Novel calibration technique for precision walker dynamometer system based on artificial neural network

Abstract

A back-propagation artificial neural network model with three layers was developed for the calibration of precision walker dynamometer system. This model adopted the output voltages from 12-channel strain gauge bridges in the dynamometer system as the network input vector and the six load components as the network output vector. The neuron number of the single hidden layer in this model was optimized by comparing the absolute error summations under the target error. Relevant error check results showed that, after the calibration using this neural network, the maximal system precision error with single-direction force was 7.78% and the maximal crosstalk was 7.49%. In comparison with traditional linear calibration method, the proposed technique can effectively increase the measurement precision of walker loads and greatly decrease the crosstalk error, which might be helpful for accurately monitoring and evaluating the rehabilitation training effect of walker-assisted walking in the future.