Micro drill defect detection with hybrid BP networks, clusters selection and crossover

According to the solution requirements, linear BP neural networks are designed which are consistent with the feature curves of the fitted equation, when the neural networks reach the equilibrium and stable state, so a optimization problem is transformed into the process of BP neural network reaching its equilibrium point. In order to obtain the global optimal solution, all the individuals of the 210 parallel neural networks are classified into several clusters, whose centroids are perturbed by the Levy flight, which is beneficial for the system to jump out of local extremum. At the same time, new clusters are obtained by clusters’ selection, and new individuals with crossover operator within each cluster and between clusters, so the computation decreases significantly. Then, new individuals are accepted by Metropolis criteria. Finally, when the BP neural network reaches the global optimal equilibrium state, the corresponding feature curves of the micro drill’s end faces is obtained. As such, the detection of the chip, rounded corner and other defects of the micro drill and some technical indexes are gotten.