Fault detection of multimode process based on local neighbor normalized matrix

In recent years, a variety of fault diagnosis methods of multimode process have been developed. However, these multimode fault diagnosis methods need to assume that the data of batch production process is even-length, and there is no pollution in the data. To obtain better monitoring performance in a batch process with uneven-length data, a fault detection algorithm of multi-mode process based on local neighbor normalized matrix (LNNM) is proposed in this paper. The method highlights the contour features of various modes, accurately captures the nonlinear position relationship between modes and within modes. The local weighted algorithm (LWA) is first used to preprocess the uneven-length batch data. Then the main local neighbor normalized matrix is constructed for the training set of equal length. The K-means algorithm is used to mode clustering. In each mode, the local outlier factor (LOF) method is used to determine the first control limits for removing outliers. The MPCA model is established and the second control limits are determined for each mode. Furthermore, the matching coefficients of the control limits of each mode are calculated, and the unified statistics and control limits are determined. The fault detection of multimode process is carried out under the unified control limits. The algorithm is applied to the actual industrial semiconductor process. Simulation results show that the proposed algorithm improves the fault detection rate relative to the traditional fault detection algorithms. The effectiveness of the method is verified. (C) 2016 Elsevier B.V. All rights reserved.