Quantitive Nondestructive Testing Of The Metal Hardness Using Wavelet Networks

Electromagnetic nondestructive testing on the quality of steel has been studied for many years and acquired a great deal of achievements. However there are many problems in the testing of hardness and cracks, and mess problem of work pieces. Especially the unlinear relation between the degree of hardness of aluminium and initial magnetic permeability, these all affected the testing of the metal hardness seriously. To settle the above problems, this paper puts forward a kind of quantitative nondestructive testing method of the metal hardness using wavelet networks. Proposes wavelet neural network models and weights study algorithms .Wavelet functions act as weights but not using wavelet nonlinearity between input and hidden layer in qualitative testing, while the input layer of neural network is inner product of feature vector and wavelet using wavelet nonlinearity in quantitative testing. Proceed once with the WGF- Ⅰ electromagnetism instrument to test, Withdraw the characteristic signal, applied wavelet networks theories, establish the degree of hardness characteristic signal and degree of hardness unlinear mapped, the hardness of aluniniun alloy can be directly displayed. It is by simulated experiment proved that wavelet networks have faster convergence speed for network training, more generalization capacity and accurate inspection than the general other neural networks. Wavelet networks can quantitative testing the hardness of LY12 aluminium alloy, the precision of the hardness is about HRB±0.8.