Abstract

With fiber composite components, even relatively weak impacts can lead to delaminations. These are often barely visible and have to be detected at great expense. To avoid such damage from being detected only during a maintenance cycle, a fiber composite component can be monitored by means of structural health monitoring and the live data can be evaluated in real time with the aid of artificial intelligence. This can be done by monitoring local accelerations using integrated sensors. If the sensors detect an event that exceeds a predefined acceleration value, the signal curve around this maximum event can be evaluated. For this purpose, the signal course can be converted into a pixel-poor spectrogram, which can then be trained into a convolutional neural network. Previous work has investigated three piezo elements or three MEMS sensors for live monitoring of the acceleration values. Although the different sensor types have a large difference in maximum sampling rate, both were able to achieve very high accuracies in position determination. In this work, the required data quality regarding the sampling rate of the accelerometer and the required recorded time interval around the impact event is investigated and evaluated. It is shown that a sampling rate of less than 1000 acceleration values per second is already sufficient to be able to reliably determine the impact position for the present use case. In addition, it is shown that mainly the first oscillation of the acceleration impact is important for the position determination. The next oscillations have little effect on the quality of the algorism. With these results, both, the amount of data needed for training and the amount of data needed for field use can be defined and thus greatly reduced. The authors wish to acknowledge the funding provided by the Federal Ministry of Education and Research Germany within the Research campus ARENA2036 – Digitaler Fingerabdruck.