基于RBF神经网络的强流LIA故障诊断与性能评价技术

Fault diagnosis and performance evaluation for high current LIA based on radial basis function neural network

用于流体动力学诊断的强流LIA是庞大而复杂的系统,其性能预测和评估是十分困难的。针对强流LIA大量的单次快脉冲非平稳信号,提出基于小波包分析与RBF神经网络技术相结合实现故障智能诊断和性能评价的方法。该方法以强流LIA高维信号的小波包结点能量提取的特征向量来表征信号平顶、脉宽以及暂态特性。在此基础上,建立了“神龙一号”加速器腔电压及注入器出口束流故障诊断与性能评价原型系统,该系统不仅可进行故障诊断和性能评价,还可探测到加速器运行参数的变化趋势,为加速器的精细维护提供预测信息。

High current linear induction accelerator（LIA） is a complicated experimental physics device. It is difficult to evaluate and predict its performance. This paper presents a method which combines wavelet packet transform and radial basis function （RBF） neural network to build fault diagnosis and performance evaluation in order to improve reliability of high current LIA. The signal characteristics vectors which are extracted based on energy parameters of wavelet packet transform can well present the temporal and steady features of pulsed power signal, and reduce data dimensions effectively. The fault diagnosis system for accelerating cell and the trend classification system for the beam current based on RBF networks can perform fault diagnosis and evaluation, and provide predictive information for precise maintenance of high current LIA.