Authors: E.I. Gaura, N. Ferreira, R.J. Rider and N. Steele
Affilation: Coventry University, United Kingdom
Pages: 513 - 516
Keywords: micromachined accelerometer, neural network, model reference control
In this paper, a closed-loop, smart transducer design is proposed, based on artificial neural network (ANN) techniques. The design aims to improve the performance of open-loop, off-the-shelf capacitive acceleration sensors and increase their robustness to manufacturing tolerances. A "model reference" control strategy was adopted for the design of the smart transducer. Multilayer perceptron (MLP) type networks were chosen for implementing the control strategy. While a static MLP was used for the feedback arrangement, a tap delayed lines MLP was necessary for implementing the controller due to the dynamic nonlinear behaviour exhibited by the sensing device. A dynamic version of the back-error propagation algorithm was used for training the networks. The resulting closed-loop transducer had a dynamic range of ±10g and a stable behaviour for input stimuli up to ±100g.