Infrared Thermal Wave Detection Of Plate Cavity Defects And Recognition Technology Based On PNN

At present, the board has become the main raw material from machinery manufacturing,shipbuilding, automobile, aerospace and chemical industry areas. In the process of plateproduction, processing and use of all can produce all kinds of different degrees of defects.However, in the plate, the surface cavity is a kind of typical defects. Once the plate surfaceappear cavity damage, the defect cross section stress distribution will tend to be very unevenand result in stress concentration. These defects will reduce the sheet metal corrosiondegradability, abrasion resistance and fatigue strength. Therefore, we need to achieve cavitydefect nondestructive testing and evaluation。Research based on infrared thermal wave testing technique, proposed a temporalcharacteristics of the infrared thermal wave detection method, and combined with principalcomponent analysis (PCA, principal component analysis) and probabilistic neural network(PNN, probabilistic neural network), in pixels, to achieve cavity defect recognition and areaof quantitative evaluation. According to typical subsurface defects in materials, research onthe aluminium plate as the object, the design and preparation of the four types of differenttypes of cavity defects. According to infrared thermal wave to test the specimen, and analyzedsequence image for the different types of cavity surface defects. In addition, the paper hasexplored defect recognition, which achieved the defect recognition.First of all, research during the cooling process of heating aluminum plate，the initialcharacteristics were obtained from the sequence grey value of normal and four kinds of cavitydefects area on the basis of sequence infrared image. And the principal component analysiswas used to extract temporal characteristics. Finally, combined with the probabilistic neuralnetwork, the cavity defects were identified and quantitatively evaluate in pixels. And usingthe support vector machine has carried on the comparative study. The research results showthat using PNN for test samples of normal and four types of cavity defect recognitionaccuracy rate were99.6%,97.3%,95.3%,93.5%and72.7%. Using the SVM of theidentification accuracy were99.9%,97.3%,89.1%,74.2and68.2%. Results verified thetemporal sequence grey value as the initial characteristics, in pixels, in combination withprincipal component analysis and probabilistic neural network for the effect of cavity defects.The research could be used to improve the plate surface defect detection and recognition toprovide theoretical guidance and useful reference.