Simulation-based parameter identification of a reduced model using neural networks

Simulation models are used to study and design real systems in order to optimize their performance. However, the lack of complexity in models, as compared to systems, leads to differences between simulation and actual behavior. Identification of the effective system parameters in the simulation models would reduce the discrepancies between the model and the real system and results in better simulation of system behavior. In this paper, radial basis neural networks are proposed for the identification of the effective parameters of the machine tool feed drive system. Neural networks are first trained with an estimated set of model parameter values and corresponding step responses of the position control loop. After the training period, the response of the model with the identified parameters is compared to the system step position response at different axis positions for validation. An application of neural network to other trajectories is done on a ramp function. The obtained results reveal considerable potential of neural networks in identifying accurately the system parameters and in reducing the discrepancies between experimental test results and the model.