基于小波包和BP网络的松脱件质量估计

摘要

In order to reduce the false alarm or omission phenomenon existing in the loose parts monitoring system,reduce the possibility of that the loose parts bring great risks to the safe operation of the nuclear reactors under the impact of high speed water flow,the collision of the loose parts and the inner wall of the equipment was simulated by the collision of the six different quality steel balls and the steel plate,and then the corresponding impact signal was obtained by the acceleration sensor.The wavelet transform was used to decompose the original shock signal in the multi-layer frequency domain,and the energy ratio was used as the mass estimation feature vector.The selected eigenvector was input as a multi-layer back propagation (BP) neural network,and a suitable steel ball mass estimation model was established after training with different samples.The results showed that the estimated errors of different quality steel balls were 0.04％,1.14％,6.76％,0.01％,0.05％ and 0.07％,respectively,the errors were all within 8％.By using the scientific combination of wavelet packet transform and BP neural network,the minimization of quality estimation error is realized under the control of system cost.%为减少现有松脱件监测系统存在的误报或漏报现象,降低松脱件在高速水流冲击下给核反应堆安全运行带来隐患的可能性.通过合理简化搭建松脱件碰撞模拟实验平台,选取6个不同质量钢球与钢板发生自由落体碰撞来模拟松脱件与设备内壁面撞击,再用加速度传感器获得相应冲击信号.采用小波包变换对原始冲击信号进行多层频域分解,利用每个频带能量占总能量的比值作为质量估计特征向量.将选好的特征向量作为多层前馈(BP)神经网络输入,经不同样本组合训练后建立较合适的钢球质量估计模型.数据处理结果表明,不同质量钢球质量估计误差分别为0.04％、1.14％,6.76％、0.01％、0.05％、0.07％,误差均在8％以内.通过将小波包变换和BP网络组合应用,在控制系统成本的情况下实现质量估计误差最小化.