I integrated IPOPT as an external optimizer, again still for the mechanical compliance minimization. I do not use the Hessian of the Lagrangian but let IPTOP to an L-BFGS approximation. The software is really nice - good documentation, open source with a good license (CPL) and a good C++ frontend with cool features.
The results are also good for 2D, the number of iterations is roughly comparable to the optimality criteria and there are about 50% more function evaluations for line search (determine the step length).
Bad news are, that in 3D the performance is much worse - but I still did not do examine what is really going on. Also I had to increase the mechanical load by a few magnitudes - otherwise I had nearly instant convergence with only slight modification of my inital guess for the desing parameters. I'll have to implement the break conditions by myself - what is supported by IPOPT
No comments:
Post a Comment