论文部分内容阅读
基于最优分类线的概念,提出了一种新的模式识别分类器构建方法——判别域界面几何法.该方法利用BP神经网络的高度非线性,将模式类样本数据从高维输入空间映射至二维判别域空间后,采用多边形中轴提取方法,构造模式类间隙多边形的中轴线,延伸至整个二维判别域空间,生成模式类决策边界.以铁路货车车轮用双列圆锥滚子轴承的故障诊断为例,介绍了判别域界面几何法的应用过程.结果表明,判别域界面几何法能在二维判别域空间上给出各不同故障模式类之间明确的界限,这就给操作者直观判断故障模式类别提供了条件.
Based on the concept of optimal classification line, this paper proposes a new method of constructing the pattern recognition classifier, ie, the discriminant domain interface geometry method. This method uses the high degree of nonlinearity of the BP neural network to map the data of the pattern class from high dimensional input space After two-dimensional discriminant domain space, using the method of polygon middle axis extraction, the central axis of the model-like interstitial polygons is constructed and extends to the entire two-dimensional discriminant domain space to generate the model class decision boundary. The double-row tapered roller bearing , This paper introduces the application process of the geometric method of discriminative domain interface.The results show that the discriminative domain interface geometry method can give a definite boundary between different failure modes in two-dimensional discriminant domain space, The conditions for visual judgment of the failure mode category are provided.