Acceleration Approaches Of The Computation For Plastic Injection Molding Simulation

The research on acceleration approaches of the computation for injection molding simulation includes a wide range of contents. The improvement of mathematical model, optimization of numerical algorithm and enhancement of computer hardware all are ways and means to accelerate the computation. This paper concentrates on mathematical model, numerical algorithm and parallel computing in plastic injection molding simulation.A BP neural network based estimation algorithm is proposed for determinating cooling time in the cooling simulation. It decreases the maximum error of the estimated cooling time to 30% in contrast to 80% while the empirical formula is employed. Thus the outer iteration number can be decreased. An acceleration method based on splitting of the coefficient matrix is proposed for boundary element method based cooling simulation. A majority of operations can be transferred to outer iteration steps from inner iteration steps and thus the simulation is accelerated. Unlike the direct rounding method and the combination method, the splitting method avoids accuracy loss.The parallel computing of the cooling simulation is also discussed in this paper. An efficient parallel computing scheme for calculating coefficient matrix elements is developed and the parallel Jacobi right-preconditioned GMRES(m) iterative solver is employed to solve boundary equations. The numerical results show that the parallel Jacobi right-preconditioned GMRES(m) iterative solver convergences more quickly than the SOR solver and the proposed parallel computing algorithm of the cooling simulation achieves a speed-up of 1.6 on dual-core computers. Moreover, a parallel SOR iterative method is proposed for solving large dense linear systems and it can achieve a practical parallel speed-up on both multi-core computers and small scaled computer clusters.The methods for solving large sparse linear equations are also discussed, the non-stationary iterative method is thought as a class of efficient solver for sparse linear equations in plastic injection molding simulation and the preconditioning technology is the focus in this thesis. SSOR preconditioner, ILU preconditioner and sparse approximate preconditioner are compared studied and are employed along with the non-stationary iterative method to solve the linear equations in filling, packing and warpage simulation. The following conclusion is drawn: the ILU(l)-PCG solver and the SSOR-PCG solver convergence quickly in solving symmetric linear equations and the ILU(l)-GMRES(m) convergence quickly in solving nonsymmetric linear equations.The mesh simplification is proposed as an approach for balancing the computational time and the computational accuracy. A triangular mesh simplification algorithm with geometric feature recognition is presented. The mesh quality is an important aspect considered and surface fitting is employed on the model surface, the simplification algorithm can achieve a great simplification ratio while the computational accuracy is mostly preserved. The numerical results show that the error of the injection pressure can be limited within 4% and the filling analysis time can be reduced to 1/7 while the simplification percentage reaches 75% on products with small geometric details.