MPCA Fault Detection Method Based on Multiblock Statistics for Uneven-length Batch Processes

In the present work, multiway principal component analysis (MPCA) fault detection method based on multiblock statistics (MBS) is developed for handling uneven-length batch processes. The first step is to obtain equal length batch records (stage I) and the uneven-length stage (stage II). Stage I is divided into p-1 smaller even-length sub-blocks. Stage II is regarded as a sub-block. The uneven-length batch data is divided into p smaller sub-blocks. The mean and variance of all the sub-blocks are calculated, and combined into a feature vector to represent the batch processes. With the proposed method, batch processes with durations of uneven-length can be effectively monitored using MPCA. The proposed method is applied to the monitoring of Al stack etch process. The experimental results show that compared with MPCA, k-Nearest Neighbor rule (kNN) and kNN based on MBS, the fault detection rate of the proposed algorithm is the highest. The corresponding fault detection time is less than 0.4s, so it has good fault detection performance.