443825 Nonlinear Partial Least Square (PLS) Methods with Generalized Likelihood Ratio Test (GLRT) for Fault Detection and Diagnosis of Chemical Processes

Wednesday, April 13, 2016: 1:30 PM
339B (Hilton Americas - Houston)
Chiranjivi Botre1, Majdi Mansouri2, Hazem Nounou2, Mohamed Nounou2 and M. Nazmul Karim3, (1)Chemical engineering, Texas A&M university, College station, TX, (2)Texas A&M University, Qatar, Doha, Qatar, (3)Chemical Engineering Department, Texas A&M University, College Station, TX

Nonlinear partial least square (PLS) methods with generalized likelihood ratio test (GLRT) for fault detection and diagnosis of chemical processes. 

Chiranjivi Botrea, Majdi Mansourib, Mohamed N. Nounouc, Hazem N. Nounoub and M. Nazmul Karima

a Artie McFerrin Dept. of Chemical Engineering, Texas A&M University, College Station, Texas 77843, USA

b Electrical and Computer Engineering Program, Texas A&M University at Qatar, Doha, QATAR,

c Chemical Engineering Department, Texas A&M University at Qatar, Doha, QATAR


Partial least squares (PLS) is an input output technique which can be used for fault detection and diagnosis in chemical and biochemical processes. In this work we have developed nonlinear PLS methods to tackle the nonlinear chemical processes. Kernel partial least square analysis (KPLS) and neural network partial least square (NNPLS) are applied with generalized likelihood ratio test (GLRT) for fault detection. Radial bases functions and polynomial functions are used as kernels in KPLS. The prediction ability of KPLS is compared with the multikernel PLS, in which rather than using one kernel function, multiple kernels are used in linear combination for better predictability of the output.

Highly correlated variables in the input matrix is of less importance to the model as we cannot extract any additional variability from it, hence we have perform pretreatment on input data by partial correlation analysis to remove the highly correlated variables from the input matrix. Contribution plots are used for fault diagnosis; contribution plots compares the residuals of the variables which result in faults. We have used mean square error values to compare the prediction power of the models. Fault detection performance and prediction power of the developed models are illustrated through Tennessee Eastman process problem, which can be used to simulate wide variety of faults occurring in a chemical plant based on Eastman chemical company. 

Keywords: PLS, GLR, KPLS, NNPLS, fault detection, statistical process control, partial correlation analysis, Tennessee Eastman process.


1.      FouziHarrou, Mohamed N. Nounou, Hazem N. Nounou and MudduMadakyaru, “Statisticalfaultdetectionusing PCA-based GLR hypothesistesting” Journal of Loss Prevention in the Process Industries 26 (2013) 129e139.

2.      S. Joe Qin and T.J. McAvoy. "Nonlinear PLS modeling using neural networks." Computers them. Engng, Vol. 16, No. 4, pp. 379-391, 1992

3.      Zhang, Yingwei, and S. Joe Qin. “Improved Nonlinear Fault Detection Technique and Statistical Analysis.” AIChE Journal 54, no. 12 (December 1, 2008)

Extended Abstract: File Uploaded