Industrial Process Monitoring and Fault Diagnosis with Nonlinear Independent Component Analysis

Considering the nonlinear characteristic of date in real industry processes,a multivariable process monitoring method based on nonlinear independent component analysis(NICA) is presented.With the help of Bayesian theorem,process data can be reconstructed by establishing multi-layer perceptrons,and statistical model of process in mathematics can be given for monitoring in real time.The proposed method is applied to the Tennessee-Eastman(TE) process.Simulation results show its availability and advantage in the aspect of fault diagnosis.