Neural network application for fault diagnosis in FCCU

In this paper, the application of an ANN to the fault diagnosis of a fluidized catalytic cracking unit (FCCU) is studied. The ANN based system successfully diagnoses the fault it is trained to recognize. It is also able to generalize its knowledge to diagnose fault combinations it is not explicitly trained to recognize. The network can also handle incomplete data. One important development in this paper is the use of object oriented programming techniques for software development. The advantages in using OOPs for such an application is expanded. An analysis of the recall capability to trained faults and the generalization capability to symptoms resulting from novel fault combinations is attempted. The generalization proficiency versus network topology is examined