Fault Detection and Diagnosis Strategies for a Cogeneration and Cooling Plant

Cogeneration and cooling plants may experience unexpected trip or reduced performance caused by sensor faults, ageing, wear, leakage, fouling, erosion, cavitations and component malfunctions. In this paper two strategies -sensor fault diagnosis system and component or plant fault diagnosis system - are elaborated to help identify the cause for the said conditions. The first suggested approach is based on linear/nonlinear principal component analysis (PCA) both applying auto-associative neural network (AANN). The second, dedicated to component fault diagnosis is, based on a Multi-Layer Neural Network and/or neuro-fuzzy modeling. Data obtained from simulation model of a counter flow heat exchanger is used to demonstrate the first approach. The results indicated that with the AANN based PCA it is possible to reduce measurement noise and diagnose sensor bias and drift errors. The two approaches applied in sequence, they can provide robust fault detection and diagnosis system.