An optimized rail crack detection algorithm based on population status

Abstract

To improve efficiency and accuracy of wavelet packet decomposition method modified by simple genetic algorithm (SGA), a novel genetic algorithm, which is based on variance of population and population entropy, is proposed. And then wavelet packet decomposition method is optimized by this algorithm to detect rail cracks. In the optimized method, internal state of population and population diversity are linked up with evolutionary operations to adjust crossover-mutation operators of genetic algorithm. Further, a mathematical model describing fault signal is established, and its parameters are optimized to effectively extract information. The proposed algorithm was tested by test functions and simulated fault signals of rail cracks. The results about simulated fault signals show that convergence probability of proposed algorithm — at best — is 45% higher than that of SGA and 28% higher than that of improved adaptive genetic algorithm (IAGA), and accuracy of crack fault detection reaches above 92%. Meanwhile, the proposed algorithm isn’t prone to stagnation and has fast convergence speed and high accuracy of fault detection. This research not only improves performance of SGA, but also provides a new detection method for fault diagnosis of wheel-rail noise.