基于频谱法和带偏差单元递归神经网络的水电机组振动故障诊断

引起水电机组振动的原因很复杂,而且水电机组的振动故障往往是多故障同时发生,使得故障诊断很困难,目前主要是应用基于模式识别的神经网络来进行故障分类,尤其是BP网络应用较多,但BP网络训练速度慢。文章提出应用带偏差单元递归神经网络的方法对水电机组的振动故障进行诊断。先对水电机组振动信号进行频谱分析,提取该信号在频率域的特征量,将频谱特征向量作为学习样本,通过训练,使神经网络能够反映频谱特征向量和故障类型的映射关系,从而达到故障诊断的目的。水电机组振动故障诊断仿真分析表明,与常规方法相比,应用带偏差单元递归神经网络进行故障诊断具有快速有效的优点。

Vibration Fault Diagnosis of Hydroelectric Generation Sets Based on Spectrum Analysis and IRN Network

The causes for the vibration of hydro-electric generation sets are complicated. What′s more,their multiple vibration faults happen simutaneously.Therefore it is difficult to diagnose the faults.Currently faults are classified mainly through applying the networks based on pattern recognition,especially BP neural networks,the training of which,however,is slow.For these reasons,the paper applied the internally recurrent network(IRN) to diagnose the vibration faults of hydro-electric generation sets.First,spectral analysis of their vibration signals was carried out to extract their spectral feature vectors in the frequency domains of the generation sets.Then the feature vectors were used as learning samples to train the IRN network to realize the mapping relationship between spectral feature vectors and fault types,thus achieving the purpose of diagnosing faults.The simulation of the vibrant faults of hydro-electric generation sets shows that compared with other diagnostic methods,the method proposed in the paper is fast and effective.