摘要
针对风电机组轴承故障信息采集过程中数据缺失导致故障类型无法识别问题,提出了一种基于四维张量模型特征分解恢复缺失数据的方法。首先,基于转速、时窗、经验模态分解和时间4个维度构建四维张量;其次,通过加权优化算法实现张量填充,修补故障数据的缺失值;然后,对张量进行Tucker分解得到核心张量及因子矩阵;最后,基于梯度优化算法进行迭代优化得到最终核心张量及因子矩阵,并利用二者对四维张量进行重构得到恢复数据。采用实验数据和实际数据来验证提出方法的有效性和可靠性。结果表明:两组恢复数据的RMSE值分别为0.3169和0.0291,远小于4种对比方法的RMSE值。利用双稳态随机共振对2组恢复数据进行故障特征提取,信噪比显著提高,分别为-13.2647和-15.5212,进一步验证提出方法的准确性。
To solve the problem that the fault type cannot be identified due to missing data in the process of collecting fault information of wind turbine bearing,a method based on four-dimensional tensor model feature decomposition is proposed to recover missing data.Firstly,the fourth order tensor of vibration fault data is constructed based on the four dimensions of rotational speed,time-domain window,empirical mode decomposition and time.Secondly,the factor matrix is obtained by Tucker decomposition of the tensor.Then,tensor filling is realized by weighted optimization algorithm to recover the missing values of bearing fault data.Finally,the core tensor and factor matrix are obtained iteratively based on the gradient optimization algorithm,and the four-dimensional tensor is reconstructed to get the recovered data.The validity and reliability of the proposed method are verified by experimental and practical data.The results show that the RMSE values of the two groups of recovered data are 0.3169 and 0.0291,which are much lower than the RMSE values of the four comparison methods.By using BSR to extract fault features from two groups of recovered data,the signal-to-noise ratio is significantly improved to-13.2647 and-15.5212,respectively,which further verifies the accuracy of the proposed method.
作者
时培明
孙航璇
许学方
韩东颖
SHI Peiming;SUN Hangxuan;XU Xuefang;HAN Dongying(Key Laboratory of Measurement Tech and Instrument of Hebei Province,Yanshan University,Qinhuangdao,Hebei 066004,China;School of Vehicles and Energy,Yanshan University,Qinhuangdao,Hebei 066004,China)
出处
《计量学报》
CSCD
北大核心
2024年第5期738-746,共9页
Acta Metrologica Sinica
基金
河北省自然科学基金(E2020203147,E2022203093)
中央引导地方科技发展资金(216Z4301G)。
关键词
信息采集
数据恢复
轴承故障诊断
张量分解
缺失值数据
特征提取
振动测量
风电机组
information collection
data restoration
fault diagnosis of bearing
tensor decomposition
missing value data
characteristic decomposition
vibration measurement
wind turbine generator
