基于FPGA及RBF神经网络的电磁无损检测技术

Electromagnetic nondestructive testing technology based on FPGA and RBF neural network

摘要:

电磁无损检测技术是无损检测领域的一个研究重点,本文针对电磁超声波的处理,提出了一种基于FPFA的参数优化的RBF神经网络。首先,通过FPGA编程实现对电磁超声波信号的采集,设计了放大电路将原始的电磁超声波进行放大处理已满足RBF神经网络的需求；提出一种采用K-means聚类算法来计算RBF中径向基函数的中心和宽度的参数优化RBF算法,K-means聚类算法的初始聚类中心难以确定会导致RBF算法的参数无法优化,提出KL散度,采用数据密度分析法来计算K-means算法的聚类中心。试验表明,改进后的K-means算法的聚类结果比传统K-means算法更准确,参数优化后的RBF神经神级网络对发动机涡轮叶片的缺陷预测比传统的RBF神经网络更准确。

Abstract:

Electromagnetic non-destructive testing technology is a research focus in the field of non-destructive testing. This paper proposes a RBF neural network based on FPFA parameter optimization for the processing of electromagnetic ultrasonic. First of all, the acquisition of electromagnetic ultrasonic signals is realized through FPGA programming, and the amplification circuit is designed to amplify the original electromagnetic ultrasonic waves. It has met the needs of RBF neural network; a K-means clustering algorithm is proposed to calculate the radial basis in RBF. The parameters of the center and width of the function optimize the RBF algorithm. The initial clustering center of the K-means clustering algorithm is difficult to determine, which will cause the parameters of the RBF algorithm to be unable to optimize. The KL divergence is proposed and the data density analysis method is used to calculate the K-means algorithm. Cluster center. Experiments show that the clustering results of the improved K-means algorithm are more accurate than the traditional K-means algorithm, and the parameter-optimized RBF neural network is more accurate than the traditional RBF neural network in predicting the defects of the engine turbine blades.