论文部分内容阅读
针对半导体工业过程多工序、变量非线性、非高斯分布等特征,提出一种基于扩散映射的K近邻(DMKNN)故障检测方法.充分利用扩散映射(DM)降维,提取低维流行特性,保留数据集内在非线性结构特性,应用改进的KNN故障诊断方法在低维流行特征空间进行检测.研究结果表明:与传统K近邻技术的统计方法相比,DMKNN的故障检测率高于其他算法,提升了对数据样本关联性信息的有效提取能力,保持了K近邻处理非线性、多模态检测问题的性能,验证了该方法的有效性.
Aiming at the characteristics of multi-process, non-linear variables and non-Gaussian distribution in semiconductor industry, this paper proposes a method to detect K-nearest neighbor (DMKNN) based on diffusion mapping, which makes full use of reduced dimensionality of diffusion map The data set has inherent nonlinear structural characteristics and is tested in low-dimensional prevalent feature space by using improved KNN fault diagnosis method.The results show that compared with the traditional K-nearest neighbor statistical methods, DMKNN has higher fault detection rate than other algorithms, In order to effectively extract the relevancy information of data samples, the performance of K-nearest neighbor processing for nonlinear and multimodal detection problems is verified, and the effectiveness of the proposed method is verified.