Non-linear filtering of ultrasonic signals using neural networks.

Structure noise from inhomogeneous micro-structures makes the detection of flaws present in highly scattering materials difficult. Several techniques have been applied to improve the signal-to-noise ratio (SNR) in order to make flaw detection easier. Linear filtering does not provide good results because both structure noise and flaw signal concentrate energy in the same frequency band. Non-linear filtering can be used to reduce the structure noise of ultrasonic signals. Therefore, neural networks are applied in this work for this purpose. In order to use neural networks for non-linear filtering, dynamic structures must be applied. The easiest way to implement a neural network with the capability of processing temporal patterns is to consider them spatial ones, applying the signal into a tapped delay line of finite extension, that is the input of a static neural network (for example, a multi-layer perceptron). In this work, a dynamic neural network has been built to filter ultrasonic signals with structure noise, and has been trained with the real-time back-propagation algorithm, using as inputs 3000 synthetic ultrasonic signals of 896 samples each. Target signals for training are the same as the ones used as inputs but without noise. The neural network is trained in order to generate as output the target signal when the noisy input one is applied. For testing the performance of the non-linear filter, a new set of 500 noisy signals has been used. The SNR improvement is about 6 dB average. The results show that this non-linear filtering method is quite useful as pre-processing stage in flaw detection systems.

Ultrasonic flaw detection using radial basis function networks (RBFNs).

Ultrasonic flaw detection has been studied many times in the literature. Schemes based on thresholding after a previous matched filter use to be the best solution, but results obtained with this method are only satisfactory when scattering and attenuation are not considered. In this paper, we propose an alternative solution to thresholding detection method. We deal with the usage of different flaw detection methods comparing them with the proposed one. The experiment tries to determinate whether a given ultrasonic signal contains a flaw echo or not. Starting with a set of 24,000 patterns with 750 samples each one, two subsets are defined for the experiments. The first one, the training set, is used to obtain the detection parameters of the different methods, and the second one is used to test the performance of them. The proposed method is based on radial basis functions networks, one of the most powerful neural network techniques. This signal processing technique tries to find the optimal decision criterion. Comparing this method with thresholding based ones, an improvement over 25-30% is obtained, depending on the probability of false alarm. So our new method is a good alternative to flaw detection problem.