摘要
结合涡流无损检测的特点,介绍了基于RBF网络的涡流无损检测系统及系统软、硬件设计方法。硬件部分采用TI公司的DSP芯片TMS320VC5410作为核心,完成信号的产生及其处理。软件部分采用了基于RBF神经网络的涡流无损检测方法,且针对常用的RBF中心选择算法不能构成全局最优、收敛速度慢等缺点,提出采用基于改进Fisher中心选择算法确定RBF网络隐层节点数及径向基函数中心。仿真结果表明:利用DSP产生处理信号,得到的波形精度高、稳定性好;利用改进Fisher算法确定RBF隐层节点数及径向基函数中心简化了网络结构,提高了分类能力和收敛精度。
Considering the characteristics of eddy current nondestructive detecting,the system configuration based on DSP and RBF network is introduced.And the hardware and software design method is given.In the part of hardware,DSP TMS320VC5410 made in TI to generate sine waveforms and process signals is adopted.In the part of software,RBF network is used to complete defect detection.For usual RBF center selection algorithms have the disadvantages of no global optimization and slow convergence speed,improved Fisher ratio algorithm is proposed to ascertain the numbers of hidden nodes and the RBF center.The results show that the waveform generated by DSP has good accuracy and stability.The neural structure is simplified strongly,and the convergence precision and classification ability is improved.
出处
《河北工业科技》
CAS
2010年第4期239-241,244,共4页
Hebei Journal of Industrial Science and Technology