Acoustic Emission Analysis of fiber-reinforced high-performance materials

Elisabeth Anne Balz, winter semester 2020/2021

Sound Emission Analysis is a quasi-nondestructive passive method to capture the initiation and growth of cracks in solid materials. The test specimen is equipped with multiple sensors in order to detect the resulting mechanical (sound) waves. As it is only possible to detect the initiation and growth of ruptures, this method is only suitable for in-situ testing. Upon degradation of the material, the occurrence of sound emissions will increase. Thus, the count rate of sound emissions indicates the damage progression of the specimen. Also, weak spots of the structure can be detected via sound emission localization.

Figure 1 shows the experimental setup for sound emission analysis. On the surface of the specimen (grey), several sensors (blue) are attached. These sensors can detect the wave signals (circles) of a crack (yellow bolt).

1 Basic principles

1.1 Failure mechanisms

Cracks are caused by internal stresses in the material. These stresses can have various causes. Sound emission analysis is usually performed on mechanically loaded specimen, e. g. in a three-point bending test (see below). Yet, the internal stresses may also be provoked by thermal and chemical effects such as repeated freezing and thawing or corrosion.

In fiber reinforced materials, the fibers predominantly bear the tensile loads, whereas the matrix carries the compression loads. The tensile stiffness of the fibers is usually higher than the stiffness of the matrix. Typical failure mechanisms in fiber reinforced materials are

• rupture of the fibers due to high tensile loads in direction of fiber
• inter-fiber failure (microcracking of the matrix) due to tensile/compression loads across the fiber or shear loads
• fiber pull-out due to weak bonding.

These failure mechanisms can be provoked in special tests. The sound emissions may be used to quantify the damage accumulation. Material properties can be compared for different specimens. The sound emissions may also be used to sample characteristic sound signals of specific failure mechanisms. Figure 2 gives examples of tests for fiber-reinforced materials.

1.2 Sound emissions of cracks

Cracks always cause tremors in the material. Elastic energy is (amongst other forms of energy, e. g. heat) converted to surface energy and sound energy (Werner, 2017). This conversion is irreversible and not reproducible (Große, 2020). The solid-borne sound can be detected by the sensor network on the surface of the specimen. In homogeneous materials, the propagation of the sound signal occurs in a straight line. Due to geometrical spreading and dissipation, the signal strength decreases with increasing distance from the origin. Nevertheless, simple localization techniques can be found for homogeneous materials. In nonhomogeneous materials on the other hand, the solid-borne sound can be deflected by surfaces and interfaces such as fibers or aggregates. Already existing cracks may also shadow new cracks from the sensor. This complicates the analysis of the signals and a simple localization method cannot be provided.

1.3 Sensors

There is a wide range of sensors suitable for detecting sound emissions. The sensors most commonly used are PZT piezoelectric sensors. Usually they are operated in resonance or multi-resonance (Grosse & Ohtsu). Another possibility to pick the signals is a LASER-system. On account of contactless measurement there are various advantages over piezoelectric sensors which for example are not suitable for high temperatures. Yet, LASER-systems are less sensitive than piezoelectric sensors (Grosse & Ohtsu, 2008).

2 Localization techniques

2.1 Hyperbolae localization technique (homogeneous materials)

The differences in the time of arrival of crack noises may be used for crack localization. Of course, synchronous taping is needed.

For zonal localization, it is assumed that the source of the solid-borne wave (= crack) must be somewhere near the sensor that detects it first. This so called 1^st hit method can be applied on quasi-1-dimensional objects (e. g. tubes) as well as on quasi-2-dimensional objects (e. g. sheets). The 1^st hit method is not suitable for an exact localization.

Two-dimensional (planar) localization can be achieved using three sensors in a plain. It is essential that all three sensors can detect the signal. Dissipation and geometrical spreading must not dampen the signal so as to fall below the detection threshold. For this technique, homogeneous and isotropic material is assumed. Due to the high number of sound emission events and multiple sensor channels, for this technique an automated onset detection is essential (Grosse & Ohtsu, 2008).

In Figure 2 an exemplary analysis is shown. Each sensor (A, B, C) detects the signal at its own instant of time. For three sensors there are three time spans (delayAB, delayAC, delayBC) giving the differences of arrival time. Each delay time gives a hyperbola between the corresponding sensor locations (by definition, a hyperbola is a line of constant difference). For ideal conditions, the three hyperbolae meet at one point which is the source of the signal (red circle in Figure 2). More than three sensors can be used to statistically improve the localization on non-perfect conditions. However, a unique solution is no longer possible.

This technique can be adapted for three dimensional problems. Here, four sensors are needed. The sensors must not be placed in one plain; otherwise no new information would be achieved, but only redundant information.

2.2 Localization techniques taking non-homogeneous and anisotropic materials into account

One approach for improved localization of sound sources is the use of artificial neural networks (Kalafat & Sause, 2015). These “supervised learning” processes utilize representative training data for modelling the material transfer function. Consequently, non-homogeneous and anisotropic materials can be analyzed as well as homogeneous ones. The training data may be gained using different test sources such as Hsu-Nielsen source (pencil lead break) or a piezoelectric pulser. Once the model is established it can be used to localize real ruptures in the structure.

Another, more theoretical, approach is creating a finite element model of the structure. This model can simulate local differences in material parameters such as sound velocity. A homogenized (“smeared”) anisotropic material behavior may be considered as well as a very detailed multi-scale model. These models can be used to reconstruct signal runtimes more accurately than the geometry-based hyperbola method (see above).

3 Crack type identification techniques

The signal of each sensor in the network is recorded. Whether or not the complete signal is stored or only a few parameters are extracted, it is differentiated between parameter-based methods and signal-based methods.

3.1 Parameter-based identification techniques

For parameter-based methods, insightful variables (e. g. arrival time, rise time, signal duration, max. amplitude) are gained form the signal in real-time. These parameters are stored and later analyzed, whereas the complete signal is not saved and thus not available for further examination. Examples for these parameters may be taken from Figure 3. This method obviously needs less memory capacity than storing the whole signal. Also, a higher number of events can be recorded as compared to signal-based approaches (Grosse & Ohtsu, 2008).

3.2 Signal-based identification techniques

For signal-based methods, the complete signal is stored and analyzed. No severe data reduction (except time discrete and value discrete sampling) is performed before analyzing the signal. From wavelet form and duration, it can be concluded which crack type is at hand. A detailed analysis, e. g. similarity measure or frequency analysis, can be performed. Also, the directional characteristic may be considered.

Similarity measures are used to assign wavelet forms to fracture mechanical happenings. Sound emission signals are compared to one another. Similarities in the frequency content signalize a similar source of the signal. Similarity can be measured by the so-called coherence sum, which is the integral over the coherence spectrum:

$//<![CDATA[ \begin{array}{l}\^{C}_{xy} (f_{min},f_{max})=\frac{1}{f_{min}-f_{max}} \int_{f_{min}}^{f_{max}} \frac{|G_{xy} |^2}{(G_{xx} (f)\cdot G_{yy} (f)) } df\end{array} //]]>$

where $//<![CDATA[ \begin{array}{l}f\end{array} //]]>$ denotes to the frequency, $//<![CDATA[ \begin{array}{l}G_{xx}\end{array} //]]>$  and $//<![CDATA[ \begin{array}{l}G_{yy}\end{array} //]]>$ the auto spectral density and $//<![CDATA[ \begin{array}{l}G_{xy}\end{array} //]]>$ the cross spectral density (Grosse & Ohtsu, 2008).

Values of $//<![CDATA[ \begin{array}{l}\^{C}_{xy}\end{array} //]]>$  greater than 0.6 are said to have “similar” sound sources and similar travel paths (Grosse, 1996).

The directional characteristic considers the polarity of the first deflection of the signal. This denotes to either a compression or an elongation of the specimen in the location of the sensor. Considering more sensors, the crack type can be derived from this information (Große, 2020). In Figure 4 an exemplary directional characteristic is shown. The arrows indicate the movement of the crack edges (here: shear crack). In this case, the fracture surface is vertical in between the arrows and perpendicular to the plane of projection. The dotted area shows the part of the specimen that will be compressed due to the crack formation whereas the non-dotted area shows the part that will be elongated. The wavelets on the right show an exemplary sensor signal of a sensor in this area. Note that the first deflections of the wavelet signals go to the opposite direction.

As already mentioned in the localization chapter, pattern recognition techniques such as machine learning are also used for identification of crack types.

References

Grosse, C. (1996). Quantitative zerstörungsfreie Prüfung von Baustoffen mittels Schallemissionsanalyse und Ultraschall [Dissertation]. Universität Stuttgart.

Grosse, C., & Ohtsu, M. (2008). Acoustic Emission Testing. Springer Berlin Heidelberg. https://doi.org/10.1007/978-3-540-69972-9

Große, C. U. (2020). Einführung in die Zerstörungsfreie Prüfung im Ingenieurwesen: Grundlagen und Anwendungsbeispiele. Vorlesungsbegleitendes Skriptum. Lehrstuhl für Zerstörungsfreie Prüfung.

Kalafat, S., & Sause, M. G. R. (2015). Acoustic emission source localization by artificial neural networks. Structural Health Monitoring: An International Journal, 14(6), 633–647. https://doi.org/10.1177/1475921715607408

Werner, E. (2017). Werkstoffkunde 1: vorlesungsbegleitendes Skriptum. Fakultät für Maschinenwesen, Lehrstuhl für Werkstoffe und Werkstoffmechanik.