Detection and Diagnosis of Faults in a Four-Tank System using Artificial Neural Networks

This paper proposes an artificial neural network-based method for fault detection and diagnosis of a MIMO four-tank process in a non-minimum phase. The approach considers two stages: a model output estimation error stage, where a nonlinear autoregressive exogenous neural network is used to model the system, and a fault detection and diagnosis stage based on the model output error estimates, where a standard feed-forward neural network is used to classify the kind of fault. Faults due to added noise and parametric changes are combined as a benchmark to be detected in real-time to evaluate this proposal. Therefore, the whole system considers two setpoint inputs, four transfer functions, two NARX neural networks, four feed-forward pattern recognition networks, and four outputs, each associated with a specific fault.