人工神经网络用于核电站松动件定位技术

Loose part localization in NPPs using neural network techniques

为了提高核电站松动件故障诊断的能力 ,结合时频变换理论和神经网络理论 ,提出了对松动件碰撞位置进行估计的新方法。通过对安装在反应堆压力壳上的多个加速度传感器的信号进行采集 ,并经过信号预处理、时频变换、神经网络计算等过程 ,实现对核电站松动件碰撞位置的定位。经过对 10 0多次实验数据的学习 ,发现神经网络计算结果能够满足核电站松动件定位精度要求。该方法适用于复杂连接结构、以及复杂几何曲面装置的松动件定位问题 ,减少了人为因素的影响 。

The anomaly diagnosis capability for loose parts in NPPs can be impvored using a new method to estimate the impact positions of loose parts based on a time frequency transform and neural network theories. The loose part impact positions are located by analyzing signals from accelerometers mounted on the reactor vessel with signal pre processing, time frequency transforms, and neural networks. The neural network results using more than 100 experimental data sets can supply the required precision for loose part localization in NPPs. Therefore, this method can be used to locate loose parts in complex geometies. The method can also reduce operator intervention with more automatic processing ability and better precision for locating loose parts in NPPs.