基于高维随机矩阵特征值之差的滚动轴承状态异常检测算法

A rolling bearing state anomaly detection algorithm based on the difference of high-dimensional random matrix eigenvalues

针对滚动轴承异常检测准确性差、精度低及数据维度灾难造成检测困难等问题,提出一种基于随机矩阵特征值之差指标的滚动轴承状态异常检测算法。运用平移时间窗对不同时刻的轴承信息锁定,并通过分段、随机化、扩增和维度重构等方法构造出高维随机特征矩阵;利用随机矩阵理论对高维数据良好的处理能力,给出了滚动轴承特征值之差指标的构造方法及动态检测阈值的数学公式,可降低噪声的干扰,提高检测指标的鲁棒性与检测结果的准确性。采用辛辛那提大学智能维护系统(intelligent maintenance system, IMS)中心的滚动轴承全寿命数据进行应用研究,分析了不同误警率对检测结果的影响;从指标构建、阈值设定及异常检测等方面,将特征值之差算法与特征值之比算法进行比较。结果表明,最大、最小特征值之差算法中检测指标构建及阈值设定更符合实际工况,对滚动轴承异常状态检测更准确,对早期异常状态的识别更敏感。

A rolling bearing state anomaly detection algorithm based on the difference index of random matrix eigenvalues was proposed to solve problems of poor accuracy, low precision, and difficulty of detection caused by data dimension disasters in rolling bearing anomaly detection. The bearing information at different times was locked by using a moving time window, and a high-dimensional random feature matrix was constructed through methods such as segmentation, randomization, amplification and dimensional reconstruction. The use of random matrix theory has a good processing ability of high-dimensional data, and the construction method of the difference index of the rolling bearing eigenvalues and the mathematical formula of the dynamic detection threshold were provided, which can reduce the interference of noise, improve the robustness of the detection index and the accuracy of the detection result. By using intelligent maintenance system(IMS) rolling bearing full-life data for application research, the impact of different false alarm rates on the detected results were analyzed. From the perspective of index construction, threshold setting and abnormal detection, the difference between the eigenvalue algorithm and the eigenvalue ratio algorithm were compared. The results show that the construction of the detection index and threshold setting in the algorithm of the difference between the maximum and minimum eigenvalues are more in line with the actual working state, more accurate detection of abnormal state of rolling bearings, and more sensitive to the identification of early abnormal state.