摘要
针对线性鉴别分析在提取非线性特征时不能取得很好效果的问题,提出了基于遗传编程的线性鉴别分析方法。首先利用遗传编程对传统的时域指标进行特征提取,得到更能反映信号本质的复合指标,然后通过线性鉴别分析提取最佳特征向量,作为识别特征输入多类支持向量机,实现了对机器不同类型故障的识别。实验表明,经过基于遗传编程的线性鉴别分析提取的特征对轴承的故障具有很好的识别能力,进而提高了多类支持向量机的分类准确性。
Aim. GP help us to make LDA method effective for extracting nonlinear features. In the full paper, we explain our LDA in detail. In this abstract, we just add some pertinent remarks to listing the two topics of explanation. The first topic is： the extraction of composite features through re-organizing initial parameters with its classification ability. The second topic is： LDA analysis. In the second topic, we explain that LDA selects and constructs the best feature vectors that accurately describe the mechanical signals of machine; the feature vectors are then input to multi-class support vector machines to recognize the machine faults of various kinds. Finally we conducted experiments on roller bearings using the features extracted with three different methods： （1） the generalized discriminant analysis of Ref. 1 by Baudat et al; （2） the nonlinear discriminant analysis （NDA） of Ref. 2 by Roth et al; （3） our effective LDA. The experimental results do show preliminary that： （1） our effective LDA method is much better than the other two methods and （2） the classification capability of multi-class support vector machine is improved with the nonlinear features extracted with our LDA.
出处
《西北工业大学学报》
EI
CAS
CSCD
北大核心
2007年第3期363-367,共5页
Journal of Northwestern Polytechnical University
关键词
故障诊断
特征提取
支持向量机
遗传编程
线性鉴别分析
fault diagnosis, feature extraction, support vector machine, genetic programming（GP）, linear discriminant analysis （LDA）
