Date of Award
Master of Science
Carbon composites, particularly carbon fiber reinforced plastics (CFRPs) are increasingly being used in many commercial applications such as automobile, aerospace, and civil infrastructures. This increase in the demand of CFRPs has led the manufacturers to research for ways of early detection and classification of defects in CFRP. This thesis work discusses the implementation of a Hopfield neural network (HNN) to automatically detect and classify defects in CFRP by ultrasonic testing (UT). Three of the most common and critical defects found in CFRP, i.e. foreign object (FO) inclusion, impact damage (ID), and porosity, were the main focus of this research. These defects were engineered into three different CFRP panels during manufacturing and were provided by an outside supplier. These panels were inspected on a standard immersion ultrasonic testing system in pulse-echo mode. One hundred time-amplitude based ultrasonic A-scan signals were recorded from the defected areas of each of the panels. Additionally, a hundred ultrasonic A-scan signals were recorded for the defect-free region of the CFRP. Signal preprocessing in terms of signal alignment is a vital process in any classification procedure as unaligned signals are prone to bad classification accuracies. Therefore, the raw A-scan signals were start-point aligned prior to any classification using cross-correlation technique. This research also focused on developing a feature extraction technique that could be used in conjunction with the HNN to classify the UT signals with high classification accuracy. A feature extraction technique based on gating technique was developed which consisted of five features, namely mean value of the signal, standard deviation of the signal, peak-to-peak amplitude value of the front wall echo (FEW), peak-to-peak amplitude value of the back wall echo (BWE), and time of flight (TOF) between FWE and BWE of each signal. Four other feature extraction techniques based on mean value of the signal, standard deviation value of the signal, fast Fourier transform (FFT), and discrete wavelet transform (DWT) were also used for the comparison and validation of results obtained by the developed feature extraction technique. The 100 signals for each category were randomly partitioned into equal-sized training and testing data sets. The partitioning was repeated for 1000 iterations, to average the results and obtain a robust estimate of the classification performance. The classification of the signals was carried out by implementing two approaches. The first approach utilized a dichotomous classification between each defect type and the non-defected region. And the second approach utilized a 4-class classification in which the defect types and the non-defected signals were classified at the same time. The results of this research showed that the classification accuracies for the 2-class problem obtained through the developed feature extraction technique exceeded 99%, which were in agreement with the results of classification obtained through the four conventional feature extraction techniques. The results of the 4-class problem obtained through the developed feature extraction technique exceeded 96% classification accuracy. For direct comparison, the results obtained from the four conventional feature extraction techniques exceeded 99% classification accuracies. Based on the results obtained, it can be concluded that the developed feature extraction technique can be used in conjunction with HNN to successfully classify defects in CFRP. With few modifications, the developed technique can be implemented to classify other types of defects in CFRP and can be implemented in different applications.
This thesis is only available for download to the SIUC community. Current SIUC affiliates may also access this paper off campus by searching Dissertations & Theses @ Southern Illinois University Carbondale from ProQuest. Others should contact the interlibrary loan department of your local library or contact ProQuest's Dissertation Express service.