By Jared Conway
Usually this isn't the question that we get on support or when talking to our customers about optimizing their analysis workflows, instead our customers ask “Why does this analysis takes so long to solve?” or “What hardware do I need to make this analysis solve faster?"
While you can throw hardware like SSDs and faster CPU cores at your analysis, eventually there are diminishing returns and by far the biggest impact you can have on solve time is optimizing your analysis setup, starting with the level of detail you include in your model.
To start what I recommend is looking at where you are in the design cycle. This should generally define the level of detail you’re shooting for and level of accuracy you should be looking to get out of your analysis. In addition to that, it should set some expectations about solve time. For example if you’re working on a design concept and solve times are very long, you’ve probably included too much detail in your model or made your setup too complicated. Conversely if you are finalizing a design, shorter solve times shouldn't necessarily be expected but by applying the recommendations below you will ensure that your solve times are as optimized as possible.
|Stage||Definition||Model||Physical Accuracy||Solve Time|
|Design Concept||“Will this idea work?”||Main Components||Broad Accuracy||Low|
|Design Solutions||“How can we refine the design further? How will it behave under different conditions?”||Main Components and Coarse Details||Broad Accuracy but Consistent Between Tested Designs||Low to Medium|
|Design Candidate||“Before finalizing the design, will it pass certification?”||Main Components and Fine Details||Minimal Assumptions||Medium|
|Finalized Design||“Design is complete and is either certified or manufactured.”||All Components and Details Included||Near Absolute||High|
Once you have a broad understanding of the level of detail your model should have based on the stage in the design cycle that you’re in, the next step is to figure out exactly which components should stay and which should go. The ones that you will want to keep should fall into one of these 2 categories:
- The load path. What components need to be part of your model to ensure that your loading condition is characterized? In structural analysis that means the path from your force to your restraint, in thermal analysis where your heat goes in to where it goes out and in flow analysis which components create the flow path.
- Measuring locations, areas of interest and areas of concern. In all likelihood, these are the features or components that you’re really looking to evaluate, so they are a given and need to be part of your analysis. But this also helps you hone in on the scope of your analysis. If you’re analyzing an electronics enclosure within a vehicle, your area of interest is around that electronics enclosure, it is unlikely that you need to model the whole vehicle.
Beyond that, we get into the nitty gritty of idealization and simplification best practices when it comes to the coarse and fine details. For example it is always recommended to remove small features, holes, fillets, gaps…etc and suppress components that aren’t of importance. And similarly it is always best to use the proper idealization like shells and beams or smart parts like bolt connectors or fans rather than discretely modeling them. But if you’ve followed the recommendations above, your work here should be minimal and if you run into something that is going to be a bear to idealize or simplify, you’ve probably made up for it elsewhere such that it is ok to keep it around and not significantly affect the solve time.
So that’s it right?
Well, almost. There really aren’t any more steps for optimizing the level of detail in your model but if you have been following along closely there is one loose end that needs to be wrapped up. In the back of your head you should be thinking, “Great, I can decrease solve times but what about the accuracy of my analysis? Doesn’t removing these components, features and details decrease the accuracy of my results relative to the physical/absolute behavior of my design?”
The answer to that question is yes but if you’ve done a good job planning your analysis and followed the recommendations above, the impact on accuracy will be minimal and any deviations are easily explained by assumptions in your final report. Worst case, as you move through the design cycle, as more information becomes available and more details are added to the model, you have will have a solid, optimized analysis foundation to build upon and can easily add accuracy and detail without significantly increasing solve time.