A nonlinear dynamic principal component analysis (ND-PCA) approach is devel
oped in this paper based on dynamic PCA and the sigmoid basis function feed
forward neural network (SBFN). Through ND-PCA an integrated framework for
on-line monitoring and root-cause diagnosis is developed. The approach is v
erified and illustrated on the Tennessee Eastman benchmark process as a cas
e study while noises were added on sensor readings. Results show that the p
roposed ND-PCA approach performs good incipient diagnosis capability and ov
erall diagnosis correctness rate. (C) 2000 Elsevier Science Ltd. All rights
reserved.