论文部分内容阅读
本文基于时序分析中的Euelide距离、残差偏移距离、Mahalanobis距离,Itakura距离和kullbaek-Leibler信息量等五种判别函数,针对齿轮振动信号的特点,提出了频域的FFT-AR分类方法和倒频域的CEP-AR分类方法。将这两种特征提取分类方法同时用到上海SH 760A型轿车变速箱的状态分类与识别中,实验结果表明:后者比前者误判率低,CEP-AR特征能更确切反映齿轮工作状态,它适合齿轮状态的自动分类识别。文章最后还讨论了影响分类识别精度的几个主要因素。
Based on the five discriminant functions of Euleride distance, residual offset distance, Mahalanobis distance, Itakura distance and kullbaek-Leibler information in time series analysis, this paper proposes a frequency domain FFT-AR classification method based on the characteristics of gear vibration signals CEP-AR classification method in the scrambling frequency domain. The classification and identification of these two feature extraction and classification methods are also used in the state classification and identification of Shanghai SH 760A sedan transmission. The experimental results show that the latter has a lower false-positive rate and the CEP-AR feature can more accurately reflect the gear working state, It is suitable for automatic identification of gear status. Finally, the article also discusses the main factors that affect the classification accuracy.