Data-driven Real-time Status Monitoring Of Batch Process

This paper investigates the application of the multivariate statistical process monitoring and control technology, which employs both Multiway Principal Component Analysis (MPCA) and Kernel Density Estimation (KDE), to real time status monitoring and fault diagnosis of batch production processes. KDE is a non-parametric method which is capable of extracting the population's Probability Density Function (PDF) based on data sample only without any a prior knowledge about the statistic properties of the data regime. In this thesis, it is conducted the implementation of the KDE for monitoring the performance of batch production processes.The conventional real-time monitoring method does not use the non-parametrical PDF of the principal components, which are capable of indicating the real-time changes of batch production processes. To improve the monitoring sensitivity, it is for the first time to propose the utilization of the PDF of the principal components in real-time to monitor batch production process.Besides the above-mentioned efforts, this thesis also applies the PDF of the principal components to detect the faulty batches of the batch process and the result is encouraging, which provides a new option for process inspection.