Optimization is a powerful tool, allowing engineers to discover better designs quickly and efficiently. Here, we look at 5 tips to ensure that your optimization studies yield the best possible results in a short timeframe.
Tip #1: Simplify the simulation as much as possible
In order to explore a large number of designs it’s important that the simulations are as efficient as possible. Think about what aspects of the design are most important, and try to cut out unnecessary details. Typically, we look to:
- Apply symmetry wherever possible to reduce simulation size
- Defeature CAD to remove unnecessary details
- Use 2D approximations where appropriate
Treat the optimization study as an opportunity to explore a wide range of designs, and remember that the good candidates can be followed up with a more detailed simulation to fully investigate their performance.
Tip #2: Find a small set of meaningful KPIs
Remember that the optimization process is driven by the outputs you select and how your fitness function generates a score from them. Trying to incorporate too many outputs into a fitness function can steer the optimization algorithm in the in the wrong direction, or even exclude designs which might actually be good candidates.
Instead, try to narrow down the outputs you analyse to one or two Key Performance Indictors – the things that are going to differentiate your product from your competitors’.
Where it’s difficult to decide how to balance the value of one output versus another consider multiobjective optimisation as a method for calculating a Pareto front, which gives the optimal trade-off between outputs.
Tip #3: Consider your choice of input variables carefully
There’s no getting around it: adding input variables dramatically increases the size of the problem space. This means you’ll need to run more simulations to arrive at your goal.
When choosing design parameters to vary consider whether they influence the KPIs you’re measuring. If you’re not sure, run a small design sweep to check. Also, remember that many optimisers allow the range of each input variable to be constrained, allowing you to limit the study to designs which you consider practical.
Finally, consider how your choice of input variables will affect your ability to compare device performance. For example, changing the surface area of a resonator will alter its capacitance. It may be desirable to keep the area of your devices equal throughout the study to facilitate direct comparison of designs.
Tip #4: Use an integrated workflow
Manually transferring information between different packages is time consuming and ties up valuable engineering time. An integrated working environment will allow you to setup, execute and analyse your study with minimum overhead.
A good example is FlexConnect which allows Matlab® to talk direction to PZFlex. FlexConnect’s functionality makes it easy to filter simulation outputs to extract the best KPIs and to analyse the results of a study.
Tip #5: Use cloud compute resources when you become compute constrained by local resources
From time to time you’ll run into problems that are large, with no obvious routes to simplification. In these cases it can be tempting to cut the optimization short to reduce total study time, or to simply skip the study and go with the design that you have. However, neither of these options are ideal, as it’s often the most complex design problems which require the most analysis.
A better solution is to push simulations to the cloud, allowing the ability to run multiple jobs in parallel on fast hardware. With PZFlex Cloud an entire population of designs can be simulated in parallel, significantly reducing the time taken to perform population based optimization studies.