Neural estimators of two-mass system optimized using the Levenberg-Marquardt training and genetic algorithm

This paper presents application of neural networks for state variables estimation of two-mass system. Two stages of applied design process can be distinguished: the Levenberg-Marquardt method was implemented for weights adaptation, moreover topologies of the neural models were optimized using genetic algorithm (number of the nodes in each hidden layer was determined). Analyzed estimators were tested in structure with state space controller. Simulation results, presenting robustness against changes of the plant parameters, are presented. Furthermore, implementation of neural estimators in programmable device (dSPACE1103 card) was done and experimental tests were prepared.