Classifying ultrasonic resonance spectra using a neural network

In this paper we present a neural network based spectrum classifier (NSC) a nd its application to ultrasonic resonance spectroscopy (URS). URS is a met hod for testing stiff materials by exciting the sample under test into mech anical resonance by progressively sweeping the frequency of a driving trans ducer across a certain frequency range. An artificial neural network (ANN) is used for spectrum classification to meet the requirements of high sensit ivity for small but relevant changes in the spectra, and simultaneous robus tness against measurement noise. Among several types of ANNs that could be used for classifying the spectra we have chosen a multilayer perceptron (ML P). Although the MLP itself can perform feature extraction, we included an optional preprocessor for this purpose. The NSC is essentially model free a nd can be trained using real and modeled spectra. The classifier uses both amplitude and phase information in the spectra. The performance of the clas sifier has been verified using a number of practical applications, such as, aerospace composite structures, ball bearings and aircraft multilayer stru ctures. Here, we present in more detail results of NSC application to detec tion of disbonds in adhesively joint multilayer aerospace structures using Fokker Bond Tester resonance instrument. In this case the classifier is cap able of detecting very small disbonds (larger than 25 percent of the sensor area) and correct identifying their position in the structure (identifying the joint with the discontinuity).