Feature evaluation and selection of penetration arc sound signal based on neural network

In welding processing penetration detection and diagnosis based on arc sound, how to choose its proper parameters is vital to diagnosis. The feature evaluation and selection methods were presented, the results trained by neural network were used to evaluate feature parameters. Because neural network satisfied the nonlinear mapping requirement for high-resolution information compression, the complex classification problem in welding penetration pattern recognition was transferred to feature processing stage, and feature extration was realized by neural network effectively. An illustration validated the effectiveness of the method.