论文部分内容阅读
在激光超声缺陷检测技术中,不同类型缺陷采样信号的准确分类至关重要.针对激光超声表面波实验采样信号高维小样本的特点,采用了一种有监督学习的Kohonen神经网络(S_Kohonen)自适应分类方法.在S_Kohonen网络自组织学习的过程中,通过改进网络的学习率提高了网络的收敛速度.通过采用一种无需邻域半径判断的自适应权值调整方式来实现竞争层神经元权值不同程度的调整,从而更有效的表征输入样本的分布特征.通过对不同类型缺陷探测样本的多次实验,验证了所述方法具有良好的分类预测效果,多次交叉验证分类正确率均能达到100%.
In laser ultrasonic flaw detection technology, it is very important to accurately classify the different types of defects in the sampled signals.Aiming at the characteristics of high-dimensional small sample of the laser ultrasonic surface wave experimental sampling signals, a supervised Kohonen neural network (S_Kohonen) Adaptive classification method.In the S_Kohonen network self-organizing learning process, the network convergence rate is improved by improving the learning rate of the network.An adaptive weight adjustment method without neighborhood radius is adopted to realize competitive neurons Which can be used to characterize the distribution characteristics of input samples more effectively.Through experiments on different types of defect detection samples, it is verified that the proposed method has a good classification and prediction effect, and the accuracy of multiple cross-validation and classification can be Reached 100%.