Structural Health Monitoring of CFRP Structures - can Artificial Intelligence be the Key?

A constant challenge for the design and operation of CFRP primary structures is their sensitivity towards impact loads (e. g. gravel impact, tool drop impact etc.). Especially, if the load is perpendicular to the laminate plane, the resulting forces between the layers can lead to delaminations formations and fibre breakages not always visible from the outside. These defects weaken the structure and thus pose a risk for the user. In practice, this circumstance is often countered by regular inspections or conservative design. Structural Heath Monitoring (SHM) systems are an alternative. Here, the structure is equipped with sensors for the purpose of permanent monitoring. Thus, the user is informed about critical condition changes at an early stage. This approach is becoming more and more attractive in the age of digitalization, in which the necessary computing power and storage capacity are available for these systems. Not only maintenance costs could be saved, but also additional lightweight potential could be gained with such a system.

Despite these advantages, however, systems like that are not yet being commonly used in the field of fibre reinforced composites. Currently, biggest market entry barriers are system costs and reliability. Only if the user can rely 100% on the statements of the system, the above mentioned advantages can be exploited. The challenge for a SHM systems developer is to demonstrate this reliability economically. 

Within the framework of the Listen2theSOURCE project, IVW has faced the task of developing a demonstrator with which damaging events can be imitated and reliably located in a complex CFRP structure. The detection is based on the resulting acoustic emission of the damaging events. These events propagate in form of waves in the structure and can be detected via piezoelectric sensors. By evaluating the differences in arrival times between the sensors, a statement about the position of the source or the damaging event can be made.

The challenge of source localization in thin-walled CFRP structures lies in the consideration of wave dispersion, anisotropic material properties and variable component geometry. In this project, this complexity is captured by the training of a neural network. For this purpose, artificial acoustic emission sources (pencil lead breaks) which reproduce sound emissions of typical damaging events in frequency and modal content are used (see Figure 1 as an example). 

The structure to be monitored consists of an omega profile equipped with a network of piezoelectric sensors designed for reliable localization within a defined window. Signal processing is performed on a single board computer (e.g. Raspberry Pi), which completes the measurement chain together with a digital oscilloscope, shown in Figure 2.

For the training of the network a total of 15 points (grey) at which pencil lead breaks were carried out were marked on the profile at 50 mm intervals in order to record the resulting differences in arrival times between the four installed sensors (see Figure 4). Using the known (x, y) positions of the pencil break and the arrival time differences of the six sensor pairs, the feed-forward network with two hidden layers - shown in Figure 3 - can be trained. 

For demonstrator validating, five further test points (red) which the network had not yet seen during the training were defined. The deviations of the localization results (target and actual position) in the training and test data are shown in Figure 4, using the black lines. In total, the artificial emission sources could be localized with an average error of approx. 10 mm within the 100 x 200 mm zone. With these initial results, the limits of artificial intelligence for this application will be examined in more detail using the demonstrator.

The project "Listen2theSOURCE - Development of Measuring and Evaluation Modules for theIdentification of Damaging Events in Fiber Reinforced Plastics via Acoustic Emission Analysis" was funded by the Federal Ministry of Economics and Energy on the basis of a resolution of the German Bundestag (funding code ZF4052302WM5).

Further information:
Dipl.-Ing. Benjamin Kelkel
Institut für Verbundwerkstoffe GmbH
Tailored & Smart Composites
Erwin-Schrödinger-Straße 58
67663 Kaiserslautern
Phone: +49 631 2017-318

[Translate to English:]

Figure 1: Resulting time signal (a) and time-frequency signal in wavelet representation (b) after excitation by pencil lead break

[Translate to English:]

Figure 2: Hardware components of the demonstrator for the localization of artificial emission sources (pencil lead break) on a CFRP structure

[Translate to English:]

Figure 3: Feed-forward network with two hidden layers for determining the position of artificial emission sources (output) based on 6 arrival time differences (input)

[Translate to English:]

Figure 4: Comparison of real training (grey squares) and test data points (red diamonds) with the points predicted by the network. Deviations between target and actual positions are indicated by black lines.

↰ News