An ANN fault detection procedure applied in virtual measurement systems case

The prompt detection of anomalous conditions of the Virtual Measurement Systems (VMS) elements, as sensors or transducers, involve a specialised implementation of the VMS software part. One solution for this software implementation is based on the neural processing structures. The implemented Artificial Neural Networks (ANN) are supplied with the voltage signals delivered by the conditioning circuits of the VMS sensors. The signal acquisition was performed using a data acquisition board or a programmable voltmeter. For the acquired signals the ANN delivers the values which can be used for fault detection and localisation of faulty elements. Referring to ANN architectures a study concerning the number of layers, the number of processing neurons, the type of neuron activation functions and the possibilities to optimise those parameters is included in this paper. The performance of proposed ANN fault detection solution was experimentally evaluated in the particular case of a VMS based on a data acquisition board (DAQ) and on a GPIB controller