An application of a progressive neural network technique in the identification of suspension properties of tracked vehicles

Abstract

The paper demonstrates that a progressive neural network (NN) technique can be applied effectively for identification of suspension properties of tracked vehicles. A three-dimensional multi-body tracked vehicle is firstly modeled with an advanced ADAMS Tracked Vehicle (ATV) toolkit. The displacements of roadwheels are selected as inputs for the NN model and the outputs are parameters that can describe suspension properties. The NN model consists of two-hidden-layer neurons connected between the input and output neurons and is trained with a modified back-propagation (BP) training algorithm. After the initial training, the suspension parameters are characterized by feeding the measured displacements into the NN model. The NN model will go through a progressive retraining process until the displacements of roadwheels obtained by using the characterized parameters is sufficiently close to the actual response. Simulation results show that the identification procedure is practically feasible to solve such an inverse problem in the suspension systems of tracked vehicles.