基于特征评价的ADMOW模式识别方法及其应用.PDFVIP

第38 卷 第4 期 噪 声 与 振 动 控 制 Vol 38 No.4 2018 年8 月 NOISE AND VIBRATION CONTROL Aug. 2018 文章编号：1006-1355(2018)04-0192-06 基于特征评价的ADMOW 模式识别方法及其应用 张 建，潘海洋，郑近德，潘紫微 （安徽工业大学 机械工程学院，安徽 马鞍山 243032 ） 摘 要：针对以往模式识别方法的不足及特征数据存在“异常值”导致的模型失真问题，提出了基于特征评价的优 化加权代理判别模型（Agent discriminate model based optimization weighted ，ADMOW ）模式识别方法。该方法根据同 一状态类别中各特征值之间的对应关系（不同类别有不同的对应关系）建立数学预测模型，然后计算各特征值的类相 似度评价指标，根据评价指标对特征值进行加权处理，从而削弱“异常值”对模型的影响，建立更加准确的代理判别模 型，提高分类准确度。实验结果表明，经过特征加权处理的ADMOW 方法对滚动轴承的状态识别具有更高的识别率。 关键词：振动与波；ADMOW ；特征加权；异常值；滚动轴承；故障诊断 + 中图分类号：TH 113；TH 165 .3 ；TN911.7 文献标志码：A DOI 编码：10.3969/j.issn. 1006-1355.2018.04.037 Study and Application of ADMOW Pattern Recognition Method Based on Characteristic Evaluation ZHANG Jian , PAN Haiyang , ZHENG Jinde , PAN Ziwei ( School of Mechanical Engineering, Anhui University of Technology, Ma ’anshan 243032, Anhui China ) Abstract : In view of the shortcomings of the traditional pattern recognition methods and model distortion due to the outliers in the feature data, a pattern recognition method of agent discriminate model based optimization weighted (ADMOW) was proposed on the basis of characteristic evaluation. According to the corresponding relation between the characteristic values in the same category, the mathematical prediction model was established. Then, the similarity evaluation indexes of different eigenvalues were calculated. Meanwhile, the eigenvalues were processed with feature weighting according to the evaluation indexes so that the influence of the outliers was weakened. Fin