Random vibration analysis of vibro-impact systems: RBF neural network method

The recent success of the radial basis function neural networks (RBFNN) method for random vibration analysis of smooth systems fuels in speculations that this approach may be extended to non-smooth problems. However, very little is known on the applicability of this approach to non-smooth systems. This work generalizes the RBFNN method to the randomly excited non-smooth vibro-impact system (VI-S). We first transform the non-smooth VI-S system to a continuous nonlinear system. Then, the solution of the reduced Fokker–Planck–Kolmogorov (FPK) equation for the transformed VI-S is expressed in terms of the RBFNN with Gaussian activation functions. The weights of the RBFNN are determined by solving an optimization problem to minimize the reduced FPK equation residual subjected to the constraint of the normalization condition. Three examples are presented to demonstrate the validity of the suggested scheme. Several remarks on the solution process also presented. All the results confirm the applicability and validity of the RBFNN method in dealing with the randomly excited non-smooth VI-S.