 1 mm aluminium sheets bonded with a 1 mm adhesive bondline.
Figure 2b. Simulation of figure 2a with a bottom substrate void disbond.
Figure 2c. Simulation of figure 2a with a top substrate void disbond.
There have been many publications in the past that have dealt with practical measurements on adhesive bonds [4,5] and the analysis of the echoes described in the above simulations and this work established the principal of using pulse echo ultrasound. However the limiting factor in many of the cases described has been that thick substrates and thin bondlines have been used, thus permitting easy isolation and analysis of the echoes excited in the structure. In automotive applications sheet steel with a typical thickness of 0.9 mm is used and in order to isolate echoes, well damped transducers operating with center frequencies in excess of 20 MHz would be required. However, due to excessive absorption in the adhesive layer, which may be two or three times thicker than the adherends, it is impossible to propagate high frequency signals across the bondline for feature extraction. Thus a compromise exists between signal amplitude and echo resolution. Based on measurements of the absorption in many commercial adhesives [6,7] and considering typical joint dimensions, experiments and simulations have indicated that the highest center frequency transducer that can be used without incurring poor signal to noise is 10 MHz. This then defines the main problem, that of pulse overlap and interference in the thin adherends. This is illustrated in figure 3 where the simulation of figure 2a is now convolved with the response of 10 MHz transducer.
Figure 3. Simulation of figure 2a after convolution with the response of a 10 MHZ transducer.
It can be seen that it is now very difficult to isolate and gate the signals for amplitude and position information. Many workers have considered this issue and a range of solutions have been suggested. Kinra  proposed a inverse solution where a model of the joint response is fitted to measured data. This requires some knowledge of either the velocity or thickness for one or more of the layers but solutions are possible. Others [9,10] have concentrated on trying to improve the resolution of the signal components by deconvolution of the transducer response. Some methods can greatly improve the resolution but require significant computation effort and introduce computational noise which can mask low amplitude bondline components. In this paper the authors have developed a technique which overcomes the pulse overlap problem and permits the analysis of the bondline echoes.
y[n]=x[n] - R x[n - P] Equation 1.
where x[n] is the input data, y[n] is the filtered data, P is the pole spacing (reverberation time) and R is the pole radius (reverberation decay). The adaption is based on the root mean square (rms) of the filtered data and on completion will return the values of P and R for the substrate in question. This data is used to calculate the sheet thickness and the acoustic impedance of the material behind the sheet. The R term is very sensitive to top substrate disbonding and is used in the bond analysis. After filtering the data should only contain the bondline echoes which can be processed using conventional pulse location and gating methods. Figure 4 shows the result of applying the filter to the data of figure 3.
Figure 4. Data of figure 3 after application of the filter.
It can be seen that the reverberations are removed and that the bondline component C is now visible. Typical data processing times are less than 100 ms on a 486 100 MHz pc.
Figure 5a. Measured raw and unfiltered data for 1 mm aluminium sheets bonded with 1.2 mm adhesive.
Figure 5b. As figure 5a but with a bottom substrate disbond present.
It can be seen that in both cases after filtering the bondline components C are visible. There are other components that are not related to the bondline or top substrate reverberations which are also presented. These `residual' components are compression and shear modes excited by the transducer edge wave source Weight . These components become more evident as the substrate thickness is reduced and can make analysis of the bondline echo difficult particularly when the components are small in amplitude. Many factors effect the ability to to locate the bondline component and test for disbonding and these are briefly considered in the following sections.
Limits of testability for a bottom substrate disbond for steel substrates
Table 6b. |
As table 6a but for aluminium substrates.
(ii) Adhesive absorption
Absorption in the adhesive can severely affect the ability to detect echoes that have travelled across the bondline. In most cases it is extremely difficult to detect bottom substrate disbonding when low modulus, rubbery adhesives are used. Fortunately, high modulus structural adhesives have significantly lower absorption and detection of the bondline echo is possible. To illustrate the difference in absorption, figure 7 shows the compression wave absorption coefficient as a function of frequency for a typical high and low modulus adhesive respectively.
Figure 7. Absorption coefficient for a high modulus structural adhesive (Ciba-Geigy Araldite AV119) and for a low modulus rubber adhesive (Elastosol M51).
(iii) Paint coatings
The observed effect of paint coatings was the alteration of the decay rate of the top substrate reverberations and the amplitude of the front face echo (A). Despite this it was possible to detect the change in decay due to disbonding. Figure 8a to 8d shows the echo response in an unpainted aluminium 1.6 mm sheet, and the sheet with paint coatings of thickness 10,54 and 81 microns respectively.
Figure 8a. Time domain response of a 1.6 mm aluminium sheet (unpainted).
Figure 8b. As figure 8a but with a 10 micron paint coating.
Figure 8c.As figure 8a but with a 54 micron paint coating.
Figure 8d.As figure 8a but with a 81 micron paint coating.
For very thick paint coatings extra residual components were generated by the filtering process making analysis of weak bondline echoes more difficult. However in these cases it was still possible to detect top substrate disbonding.
The effects of transducer coupling were investigated and it was discovered that variations in coupling had no effect on the ability to detect the bottom substrate disbond but did introduce small variations in the measured decay rate of the top substrate reverberations. These would be significant if sensitive measurements of decay rate were required, ie to measure the acoustic impedance of the adhesive.
Current work is moving towards the development of lower frequency techniques to overcome the problem of excess absorption in the adhesive and the design of scanning systems suitable for implementation in an automotive body production plant.
The Paper was presented on the UTonline Application Workshop in May '97
|| UTonline ||