Level: Intermediate
The Model-Based Multivariate Control Chart (MDMVCC) platform creates control charts based on Principal Component Analysis (PCA) and PLS models, which can be used for fault detection and diagnosis in multidimensional data. Here, we demonstrate MDMVCC monitoring based on PLS models using the Tennessee Eastman process, a simulated chemical industrial process. In this simulation, quality and process variables are measured as a chemical reactor produces a liquid product from gaseous reactants.
First, we perform offline fault diagnosis, which can be done very easily with the MDMVCC platform, by alternating between multivariate and univariate control charts and process diagnostic reports. Next, we connect JMP to an external database and demonstrate online monitoring with the MDMVCC platform. Product quality variables have a time lag until all measurement results are available, so fault detection is often delayed. In MDMVCC based on the PLS model, the variability of quality variables is monitored as a function of process variables, which are generally available relatively quickly, making it useful for early fault detection.
Jeremy Ash holds a degree in Bioinformatics from North Carolina State University and currently works as an Analytics Software Tester for JMP. His dissertations include computational methods in chemoinformatics, chemometrics, and bioinformatics. He also holds an MS in Statistics from North Carolina State University and a BS in Biology from the University of Texas at Austin.
