基于神经网络熔透电弧声特征参数评价与选择

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

在进行基于电弧声的焊接过程熔透监测与诊断过程中,恰当地选择电弧声特征参数是诊断成败的关键.基于神经网络的特征评价和特征选择方法,利用神经网络的训练结果对特征参数进行合理的评价.神经网络能满足高分辨率信息压缩所需的非线性映射条件,通过特征选择将焊接熔透模式识别中复杂的分类问题转移到特征处理阶段,利用神经网络有效地实现了特征参数的降维.诊断实例验证了该方法的有效性.

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.