In this paper, the wavelet-sigmoid basis neural network(WSBN) proposed
by authors earlier is with expert system(ES) for dynamic fault diagno
sis(DFD). Meanwhile, a two-level integration strategy(TLIS) for DFD is
also presented. At the first level of TLIS, a WSBN detects the fault
sources through data from the dynamic processes, and outputs the corre
sponding fault degrees. At the second level, ES interprets the results
from the WSBN and predicts the time period for fully developing the f
aults and the product qualities as well as the reactor states at the m
oment when the fault is fully developed by using a dynamic simulation
package. If the predicted values exceed their acceptable ranges, then
ES gives some related proposals for removing the fault causes or cance
ling their effects by using the reasoning context consisted of product
ion rules. DFD is applied to a hydrocracking process showing the effic
iency of the TLIS.