Neural model of the dynamic behaviour of a non-linear mechanical system

The availability of a model to prepare, supervise and analyse vibration tests of complex non-linear mechanical systems is vital. Analytical models, however, have some limitations when it comes to integrate the complex mathematical and physical parameters of non-linear mechanical systems. In this paper, an original approach based on multilayer perceptron neural network with a time regression input vector is proposed as an alternative solution. An experimental set is designed for this purpose: a non-linear mechanical system is subjected to random, shock and swept sine excitations on a vibrating platform. Its mechanical response is measured with accelerometers. The model is trained and validated based on the raw experimental data. Finally, simulation results from the analysis are presented and discussed.