基于ILNS-SVDD的多工况过程故障检测应用研究

Research on application of fault detection based on ILNS-SVDD in multi-state process

为了提高支持向量数据描述(Support Vector Data Description, SVDD)方法在多工况过程故障检测中建模的准确性,提出了改进的局部近邻标准化(Improved Local Neighbor Standardization, ILNS)和SVDD结合的过程检测方法。首先寻找每个样本的第一近邻样本,再寻找第一近邻样本的局部前k近邻集,用近邻集的均值和标准差进行数据标准化,然后对标准化数据利用SVDD进行数据检测。改进的局部近邻标准化方法能够将多模态数据融合为单模态数据,建立更为准确地SVDD模型,提高了SVDD多工况过程检测精度,通过数值仿真和半导体数据实验,验证了ILNS-SVDD方法的有效性及优良性。

In order to improve the accuracy of the Support Vector Data Description(SVDD)method in multicase process fault detection,a process detection method based on Improved Local Neighbor Standardization(ILNS)and SVDD was proposed.First,look for the first neighbor sample of each sample,find the local top-k nearest neighbor set of the first neighbor sample,use the mean and variance of the neighbor set to normalize the data,and then use the SVDD to perform data detection on the normalized data.The improved local neighbor normalization method can fuse multi-modal data into single-mode data,so that the data satisfies the application conditions of SVDD,and the ILNS method improves the detection capability of the SVDD multi-operation process.Through numerical simulation and semiconductor data simulation experiments,the effectiveness and feasibility of the method are verified.