基于互相关能比熵和BiGRU-GRU的轧机关键零部件早期故障诊断

Early Fault Diagnosis of Key Parts of Rolling Mill Based on Cross-Correlation Energy Ratio Entropy and BIGRU-GRU

轧机作为机械制造行业的重要设备,工况环境复杂,其关键零部件极易发生故障,对其进行早期故障诊断,趋势预测存在困难;对此文章以轴承为例,提出了一种新型性能退化指标用于检测出现早期故障的时刻;对于防止轧机工作环境复杂的问题,首先要对采集到的样本信号进行降噪,实现对噪声信号的去除,之后利用互相关函数对样本前后数据进行互相关分析,然后求分析所得数据的所有极值点能量与总能量得比值,最后将做的比值带入信息熵公式,即为最终得性能退化指标,即互相关能比熵,并通过包络谱分析验证指标的有效性;针对轴承性能退化趋势预测的问题,利用门控循环单元网络(GRU)和双向门控循环单元网络(BiGRU)各自的优点建立了BiGRU-GRU网络。将采集到的数据分为训练数据和测试数据,在训练数据中训练之后,对测试数据进行预测,实现了对轴承性能退化趋势的预测。并通过对比实验证明了所提性能评估指标和网络比一般指标和网络具有更好的效果。

As an important equipment for metallurgy,rolling mill has complex working conditions,and its key parts are easy to break down,so it is difficult to make early fault diagnosis and trend prediction.Taking bearing as an example,a new performance degradation index is proposed to detect the moment of early failure.To prevent the mill work environment complex problems,first of all samples were collected for signal de-noising and implementation,the noise signal is eliminated after the cross-correlation function is used to analyzed the data before and after the sample cross-correlation,then the ratio data is analyzed from all the extreme value and total energy,which taken into the information entropy formula,finally the ultimate performance degradation index is made out,namely the cross correlation energy ratio entropy,and the index validity is verified by the envelope spectrum analysis.Aiming at the problem of bearing performance degradation trend prediction,the BiGRU-GRU network is established based on the respective advantages of Gate Recurrent Unit(GRU)and Bidirectional Gate Recurrent Unit(BiGRU).The collected data are divided into training data and test data.After training in the training data,the test data are predicted to realize the bearing performance degradation trend.Comparative experiments show that the proposed performance evaluation index and network have more effective than the general index and network.