基于GA-BP模型的锆合金管材超声检测参数优化

Optimization of Ultrasonic Testing Parameters of Zr-4 Alloy Tubing Based on GA-BP Model

针对实际生产中锆合金管材超声检测参数优化问题,基于少量超声检测数据,利用遗传算法(GA)改进的BP神经网络(GA-BP)建立了管材超声检测参数优化模型。模型结构选取三层网络,其中输入层为导套间隙、送进速度、重复频率、旋转速度和水层厚度,输出层为人工缺陷重复检测的标准偏差。为避免因输入层节点数过多带来噪声影响预测精度,采用主成分分析(PCA)法对输入层进行了降维,最终形成2-5-1拓扑结构。利用该优化模型,确定出Φ8.08mm×0.55mm的Zr-4管材超声检测的最佳参数为:管材旋转速度4 000r/min、送进速度6m/min、仪器重复频率15kHz。进一步生产检测表明,新型GA-BP模型的应用可为核用锆合金管材超声检测参数优化提供一种新的技术手段。

Based on a small number of ultrasonic data,a ultrasonic testing parameters model for Zr-4 alloy tubing is built based on BP neural network optimized by genetic algorithm.The model employs a three-layer network,in which the input layer consists of guide gap,feeding speed,repetition frequency,rotation speed and water layer thickness,whereas the output layer is the standard deviation of the repeat test for artificial defect.In order to avoid the influence of noise on the prediction accuracy due to the excessive number of input nodes,the principal component analysis（PCA）method is used to reduce input nodes,and finally the 2-5-1 topology is formed.The model is optimized by ultrasonic testing parameters,and it is determined that the optimal ultrasonic testing parameters forΦ8.08×0.55 mm nuclear Zr-4 tube are 6 m/min as the feed speed,4000 rpm as the rotation speed,15 kHz as the repetition frequency.Further testing experiment shows that the application of new GA-BP model is a new way to optimize the ultrasonic testing parameters of nuclear zirconium tubes.