一种不等长的多模态间歇过程故障检测方法

摘要

提出一种不等长的多模态间歇过程故障检测方法。首先，运用局部加权算法对不等长批次数据进行预处理。在训练样本中确定不等长数据的最大可保留长度，利用 k 近邻信息，通过加权重构出不等长批次缺失的数据点。其次，对等长的训练集构造局部近邻标准化矩阵，运用 K-means 算法进行模态聚类，使用局部离群因子方法确定第一控制限，并剔除离群样本。最后，对各个模态建立 MPCA 模型并确定第二控制限。根据各个模态控制限的匹配系数计算统一的统计量和控制限，在统一的控制限下进行多模态故障检测。将提出方法应用于半导体工业过程，仿真结果表明，与传统的故障检测算法相比，本文算法提高了故障检测率，验证了该方法的有效性。%A fault detection algorithm for uneven-length multimode batch processes is proposed. First, the local weighted algorithm is used to preprocess the uneven-length batch data. In the training sample, the maximum retention length of uneven-length data is determined. Using the k-nearest neighbor information, the missing data points are reconstructed by weighting reconstruction. Secondly, the local neighbor normalized matrix is estimated for the training set of equal length. The K-means algorithm is used to classify the models. The local outlier factor method is used to determine the first control limits and remove outliers. Finally, the MPCA model is established and the second control limits are determined for each model. The unified statistics and control limits are calculated according to the matching coefficients of the control limit of the various models. The multimode fault detection is carried out under the unified control limits. The algorithm is applied to the semiconductor industrial process. Simulation results show that the proposed algorithm improves the fault detection rate relative to the traditional fault detection algorithms. The effectiveness of the proposed method is verified.