基于堆叠降噪自编码的给水泵轴承故障诊断

Fault Diagnosis of Feed Pump Bearing Based on SDAE

给水泵是电厂锅炉供水系统中十分重要的设备,其长时间运行后难免会出现故障。通过某电厂多年维修记录发现,给水泵发生故障大部分都是出现在轴承上,因此有必要对给水泵轴承进行故障诊断。近年来深度学习作为一种新兴的机器学习方法,逐渐在机械设备故障诊断领域中得到了推广与应用。由于电厂给水泵采集到的数据和测得的信号总掺杂着外部噪声,提出了一种基于堆叠降噪自编码(SDAE)的水泵轴承故障诊断。堆叠降噪自编码的原理是先通过对原始数据进行破坏,即添加“损伤噪声”,然后通过基本的自编码网络进行数据还原,从而得到更具有鲁棒性的特征表示。以西储大学轴承故障数据为例,构建堆叠降噪自编码网络,识别所测得的时域信号的特征,从而对轴承故障状态进行精确识别。试验结果表明,SDAE方法可以得到更鲁棒性的特征表示,易于进行故障分类,并且轴承故障状态识别准确率较高。

Feed water pump is a very important equipment in boiler water supply system of power plant.According to the maintenance records of a power plant for many years,it is found that the faults of the feed pump are mostly on the bearing,so it is necessary to diagnose the fault of the feed pump bearing.As a new machine learning method,deep learning has been gradually applied in the field of fault diagnosis.Because of the data collected from feed water pump in power plant and the measured signal are mixed with external noise,a kind of pump bearing fault diagnosis based on stacked de-noising auto-encoders(SDAE)was presented.The SDAE algorithm added“damage noise”to the raw data and then reconstructed the data through a self-coding network to achieve a more robust signature.Taking the bearing fault data of Western Reserve University as an example,a stacked noise reduction self-coding network was constructed to identify the characteristics of the measured time-domain signals,so as to accurately identify the bearing fault state.The experimental results show that SDAE method can obtain more robust characteristic representation,which is easy to classify faults,and has higher accuracy in bearing fault state recognition.