Identification of neural dynamic models for fault detection and isolation: the case of a real sugar evaporation process

Towards Robustness in Neural Network Based Fault Diagnosis

“…The first layer includes neurons with IIR filters while the second one consists of neurons with FIR filters. In this case, the second layer of the network is not a hidden one, contrary to the original structure of locally recurrent networks (Patan and Parisini, 2005). The following result presents approximation abbilities of the modified neural network: Theorem 1.…”

“…Neural networks also have high computation rates, substantial input error tolerance and adaptive capability. These features allow applying neural networks effectively to the modelling and identification of complex nonlinear dynamic processes and fault diagnosis (Marcu et al, 1999;Patan and Parisini, 2005).…”

“…In this situation, the neuron should have an ability to represent the dynamics. One way out of this problem is to use dynamic neurons (Patan and Parisini, 2005). Dynamics in this neuron are realized by introducing a linear dynamic system-an IIR filter.…”

“…3(b). To evaluate the quality of the proposed solution, another decision making technique based on constant thresholds is also applied (Patan and Parisini, 2005). Decision making using constant thresholds is illustrated in Fig.…”

“…3(a). In both methods, a number of false alarms represented by the false detection rate r fd was monitored (Patan and Parisini, 2005). The achieved indices are r fd = 0.012 in the case of adaptive thresholds and r fd = 0.098 in the case of constant ones.…”