Enhanced Data-Driven Fault Diagnosis of Chemical Process via Information Fusion in Multiple Neural Networks and Andrews Plot

An enhanced data-driven fault diagnosis approach combining multiple neural networks and using Andrews plot for information pre-processing is presented in this paper. The system first applies Andrews plot to the process data to extract features and the normalized features are used as the neural network inputs. To overcome the non-robustness of an imperfect single shallow neural network, multiple networks are combined using information fusion for enhanced fault diagnosis performance. The suggested scheme is applied to a simulated CSTR process and the results show that the suggested scheme gives more robust and faster diagnosis than the conventional neural network based fault diagnosis scheme.