论文部分内容阅读
本文利用计算机辅助进行在役管线焊故障缝缺陷检测,在缺陷特征提取中提出圆形度、长宽比、填充度、尖部尖锐度、对称度、灰度比以及缺陷的重心坐标相对焊缝中心的位置等7个参数作为缺陷的特征值,可有效地分类识别不同故障缺陷。在缺陷分类的解决方案上,采用具有自组织、自适应的3层前馈式神经网络,运用改进的BP算法,以焊缝缺陷的特征参数作为神经网络的训练样本。本文还通过实验的方法,分析了初始权值、隐含层的神经元数量、动量系数、误差水平及学习速率对网络训练的影响。
In this paper, computer-aided pipeline fault detection in-service fault detection, proposed in the defect feature extraction circularity, aspect ratio, filling, sharpness, symmetry, gray scale ratio and the relative position of the center of gravity defects 7 parameters such as the location of the center as the eigenvalue of the defect can be effectively classified to identify different fault defects. In the solution of defect classification, a 3-layer feedforward neural network with self-organization and self-adaptability is adopted. The improved BP algorithm is adopted to take the characteristic parameters of weld defects as the training samples of neural network. The paper also analyzes the influence of initial weight, number of neurons in hidden layer, momentum coefficient, error level and learning rate on network training through experimental methods.