Statistical process monitoring using multiple PCA models

Principal Component Analysis (PCA) has been successfully used to build a multivariate monitoring model for the process usually with one operation stage. However for processes with more than one operation stages, building a single PCA model to monitor the whole process operation performance may not be efficient and will lead to a high rate of missing alarms. To remedy this situation, a monitoring strategy using multiple PCA models is presented in this article based on soft-partition algorithms. The framework of utilizing multiple PCA models to monitor continuous processes is also introduced. The application to a three-tank plant demonstrates the effectiveness of the method.