In a coupled structural-acoustic system which has a box structure and interior air the position of a defect included in the structure is identified by using the frequency response of sound pressure level in the interior sound field. For the analysis of coupled structural-acoustic system Finite Element Method (FEM) is utilized. The box including a defect is excited by dynamical load and the frequency response of the interior sound are learned by Self-organizing Maps (SOM) as an input vector and the map is drawn in the 2-dimensional plane. The frequency response for the model whose defect position is not known is given to the trained map, and the identification is carried out. As a result the good accuracy of the identification is obtained. Furthermore, it is confirmed that the identification by SOM is better than by the other neural network such as LVQ (Learning Vector Quantization) or BNN (Back Propagation Neural Network).

View full abstract

  Stability conditions and equilibrium points of the Kovalevskaya's top are decided by equilibrium point analysis of the non-dimensional Euler's equation. Two parameters which regulate an amount of disturbance of initial value are introduced into the angular velocities of the equilibrium points, because two harmonic oscillations exist near them. Then, non-dimensional Hamilton's canonical equations which are not separable, and their initial conditions are derived. The symplectic method is adopted for numerical integration followed by a compensation process. It modifies the values of the canonical variables so as to cancel the errors of the Hamiltonian only. Error of the Kovalevskaya integral is automatically controlled.

View full abstract

  Renormalization molecular dynamics method (RMD) is one of coarse-grained molecular dynamics methods to overcome limitations of calculation cost. The number of molecules is compressed by applying renormalization group conversion to the Hamiltonian, and simulations are possible at any scales by controlling the number of conversions. The main advantage of this method is that there is no need to remake the inter-molecular potential function of new coarse-grained molecules. By using this method, however, we cannot simulate hybrid systems that include molecules at different scales. In this study, we propose a simple scheme to perform hybrid molecular simulations with MD and RMD. In this method, we find the nearest N MD molecules from each RMD molecule and define them as closest MD molecules. We place a virtual RMD molecule at the center of gravity of the closest MD molecules, and calculate the interaction force between RMD and virtual RMD molecules. To test our method, we applied MD, RMD and MD-RMD Hybrid simulation methods to argon liquid and gas. Then, we calculated radial distribution function and viscosity, and compared the results. The results of the MD-RMD hybrid were similar to those of all-atom MD, and the simulation time was shortened to 1/4.

View full abstract