Partial fault diagnosis in a chemical plant using artificial neural networks

The use of artificial neural networks (ANNs) for partial fault detection is explored. Fault detection is a type of pattern recognition problem. An ANN represents nonlinear correlations between its inputs (measured process variables) and its outputs (faults) and acts as a pattern recognizer. As a test case, a realistic multicomponent distillation column is studied. Tbe column is simulated by using a full-order rigorous model with tray-by-tray calculations. The neural network successfully diagnoses the faults if is trained upon. In addition, if the network is trained by using data representing the full complexity and nonlinearity of the system, it is then able to generalize its knowledge to detect novel partial fault patterns.