A new fault detection algorithm based on weighted differential principal component analysis(PCA) was proposed for studying multimode and nonlinearity of industrial production processes. The closest neighbor xfof the original data sample and the closest k points in front of xfwere first selected. The mean value mjand the weighted value wjof the j points in front of xfwere calculated.The weighted differential method was then used to preprocess the original data to eliminate multimodal and nonlinear characteristics. The principal component analysis algorithm was applied, and the loading matrix P and the control limits of detection indexes SPE and T2were calculated to establish the PCA model. Finally, the test data (after preprocessing based on weighted differential PCA) was projected onto the PCA model to calculate the detection index. Fault detection was carried out by detecting whether the detection index exceeded the control limit. The proposed method was applied for numerical examples and semiconductor manufacturing processes to verify its effectiveness.%针对工业生产过程的多模态和非线性特性,提出了一种新的基于加权差分主元分析的故障检测算法.首先选取原始数据样本的最近邻xf以及xf的前k个近邻, 分别计算出xf的前j个近邻样本的均值mj和权值wj, 利用加权差分的方法对原始数据进行预处理,剔除多模态和非线性特征;然后利用主元分析法(PCA)计算出负载矩阵P以及SPE和T2检测指标的控制限,建立PCA模型;最后将待检测数据运用加权差分法预处理后投影到PCA模型上计算检测指标,通过检测指标是否超过控制限进行故障检测.将该方法应用于数值例子和半导体生产过程来验证其有效性.
展开▼
