Automatic detecting and classifying defects during eddy current inspection of riveted lap-joints

Uppsala University, Signals and Systems, SE-751 20 Uppsala, Sweden
NDT & E International (Impact Factor: 1.74). 01/2000; DOI: 10.1016/S0963-8695(99)00007-9

ABSTRACT This article presents a novel method for automatic detection and classification of cracks located in the second lower layer of the aircraft lap-joints during Eddy Current (EC) inspection. The cracks originating from the rivet holes were detected using a tailor-made deep penetrating EC probe. The proposed method consists of three steps: pre-processing, feature extraction and classification. The pre-processing, performed before the feature extraction included median filtering, rotation and de-biasing of the EC patterns. The rotation of the patterns was performed so that energy of the responses to the rivets was maximized along the quadrature direction, while the defect responses were maximized in the in-phase direction in the impedance plane. Feature extraction was then performed using four different methods: discrete wavelet transform, Fourier descriptors, principal component analysis (PCA) and block mean values. The classification was performed using a standard multi-layer perceptron (MLP) neural network. All the pre-processing methods showed similar classification performance on the used data set, but the PCA method compressed the data best.

  • [Show abstract] [Hide abstract]
    ABSTRACT: For defect characterization in steam generator tubes, feature extraction to interpret eddy current testing (ECT) signals has been recognized as an important step. In this paper, we propose a new feature extraction from ECT signals for defect classification and defect sizing. Using the extracted features as an input vector, a multi-layer perceptron (MLP) neural networks are used to classify defect types and to predict defect size. Although the proposed method requires relatively fewer features for the defect classification, it provides not only a high level of classification accuracy but also promising robustness to noise. Moreover, for the prediction of defect size, the proposed method yields a comparable prediction accuracy even though it needs fewer features than the previous result.
    Ndt & E International - NDT E INT. 01/2009; 42(7):658-663.
  • [Show abstract] [Hide abstract]
    ABSTRACT: This paper investigates the application of two source separation techniques, principal component analysis and independent component analysis, to process the data from the inspection of riveted lap joints by eddy currents. An eddy current array sensor is designed for the rapid inspection of lap-joints and used to test a set of flawed rivet configurations featuring 1–10mm notches, buried down to a 4mm depth. Implementation methods are proposed for processing such eddy current data by means of both the considered source separation techniques. The signal processing results obtained from the experimental data are compared in terms of source separation efficiency and detection using a receiver operating characteristic approach. In the light of this study, both the techniques appear to be efficient. However, the principal component analysis provides better defect detection results, especially for deeply buried defects.
    Ndt & E International - NDT E INT. 01/2010; 43(7):606-614.
  • [Show abstract] [Hide abstract]
    ABSTRACT: Eddy current nondestructive evaluation techniques are widely used in structural integrity and health monitoring. A novel algorithm based on kernel methods was proposed for characterising eddy current (EC) signals. In scanning inspection, the EC signals responding to the impedance change were pre-processed for noise elimination using the wavelet packet analysis method. Then, the Morlet wavelet was employed to perform the decomposition of one-dimension differential signals onto the coefficients of the wavelet transforms at different scales as input to kernel principal component analysis (KPCA) for feature extraction. After feature extraction, support vector machine (SVM) was carried out to classify EC signals. It is shown by extensive experiments that KPCA is better than the principal component analysis for feature extraction. The kernel methods using SVM by KPCA feature extraction can perform better than the other classification methods.
    Nondestructive Testing And Evaluation 03/2009; 24(1):19-37. · 0.67 Impact Factor