To evaluate the circularity error of a twist drill

the coordinates of sampling points of the margin projection were measured with universal tool microscope.The neural network was introduced into data processing in order to fit the elliptical coefficients of the margin projection.The neural network was trained with the coordinates of sampling points and constant 1 which were taken as 3 inputs

and the square of errors between the neural output and constant 1 as performance index.The weights between hidden neurons and output neuron

and weights between inputs neuron and hidden neurons were tuned

thus stable weight values could be obtained in the light of negative gradient descent mean until the desired performance index was reached

from which the central coordinates and reciprocals of elliptical long axis's and short axis's square were obtained respectively

thus the whole expression of the margin projection ellipse were gotten.The circularity errors of twist-drills could be got in higher precision by adopting conventional means.The proposed algorithm adopting neural network weight to fit elliptical expression coefficients is a news dada processing means for measurement geometrical errors of workpiece.