Are neural network techniques the solution to measurement validation, monitoring and automatic diagnosis of sensor faults?

Abstract

The appropriateness and feasibility of using artificial Neural Network (ANN) techniques to facilitate improved in-service performance of micromachined acceleration measuring devices is questioned in this research and its possible extrapolation to sensor fault diagnosis is attempted. Two examples of closed loop neuro-transducers are given: a micromachined accelerometer with capacitive pick-off and a neural network controlled tunnelling accelerometer. Based on the success of the ANN control method as applied to sensors, the authors investigate the possibility of developing self-diagnosis sensors based on ANNs and a strategy of such development is proposed.