An innovative neruofuzzy network is proposed herein for parameter recognition, specifically for steel plate's defects size inspection through different NDT sources data fusion. A neural network architecture is used to automatically deduce membership function based on a hybrid supervised learning scheme and a set of activation functions are used to adapt to different fuzzy states. The realization of this model and its characteristics are discussed in detail. The application of this model on the inspection of surface defect sizes shows that a quantitative method for determining the actual defect size is successfully developed to make full use of the measured defects sizes from different NDT sources.
An automated vision system is presented intending to detect and classify surface defects on steel strip. The framework of the system is briefly introduced and the realization, mainly focused on image processing and pattern classification, is discussed in detail. Original images of defects obtained from CCD camera are preprocessed firstly by using of DSP, which includes threshold segmentation, morphological operations, edge detection, and contour extraction. After several key features have been selected, they are inputted into fuzzy neural network functioned as classifier. The result shows that the fuzzy neural network classifier provides better classification accuracy and lower iteration times.