Research of Hopfield network & simulated annealing on micro-drill´s roundness error measurement

While the roundness error of micro-drill margins is measured, the special structure´s Hopfield neural network is designed according to normal equation obtained by pseudo-inverse. The network has 5 neurons, the weights of which are elements of normal equation´s symmetry matrix, the inputs vector of which is composed of numeral value of normal equation´s right. At the same time, via simulated annealing algorithm, iteration is carried out in the light of Metropolis criterion, and the probability of convergence to global optical solution is 1 while the present temperature approaching to desired lease value. Thus the global optimization solution is obtained, which can be taken as coefficient of fitted ellipse, and the roundness error of micro-drill margins can be solved.