基于小波包神经网络的传感器故障诊断方法

Methodology for Sensor Fault Diagnosis Based on Wavelet Package and Neural Network

讨论了小波包神经网络在传感器故障诊断中的应用问题。文中提出了将小波包分解提取各个节点特征能量与RBF神经网络进行模式分类的传感器故障诊断方法。通过三层小波包分解得到各个节点的分解系数,通过一定的削减算法使得故障的瞬态信号的特征得到加强,再根据重构的时域信号计算各个节点对应的能量,作为特征向量训练RBF神经网络。通过各种故障模式特征数据的训练,RBF网络具有了传感器故障诊断的功能。最后,通过工业锅炉流量传感器数据对训练之后的RBF神经网络进行检验,验证了这种方法的实用性和有效性。

Sensor fault diagnosis based on wavelet package and neural network is discussed. The method with wavelet package to get the pattern energy of each node and RBF neural network to classify the sensor mode is proposed for sensor fault diagnosis. After the decomposition of wavelet package, the coefficients of each node are achieved. With some filter algorithm, the instantaneous signal with fault character is strengthened. As the pattern sample, the energy of each node is calculated after the reconstruction with the coefficients above to train the RBF neural network. Then the RBF network possesses the capability for sensor fault diagnosis, which is tested with data from a flow sensor of an industrial boiler. Finally, the applicability and effectiveness of the proposed methodology is illustrated by diagnostic results.