Model-based synthesis for diagnostic neuro-classifiers

The paper presents certain approach, based on model (group of model-based methods), in the synthesis of neuro-classifier used to design a diagnostic system of an exemplary process model. As learning patterns for neural network the input/output signal samples from the process were used (method III) as well as the parameters of ARX model (method I) and eigenvalues of process model (method II). There were considered two types of the process, i.e. operating nominally (without any faults) and in the presence of some selected faults. The considered diagnostic system with neuro-classifiers has been tested to verify the correctness of its actions towards fault detection and isolation.