In an iterative simulation process, the main problem or difficulty is to use the existing results to understand its dependency on certain variables (design or just numerical). This difficulty, often causes us to rerun the simulations in a more controlled environment. Let me illustrate this with a simple problem in which we have some v1-v10 variables that we are changing to improve r1-r10 responses in some fashion. Unfortunately, the v1-v10 variable magnitudes are very hazy at the beginning in many cases so we incrementally change them over a certain span of time. At the end of this period, if we were to answer the question of ‘what is the effect of v1 against r1’, we naturally are not comfortable to do this unless the entire simulations were done by an optimization tool.

One recent practice that has given me tremendous confidence to answer above question is to run a ‘pack’ of simulations and keep them aside for later retrieval. In a recent barrier development project, there are a number of variables that were constantly changing. One particle variable of interest was the shear yield of the honeycomb that was rather difficult to obtain. In one controlled ‘pack’ of simulations, while keeping every other variable constant, a 3 reasonably varying values of the shield yield was picked and its effects on key responses was quickly understood. This “pack” of simulations can then be often “put-away” such that we can come back at a later time to restudy the effects of the variable with confidence since no other variable was changed in the input file. For a multi-variable problem, LS-OPT is a fantastic tool. Since the number of runs is a function of the number of variables while the turnaround time depends on the size of the problem, it is often important to use a fairly small model that can best duplicate the original problem. In the honeycomb barrier problem, the block of 300,000 elements was easily duplicated with just a few thousand element problem.

Not sure if I conveyed the idea but if I can summarize it it would be to work in a ‘pack’ of similar simulations in a controlled environment to easily asses our findings with confidence. Contrary to this approach, if we were to make meaningful conclusions from a random set of results, the task of finding the changes between them is rather difficult if the changes were significant.