Testing of Rounded Corner for Micro-Drill on Hybrid of BP Neural Network and Adaptive Particle Swarm Optimization

A new approach based on hybrid of linear BP neural network and particle swarm optimization algorithm for fitting of micro-drill’s margin projection is proposed. The network is structured according to fitting equation, where sampled point coordinates of micro-drill and their recombination are taken as 6 inputs, and one output is obtained. The square of difference between the output and constant 0 is taken as performance index. The weights between input neurons and output neuron are tuned in the light of gradient descent method. In order to obtain global optimal solution, improved particle swarm optimization algorithm is integrated into the fitting program, where inertia weigh is modifying adaptively and mutation operator is carried on to increase the variety of particle dynamically. While the iteration is finish and the desired performance index of BP neural network is reached, thus stable weight values are obtained, according to which expression coefficients of ellipse can be solved. The rounded corner and diameter of the micro-drill can be tested easily. The presented approach provides a new solving method for ellipse fitting with advantages of programming easily and high precision.