The Application Of Self-organizing Mapping Algorithm In Steel Separating Instrument

The relationship between the development of the steel industry and thenational economy is one prosperity or loss the both production. Safety hasbecome the focus of the development of the steel industry; the quality of the steelwork piece plays a significant role in the development of the steel industry,subject to the attention of the entire field of the use of the safety, reliability andservice life of the steel work piece. Before the steel work piece being used if wecan know whether there is a defect production is significant. With thedevelopment of the computer, a lot of testing equipments have been developedand put into the market, and achieved a certain effect, however, in theelectromagnetic field of non-destructive testing still have some steel materials thepresence of crack detection efficiency is low and high rate of false. Many reasonscan cause these problems, the main reason is that the algorithm is not ideal andthe instrument processor chip processing performance if not enough, so in orderto improve the accuracy of the crack detection and speed, we can start from thesetwo aspects of existing equipment to upgrade of redesign.In this paper, the SOM neural network technology, electromagneticnondestructive testing technology and FPGA technology were used to improvethe crack detection algorithm of the existing WGF-steel material automaticsorting instrument. WGF-has chosen Cyclone III FPGA with32-bithigh-performance processor Nios II and advanced SOPC solutions of AlteraCorporation. The current crack detection algorithm of WGF-is BP neuralnetwork, it has the shortcomings of slow convergence and easy to fall into localminimum points etc. Although the current algorithm achieved a certain effect, butthere is still a certain error rate. This paper uses the improved SOM neuralnetwork algorithm instead, to achieve a better detection effect. The initial amplitude of the permeability method is primary means to theWGF-extract the characteristic signal of the steel work piece. The signalprocessing is done in two steps, first, the characteristics of the signal extracted,then extracted signal denoising, this is the way get the reaction performanceindicators useful signal. Established SOM neural network, use thesamples(including crack and no cracks in the work piece) to train the network,extract the characteristic signal of the tested work piece and put into the trainedSOM neural network, according to the judgment rule, it will automatically get thecorrect result.Through the sample test and verification, the application of the algorithmdesigned in this paper in WGF-performance very good, realized the steelmaterial cracks accurately identify and high efficiency. It has significantadvantages than the conventional detection algorithm, and has high value ofpromotion in crack detection.