Visualization of multivariate processes using principal component analysis and nonlinear inverse modelling

Interpretation of the state of industrial processes is considered using principal component analysis as a visualization technique. A procedure for using the resulting two dimensional maps in detecting upsets and faults of the prosess is described. Nonlinear inverse models from the map coordinates back to the original process variables are studied and compared to linear modelling methods. Visualization techniques together with inverse modelling methods are shown to form a useful decision support system for the operating personnel of the plant. The visualization techniques and inverse modelling are studied using a simulated chemical process as an example.