Research And Application For The Grating Subdividing Of Moire Fringes Based On The Neural Network

Grating measure technology is a kind of measurement technology which the grating is took as the displacement sensor to complete the measurement of displace with distinguishing the direction of the Moire Fringes and counting. With the development of science and technology, the technology of displacement measurement is demanded stricter than before, for instance, the precision of the system, the ability of anti-interference, reliability. The traditional hardware subdivision methods are gradually replaced by some method of digital subdivision. Generally the digital subdivision method utilizes A/D to collect two orthogonal signals, and then disposals of the collected signals through the microprocessor or logic circuit operates for table look-up, count, and interpolation, at last receives the value of subdivision. This article introduces a method of grating subdivision based on the neural network which takes the displacement convert to the digital quantity by gathering two orthogonal signals, then transmits to the microprocessor with process of neural network to get the subdivision value.The Radial Basis Function Neural Network (RBF) which has the higher forecast precision is selected with being compared the learning results of the BP neural network and the RBF. The tangent values which multiple sample points in one period of Moire Signal correspond to be taken as the input of the network; Micro displacement of the sample point in a grid pitch is regarded as the target output. The rational neural network model is constituted by utilizing three-layer RBF neural network which is combined with DSP to achieve Moire fringe subdivision. The research has shown that the neural network which uses only a few neurons can carry out high precision through the sample points with fractional learning.A method of polymerization fitting function is proposed because of the disadvantage of the complicated about the neural network calculating.lt is taked the transfer function of RBF as the form of the multinomial. Moreover is combined with the IQmath of the TMS320C28X. With the method of neural network, it will simplify the hardware and improve the precision of subdivision. The grating displacement measuring system based on the neural network is combined with the technology of DSP simplifies the structure of the system; enhances the reliability of the system, and improves accuracy of subdivision and tracking speed of the displacement. With the experiment, simulation and debugging circuit, it is proved that the system is feasible with the experiments of non-sample points, and has application value.