基于AR自相关峰态值的一类轴承故障检测方法  

FAULT DETECTION FOR ONE CLASS OF BEARINGS BASED ON AR WITH SELF-CORRELATION KURTOSIS

作　　者：陶新民[1] 杜宝祥[1] 徐勇[1] 吴志军[1] 

机构地区：[1]哈尔滨工程大学电子通信工程学院,哈尔滨150001

出　　处：《振动与冲击》2008年第2期120-124,136,共6页Journal of Vibration and Shock

摘　　要：针对轴承故障检测系统中异常样本数据不易收集以及异常样本数据分布不均导致传统分类算法出现过适应现象等现实应用问题,提出了一种基于自回归(AR)模型自相关系数峰态特征的一类故障检测方法。该方法利用正常样本生成AR模型参数,其他样本在该模型的投影形成残差序列,计算残差序列的自相关系数并取其峰态特征作为相似性的度量。实验结果表明该方法能有效地克服以AR模型参数为特征计算复杂度高且检测性能易受样本大小影响的不足。同时,文章给出了单一故障诊断模型并提出基于粒子群优化算法的阈值设定决策方法。实验中将本方法同其他以AR模型为特征的多层感知机(MLP)及自组织映射(SOM)方法进行比较,实验结果验证了本文建议方法的正确性和有效性。Since it is not easy to collect abnormal samples in bearing fault detection and there exists an overfitting of conventional classifications due to the abnormal data imbalanced,a novel one-class detection model based on AR model with self-correlation kurtosis characteristic is presented.The AR model is employed to extract the normal samples' parameter characteristics and consequently the normal samples' AR sub-space is established.The self-correlation of errors resulted form other samples' being projected onto the AR model space is calculated.The kurtosis of the previously calculated self-correlation is used as the metric of similarity with the normal sub-space.The experiments show that the proposed approach can efficiently overcome the drawbacks that the computation is very complex and the detection rate is very sensitive to the length of samples of conventional detection method based on AR parameters.The single and multiple fault detection schemes based on AR with self-correlation kurtosis also are proposed.The model's threshold-value-setting and its determination approach based on the particle swarm optimization are investigated.The proposed detection schemes are compared with MLP and other detection techniques based on AR parameters in the experiments.The results illustrate effectiveness of the proposed techniques.

关 键 词：故障检测 AR模型 自相关系数 峰态特征 粒子群算法 多层感知机 

分 类 号：TM711[电气工程—电力系统及自动化]