Measurement-Based Model of Structural Sound Transmission in a Concrete Specimen

Speaker:
Hinrichs, R.; Leibniz Universität Hannover; Germany

Authors:
Hinrichs, R.; Leibniz Universität Hannover; Germany
Krause, T.; Leibniz Universität Hannover; Germany
Käding, M.; Leibniz Universität Hannover; Germany
Ostermann, J.; Leibniz Universität Hannover; Germany
Marx, S.; Leibniz Universität Hannover; Germany

ID: ECNDT-0148-2018
Download: PDF
Session: Meta Modeling
Room: J2
Date: 2018-06-12
Time: 15:40 - 16:00

In the field of Non-Destructive Testing and Structural Health Monitoring the propagation of sound in solids is used in methods like acoustic emission, ultrasonic testing and vibration based approaches. The acoustic path in the material has a major influence on the measured sensor signal and therefore has to be taken into account. Important parameters are the position of the source and the sensor. From the signal processing point of view the behaviour of the acoustic channel can be described by position dependent transfer functions. A typical way to predict such functions is by simulation which requires a lot of prior knowledge like material properties and the shape of the specimen. In this paper a different approach based on real world measurements is presented. The sound transmission in a concrete specimen (dimensions 1451 x 105 x 100 mm) between an actuator and an acceleration sensor was investigated in the frequency domain of 200 to 9,900 Hz. In total 195 measurements were made alongside the axis of the specimen in a 5mm grid. Based on these measurements a model of the acoustic channel was designed and evaluated which describes the system behaviour in terms of absolute amplitude frequency responses alongside one dimension of the specimen. The response at one frequency bin over the position is modelled by an absolute sine function with an amplitude modulation term. The position based periodicity is modelled assuming only one speed of sound. The parameters of the model were derived by a genetic optimization algorithm. Except for positions very close to the source and close to the end of the specimen, the model predicts the absolute amplitude frequency response alongside the specimen with good accuracy, achieving a normalized mean square error below 0.01 per sample at frequencies with sufficient signal-to-noise ratio.