Multivariable process monitoring using nonlinear approaches

The use of principal component analysis (PCA) for process monitoring applications has attracted much attention recently. The idea of compressing the process data into a few factors facilitates and simplifies the identification of an abnormal operation condition. Nonlinear factors obtained by the implementation of neural nets enhance this reduction specially in processes with broad operation conditions. This paper summarizes and compares the techniques used to obtain nonlinear factors. It also discusses the advantages of using nonlinear PCA for monitoring and calculation of confidence regions