Adaptive Data Compression of Ultrasound Data for Long-term Data Acquisition and Trend Evaluation

Holstein, Peter; SONOTEC Ultraschallsensorik Halle GmbH; Germany

Holstein, P.; SONOTEC Ultraschallsensorik Halle GmbH; Germany
Seitz, S.; Institute of Circuits and Systems; Germany
Tharandt, A.; SONOTEC Ultraschallsensorik Halle GmbH; Germany
Probst, C.; SONOTEC Ultraschallsensorik Halle GmbH; Germany
Tetzlaff, R.; Institute of Circuits and Systems; Germany

ID: ECNDT-0339-2018
Download: PDF
Session: Data Processing - UT
Room: G2
Date: 2018-06-15
Time: 11:10 - 11:30

Data compression technologies such as MP3 are extensively used for audio technologies.
There are two objectives: to maintain the quality of the original sound and two compress the data as much as possible. The degree of compression depends strongly from the signal character. The information of pure harmonic signals can be much better compressed than by considering noisy signals.
The ultrasound produced by processes in industry is used for maintenance purposes since some decades. On the one hand there are many signals with noisy character. Examples are leakage noise, jet noise, the noise of friction in bearings and gears. On the other hand there are signals with dominantly repeating impulses such as bearing faults (cracks or spalled areas) or arcing during electrical discharge. For both types of data different compressing rates has to be found.
Traditionally, ultrasound data for maintenance are evaluated by means of narrow band techniques such as heterodyning the ultrasound signal in the audible range, thereby a data reduction can be achieved.
In order to avoid the loss of real physical information, an ultrasound broadband technology has been introduced in order to consider the complete frequency range leading to a considerably increased data rate. Long-term recordings of service data would generate a huge amount of data on short time scales. Furthermore, the broadband technologies requires new approaches to make the signal audible. A vocoder technique has been modified in this contribution in order to compress the ultrasound data in real-time to a bandwidth which is compatible to the listening capabilities of testing persons. The spectra of compressed data can be used for the interpretation of measuring data. The degree of loss with respect to the information content has been estimated in comparison to the original uncompressed signal. Furthermore, ultrasound data – original and the compressed audible data – has been taken by applying pattern recognition techniques based on Cellular Neural Networks which will be exemplified by means of maintenance related data.

Peter Holstein, Andreas Tharandt, Christian Probst
(SONOTEC Ultraschallsensorik Halle GmbH)

Ronald Tetzlaff, Steffen Seitz, Jens Müller
(Institute of Circuits and Systems, TU Dresden