PROCESSING SIGNALS FROM AN INERTIAL SENSOR USING NEURAL NETWORKS

An inertial sensor is a device for determining the location of a part by measuring parameters related to its inertia. This paper describes a neural-network-based method for processing signals from an inertial s ensor to compute the orientation of a part. This method involves train ing a multi-layer perceptron (the most commonly applied neural network type) using the backpropagation algorithm to model the operation of t he sensor by mapping its natural frequency of vibration to part locati on information. The paper details the experimental procedure for acqui ring data to train and test different neural network models and the re sults obtained. The latter show that the proposed method yields orient ation values with a comparable degree of accuracy to existing techniqu es while requiring much less computational effort. Copyright (C) 1996 Elsevier Science Ltd