ABSTRACT

Acoustic emission (AE) measurements at tensile loaded specimens exhibit complicated signatures of the registered burst type signals. To draw conclusions from the signal form to the dynamics at the source, information is necessary about the transfer function of the AE propagating from source to transducer. It is assumed that the geometry of the specimen influences the signal as well as the type of source (e.g. friction or fracture). Geometrical effects are simulated by a computer model which constructs the transducer response to AE point sources. The effect of measuring systems parameters on the further signal processing is investigated and the possibilities of signal evaluation for source characterisation are discussed.

KEYWORDS

AE signal, geometrical effects, AE source characteristics

1. INTRODUCTION

It seems to be clear that the signal parameter analysis is more efficient for information gain than only the registration of signals. So, if you only record the acoustic emission as counts it is possible to say that something happens. But if you record the times of arrival of the signals at different transducers, you are able to calculate the location of the source. That means you can say not only that something happens but also where it happens.

To get more information it is necessary to put some conditions into the system, which have no relation to the monitored processes and which are arbitrary chosen. In our example of evaluation by location the most important condition is the threshold level. It can be chosen depended on the noise level of the environment or on economical conditions like the ratio between the necessarity of safety and the costs of false alarm. But the choice of the noise level is independent from the causes of acoustic emission within the specimen. Nevertheless the estimation of the times of arrival is strongly influenced by the threshold level, especially, when the signature is complicated and the signal has long rise times. Fig. 1 shows the arrival time differences for a signal measured at two transducers at a CT-specimen [1] when the threshold is varied. It can be seen how the threshold changes the time differences so that even the order of the channels may be reversed.

Fig. 1. Dependence of arrival time differences from threshold value

In front of this background we can formulate the problem as follows:

Evaluation of signal parameters gives the possibility to get more information about the processes in the test object. But so more signal parameters are registered, so more extern conditions must be postulated. The postulated conditions must be chosen so, that characteristic signal properties stay independent from changes of these conditions and that wrong conclusions in categorising the signals are impossible. When a signal cannot be related to good (real physical effect) or bad (fault, noise) it cannot be taken into account .

In acoustic emission this problem stands with particular sharpness. On the one hand the causal chain from mechanical stimulation to the measured signal is longer than in other test methods and a lot of signal modifications are to be considered. On the other hand there is no strong reproducibility for acoustic emission signals, because acoustic emission indicates changes and so the original conditions for an acoustic emission event cannot be created once again.

Therefore at first is to be considered how the measured signal influenced by environmental and measuring conditions. It is done in the second chapter.

To examine the influence of the right choose of signal parameters it is useful to make a numerical simulation. Then reproducibility and variation of measuring conditions is no problem. This numerical simulation model is introduced in the third chapter.

In the fourth chapter the comparison of real and simulated signals is considered. The usefulness of the simulation model and further applications are discussed.

2. MEASURING CHAIN

Acoustic emission measurements are done to test the integrity of structures. These structures are loaded and so acoustic emission is stimulated. After the registration and evaluation of the signals a decision on the integrity or the necessarity of further investigations on the base of other test methods must be made.

Fig.2 shows the way from stimulation to decision. The result is influenced by the given objective conditions marked with crosses and the arbitrary applied conditions marked with lines.

Fig. 2. Coinditions influencing the result of acoustic emission measurements

Already the kind of stimulation is important. If the load rises up very quickly so the probability is high to register only some few strong events which are difficult to separate from one another. If the load rises up very slowly then relaxation processes may prevent acoustic emission. So the dynamics of loading influences the result.

But here it is assumed that loading corresponds to the normal working conditions of the investigated structure and leads to single acoustic emission events.

The acoustic emission source sends pressure pulses through the structure to the surface mounted transducers. The source characteristics as well as the object geometry determine the surface motion. These conditions are given and they are not to be influenced by measuring conditions.

The location of sensor mounting is important for the registered signal form. So wider the distance to the source so more the signal is influenced by reflections, mode conversions, dispersion and damping. The sensor type considerably determines the voltage output produced by the surface motion. If there is not a linear transfer function it is difficult to get unambiguous relations between voltage and surface motion, especially in the case of resonance transducers [2].

The voltage signals arrive in the measuring system. The parameters like amplification gain and threshold are important for the possibility to register acoustic emission or not. Therefore the threshold level is the most important parameter of the measuring system. But it is very difficult to find objective methods to determine its value. If the signal form is to be investigated the signal must be stored. The length of the memory limits the signal length and the sample rate must correspond to the signals frequency content. The stored signals cannot be stored for a long time because of the limited memory capacity. It is necessary to make an extract of signal properties which enable a decision about their origin. At least from these signal parameters and their evolution in time is drawn the conclusion about the integrity of the tested structure.

It can be seen that the signal form is strongly changed and its way from source to transducer area and to the measuring system. There are two mean methods to get information about the source from registered surface motion.

The first is the deterministic method. Using the Greens function the dependence

with

            G_ji dynamical Greens function of the structure related to the j-th source at location r² and i-th transducer at r ,

            * time convolution,

            f_j source function,

            V₀ source volume,

gives the necessary relation [3]. Inverse methods allow to determine the source function [4]. But this method is difficult to use for complicated structures and requires point-like transducers with flat transfer functions. The influence of the measuring system must be considered otherwise.

The second method uses the pattern recognition. If typical signal forms are known for a determined structure and a determined measuring system the measured signals can be assigned to the patterns and categorised. But changes in the measuring conditions lead to complete new patterns, it is not possible to test the influence of the change of one only measuring condition, because test object and measuring system are taken as a whole black box.

To investigate especially the influence of the extern arbitrary defined conditions a model for the numerical simulation of signal construction was made. The aim of this model is the production of signals similar in the signature to real acoustic emission. With these signals the influence of the measuring systems could be investigated.

3. NUMERICAL SIMULATION MODEL

The test object has the shape of a flat CT-specimen taken only in two dimensions. The source may be determined as a point anywhere within this area. The signal is given by a quickly accelerated and then slowly decelerated movement of the source point with a total duration of about 150 ns. This movement is either isotropic to all directions within the specimen or defined direction dependences can be introduced. The movement propagates to the surface and is influenced by interferences and reflections at the surface.

To regard these effects single rays are considered. The particle movement propagates along this ray until it is reflected at the surface and so on. When it arrives at the sensitive surface part, at which the transducer is localised, the signal is stored at a time according to the length of the ray from source to sensor. Fig. 3 shows one example of the constructed rays within the two-dimensional specimen.

Fig. 3. Ray from the source reaching sensor 2 one time

The damping is assumed to be about 1.3 dB/m [4]. So the signal amplitude decreases with growing ray length. It is assumed that there are only longitudinal signals with total reflections. Mode conversions or dispersion are vanished. Other ray directions lead to other times of arrival of the signal. If all possible rays are considered, all the signal parts arriving at the sensor area at different times are added and a sum signal is constructed which takes into account reflections, damping and interferences. 200 us after the start from the source the signal registration time ends. In this time the ray path length is about 1 m within the 75 x 72 mm2 specimen. The movement of the surface gives rise to a voltage at the sensor output The sensor may be linear, then the voltage has the same time dependence as the mechanical stimulation at the sensor area. Or it is a resonant one, then the output signal is calculated using the differential equation for oscillators. Damping and resonance frequency of the sensor must be known. The equation is solved approximately with time differences of 25 ns. Noise can be added using a random function. So an 8 kByte long signal with 8 bit resolution and 25 ns sample time is constructed. This signal corresponds to the transient recording of the electrical sensor output produced by real acoustic emission hits. The signal trigger is determined by the time of signal start at the source in contrast to real measuring systems, where the trigger is determined by the registered signal.

4. RESULTS

The model was tested with source locations near the bolts where friction signals are expected and near the crack tip, where acoustic emission from crack growth is expected. The sensors were located symmetrically to the crack tip. The same configuration was used in real experiments [5]. So a comparison of the constructed signals to real measured signals was possible.

First the numerical stability of the model was tested. Different numerical sizes were used and the number of rays per degree was varied. The numerical stability was given. Of course, it is necessary to consider a big number of rays to construct only one signal. It was found, that steps of 1/8 degree for the emission direction are small enough. Then the signal is constructed from 2880 rays.

The next test of the model referred to the similarity of the constructed signals to real acoustic emission. A CT-Specimen under load was monitored using a five channel SPARTAN AT equipment (PAC, U.S.A.) [6] and a two channel transient recorder PC-Scope (IMTEC, Germany) with resolution 8 bit, sampling rate 200 ns and storage depth 32 kByte [7]. So localisation and signature analysis were possible. The load was manual increased in such a manner that the signal rate was low enough for the measuring systems capacity for 95% of the measuring time. The transducers had a resonance frequency of 150 kHz. Examples for acoustic emission from crack propagation processes and friction are shown in fig. 4 and 5 respectively. The assignment to the source types was made by using the measured arrival time differences. These results can be compared to the simulated signals from crack tip (fig. 6) and bolt region (fig. 7). Although the same source functions and isotropic propagation for both sources were used, the localisation of the sensors relative to the source lead to different signal types. The signatures of the constructed signals are quite similar to those of the real acoustic emission.

5. DISCUSSION

From the similarity between simulated and measured signals the simulation seems to be usable for the investigation of the influence of measuring system parameters on the modification of the original signal.

Fig. 4. Simulated crack propagation signal from the crack tip region. Sensor transfer function taken into consideration

Fig. 5. Simulated friction signal from the bolt region, sensor transfer function taken into consideration

Although the signal propagation model is simplified one finds the characteristic properties of acoustic emission signals as result of interferences and reflections in the CT-specimen. The signals from the bolt region show a more complicated behaviour then those from crack tip region in experiment and also in simulation. The reason is the relative position of source and sensors. The more direct connection leads to strong quickly arising signals in the case af crack tip sources. Looking at the signal beginnings it can be seen that it is not unambiguous to define such parameters like rise time or duration. The values for these parameters depend on the threshold and the registration time. If the measuring system is defined with too short registration times and short

Fig. 6. Measured signal from crack tip region, sensor PAC R 15 (150 kHz resonance frequency)

Fig. 7. Measured signal from bolt region, sensor PAC R 15 (150 kHz resonance frequency)

lockout times then one signal may be divided in two single signals. But if one single signal is emitted short after another single signal the overlapping of these two signals may appear as one signal. Such phenomena which could be of interest not only in acoustic emission but also in other signature analysis problems can be investigated with the above introduced simulation model.

Possible misinterpretations of acoustic emission test results according to such false conclusions appear especially if there is no experience about the signal structure from foregoing measurements and if too hard restrictions for the suppression of a part of the measured data are made.

On the other hand, if you have to carry out acoustic emission analysis measurements with always the same conditions the feedback from preceeding results and comparison with results of other test methods can make the evidence of conclusions from acoustic emission analysis more sure. In this field of repeated measurements under nearly constant conditions and with for those conditions adapted settings of the measuring system really appear the advantages of the test method acoustic emission analysis.

6. REFERENCES

[1] G. Mauersberger, R. Eckart, M. Schade, R. v. Schäwen, H.-J. Kühn
      Signalformen von mit Breitband- und Resonanzaufnehmern registrierten Schallemissionsburstsignalen Wiss. Berichte THZ 1269(1990) 24, 84-86

[2] W.I. Ivanov, W.A. Murgazov
      Digitale Modellierung des Durchgangs von Impulssignalen durch akustische Aufnehmer
      Defektoskopija 5(1990), 15-22

[3] Grabec, W.Sachse
      Application of an intelligent signal processing system to acoustic emission analysis
      J. Acoust. Soc. Am. 85(1989)3, 1226-1235

[4] H.Kühnicke
      Beiträge zur Ausbreitung von SE-Signalen
      Habilitationsschrift, TH Magdeburg 1986

[5] R. v. Schäwen, M. Schade, H.-J. Kühn
      Belastungseinrichtung für Bruchmechanikuntersuchungen bei höheren Temperaturen mittels SEA
      Wiss. Berichte THZ 972(1988)18, 98-99

[6] SPARTAN AT, Instruction Manual
      Physical Acoustics Corp., Princeton N.J., USA 1988

[7] Bedienungsanleitung PC -SCOPE T1620/T6420 V 2.30
      Intelligente Meßtechnik GmbH, D-7150 Backnang 1990