Nondestructive Testing Method in Artificial Intelligence Real Time Systems

Abstract

Modern sonar and radar measurement systems are widely used in the field of nondestructive testing for a long time. Usually reference signal is emitted towards the object to be investigated and we get a signal, which is the sum of reference signal reflected from different plies. The task of signal processing is to determine time instances corresponding positions of certain ply, which allow analyzing structure of object. Usually the cross correlation function (CCF) of transmitted sequence and received sequence is calculated. If peaks were clearly identified in the cross correlation function (CCF), it would be easy to determine time instances. Due to the noise some coherent peaks, additive to the expected peaks, appear on the CCF, which are confusing in regard to the clear distinction of target. In order to cancel effects of noise as much as possible some measures have to be taken for data manipulation noise cancellation, such as averaging, inverse filtering and so on. These signal-processing methods need a lot of floating point floating point operations and are time consuming. That is why the usage of such ultrasonic systems is limited in real time systems, which are the base for self-organizing systems. Amount of calculations depends on the length of reference signal, the length of reflected signal to be processed and the noise reduction method used in such system. A new system with reduced amount of calculations is considered in this article. In this system only parts of reflected signal corresponding peaks of CCF are processed. These parts are defined in acquisition mode and afterwards system enters measurement mode. New noise reduction method based on wavelet transforms coefficients thresholding is applied in this system. The length of reference signal impacts system noise immunity and amount of calculations. The main problem in ultrasonic non-destructive testing systems is to cancel out effects of the noise. The optimal length of reference signal for wavelet based signalprocessing method is defined. All these measures allowed significantly reduce amount of calculations in the self-organizing systems.