04 May, 2023
Having now done several Acoustic Topology Optimization projects for clients over the last year, I have had some learnings that are not necessarily revealed when working with acoustic topology optimization in a more academic setting.
Targets for the objective functions are not necessarily given by the client.
The client will express in words some of their wants and needs, but it is up to you to work with that, ask the right questions, and formulate the mathematical expressions for the objective function. This may require a lot of back and forth between the consultant and the client, and investigations have to be made into individual weighting of targets, and targets potentially working against each other.
Most industry cases have to be considered in 3D
When looking at topology optimization in academia, 2D is very often used to investigate new implementations, test cases, and so on. But in the industry it is not that common that 2D acoustics can be utilized, and even 2Daxi symmetry, which otherwise can be relevant for loudspeakers, will not be of use when the targets are to be evaluated in 3D. If you have true mirror symmetries, take advantage of them.
Meshing is a restricting factor
From the above it follows that many mesh elements will be needed for the 3D cases being investigated. Acoustics involves wave propagation, typically over most of the audio range, as opposed to academic cases where often only a single or a few frequencies are considered. And the topology optimization domain must be resolved enough to capture this as well as the overall geometry, plus it might be advantageous to have even more elements here when considering manufacturability of the final design. This, however, can cause cases to take such a long time to run that it is not feasible to reach a design and it is also difficult to track the direction an on-going simulation is taking when progress is so slow. So you have to find the right balance of element number and order to have a process that can work you in practice.
Acoustic Topology Optimization is an iterative process.
Here, I don't mean that you are using an iterative solver, but rather that each case is unique, and it is important to look at what the solver is trying to do along the way, and not just start it and leave it be until the run finishes. So while sometimes you can do other things while a simulation is running, an optimization may require you to sit and observe what the solver is trying to do, and perhaps you need to stop it as soon as you see it going astray in some way. For academic cases you can typical just mesh very finely, run fairly quickly, evaluate the final result, and run it again with some modified settings, but that is not likely to fly for these wave propagation cases with complex geometries and many targets over a wide frequency range, where some trial-and-error is needed to build a 'feel' for the case, since the case itself is likely new to you. That is also why you should of course start out simulating without any optimization, and make sure that the initial results match measurements.
Standard settings may not suit your industry case
When doing topology optimization, there are lot of different settings that can be adjusted regarding the solver, filters, and so on. For simpler cases, you might not see much of difference when changing these, but for industry cases, the right combination of parameters can be pivotal to getting a robust design out in the end. Get ready to experiment with these settings, also some of the ones buried deeper in the software.
The latest version of the software is probably the preferred version
Advances are constantly being made in the simulation/optimization software framework, and it is likely that the latest version has some improvements regarding optimization that can benefit you, both regarding the optimization itself but also the testing of the final design. There are cases, however, where the setup has to kept in a certain version, but you can still afterward test run time and end results in the latest version for comparison and learnings for the next case.
Testing the optimized field can be a hassle
While solving you optimization case, you will end up with some design variable field, which is continuous, although you are trying to force it to be binary; meaning either air or a hard-walled structure for acoustic topology optimization. So, to make sure that the final optimized design actually works somewhat as indented, it is necessary to test it. But this is not necessarily that easy for industry cases, as the design variable field needs to be converted to a some entity that can be interpreted as a geometry in a subsequent study, and certain selections, boundary conditions, and so on may have been lost, and a new mesh is also likely needed. This is illustrated below with a case from COMSOL's Application Library Model (topology_optimization_2d_room), where a new mesh and general setup is made for the testing of the topology optimized results.
This is not a big deal for this particular 2D example, but for 3D cases you likely have to for example manually stich meshes together and these operations have to be done for each topology optimization run, and for each relevant filter value. So for 3D cases I have implemented a much quicker way of assessing the optimized results that requires less manual setup, and the setup migrates through files, so that once it is implemented, I don't have to put any more effort into it for subsequent runs. It is not trivial to implement, but it has certainly been worth the initial effort.
The client cannot be expected to do this testing, as that would require them to for example 3D-print each potential design candidate with whatever manufacturing tolerances and measurement uncertainties that come with that. So make sure to have an efficient test procedure that you can trust.
NDAs must of course be respected
This is probably one of the only downsides of being a consultant; not being able to show off your work, while knowing that it has results that are well worth publishing. But the client is the most important part in any consultancy work, and so as a consultant you need to keep an airtight operation and not reveal anything that the client is not on board with.
I find, however, that many clients are actually interested in getting their work out, and I for example have audioXpress articles coming very soon done together with clients. I see such collaborations as being very beneficial for all parties involved, and I hope to do more of these going forward.
A note on pricing
If the quasi-testing is relevant to you, reach out and we can discuss prices, but it will probably be a couple of days of work to make sure that it is set up correctly for you and your needs. If you are interested in learning how to do acoustic topology optimization in general, I offer a training course, preferably running over a year's time, with teaching material, meetings, general cases, and focus on your specific cases, for an approximate price of 40-50,000 USD based on a break-even of me doing 3-4 of these optimization cases for you without any knowledge transfer regarding the setup. This training has to planned well ahead of time, so if you are ready to take the jump into this disruptive acoustic design technology in 2024, now is the time to start planning.
