Local model based KPLS with application to fault detection of batch process

To handle the problems of nonlinearity, dynamic and multiphase in batch processes, a novel online modeling and monitoring method based on local model and kernel partial least squares (KPLS) is proposed. The improved local modeling method not only utilizes the information of the current time slice, but also incorporates the information of the nearby time slices. The local model can reflect the changes of operating process, so it can be used to monitor processes with time-varying behavior and operating phase change. Besides, KPLS is employed to model the nonlinear information of the process. A case study of the semiconductor process illustrated that the proposed approach can effectively monitor the batch process.