[Kvothe28]

Hello dear community,
I'm currently in my last Master's semester (mechanical engineering) and I'm taking an entrepreneurship course in which we have to build a theoretical business as an assignment. Since I specialized in CFD simulations during my studies, I chose CAE software as my business model. For this I now have to work out a business-finance plan.

Unfortunately, I currently have great difficulty in estimating the costs of developing CAE software. Unfortunately, I have not found any figures in the annual public financial statements of the larger CAE providers that help me.

Do any of you have any knowledge about how much money has been invested in the development of CFD software before it went live for the first time? Are there any estimates for this? I would appreciate any advice or guidance on what to look for.
For CAE software I was thinking of something like Star-CCM+ from Siemens or Ansyl's Fluent. So user friendly software with a graphical user interface (GUI).

Thanks Kvothe

[sbaffini]

Interesting question, but nonetheless difficult to answer without a more specific framing.

When considering Star, or Fluent, you are talking about software built from the ground up in the 80's (Star being later rewritten is just a detail in this reasoning), with heavy lifting of most things, including original theoretical developments and visualizations. IT world was also kind of different, with no internet, no relevant open source, etc.

As of today, the theoretical recipes are clear, there are commercially friendly open source options for most things, and software development hasn't been smoother.

Still, the list of features the early codes have today is really huge, which is not something you can do more easily just because of advancements in IT.

Then, if we exclude all the remaining costs, a relevant measure to consider would be: what are the development costs of a mvp in the CFD field which doesn't use third party software for the core business? For example, using vtk for visualization but not OpenFOAM for the solver?

My answer to this question is 6 to 8 skilled people (who already work in this field doing this exact job) for 2 years. This will include pre-post, gui, solver and all the infrastructure to take that code from those experienced 8 and leave it in the hands of some other team and go straight in production.

I'm here considering the right 8 people, of course, with the optimal overlapping of skills for all areas, so that no single area of the code is exclusive competence of a single developer. Also, of course, the optimal distribution of skills that one can expect from people in this field, so that they actually cover all the needs.

Then you have to figure out their cost. Some rough estimates of the minimum salaries might be 50k-60k euros per year in europe per person, and 100k-120k usd in usa. Let's round this to 1 Mln$ per year, so you have a 2Mln$ figure.

This is pure speculation of course, and you might end up spending 1 or 4 Mln, but not 200k and not 10 Mln.

The only thing missing here is the definition of mvp in CFD. I consider as such a fully working and tested code with similar performances to the competitors, something that could be sold the day after to more than few selected users of competitor software.

[flotus1]

That estimate seems low when specifically comparing against well-established solver packages offered by Ansys and Siemens.
Let's not forget that these are multi-physics solvers, with a whole zoo of physical models. Models which can interact with each other in multiple ways, and each of them probably with several different numerical formulations. And a lot of fine-tuning behind the curtains to ensure stability. The mere fact that theoretical recipes are readily available, doesn't mean the implementation into a commercial solver is straightforward.
And then there are the meshing and geometry pre-processing capabilities...

Wit 8 star developers (the kind of people that cost more than 100k/year) working 2 years, you can certainly get something that some customers would pay money for. But compared to the likes of Fluent or CCM+, the scope would need to be narrowed down a lot.

[sbaffini]

Originally Posted by flotus1

That estimate seems low when specifically comparing against well-established solver packages offered by Ansys and Siemens.
Let's not forget that these are multi-physics solvers, with a whole zoo of physical models. Models which can interact with each other in multiple ways, and each of them probably with several different numerical formulations. And a lot of fine-tuning behind the curtains to ensure stability. The mere fact that theoretical recipes are readily available, doesn't mean the implementation into a commercial solver is straightforward.
And then there are the meshing and geometry pre-processing capabilities...

Wit 8 star developers (the kind of people that cost more than 100k/year) working 2 years, you can certainly get something that some customers would pay money for. But compared to the likes of Fluent or CCM+, the scope would need to be narrowed down a lot.

I was probably too implicit in that regard. So, not as a counter argument to your reasoning, but just speaking out loud, what I had in mind with mvp (minimum viable product) was:

- multi-zone capability with fluid flow and heat transfer for both solid and fluid, maybe some basic radiation

- steady/unsteady incompressible/compressible flows with inviscid/laminar/RANS capability

- passive scalars, multiple reference frames and a basic mechanism for user coding (like user defind functions or expressions a la Fluent)

So, no combustion, no multi-phase and/or particles, no moving meshes, etc., just a straight CFD solver with basic functionality in terms of models but, still, fully functional in that regard.

In the same way, I was not considering a full fledged mesh generator, but something like Snappy, VisCart, Harpoon or Hexpress. And a very limited subset of what paraview can do for post processing.

Also, note that I explicitly referred to people who already do this as a job. What I meant with this is that they exactly know how to do stuff for an actual commercial solver, including every detail regarding the fine tuning.

But again, that was a rough estimate. The same group of people might end up needing 4 years instead of 2, or you might need to spend more on some superstar developer than the rest of the team, or maybe hire the 9th or 10th junior developer for some time. But I think that anything north of such 4 Mln figure, round it to 5 Mln at most, is overspending for the kind of product I was thinking of.

Obviously, I was thinking of a lean startup setup, a group that works with no interruption or managing whatsoever because they all exactly know what to do and how to do it. Let's say, some sort of ideal conditions.

If we want to also consider some heuristic productivity measure to double check these figures, let's say each developer codes for 200 days per year (leaving the rest for group meetings etc.) at 50 lines of code (LOC) per day, this means 10k LOC per year, very low indeed. So, allowing for some flexibility, let's say we expect each developer to write in 2 years something between 20k and 50k LOC. Would be more if allow for more than 2 years of development time. This then puts our software in the 160k-400k LOC range, which I think is compatible with the sort of product I am thinking of, according to the codes I have seen and considering the use of open source libraries like Qt, VTK and OpenCascade.

[LuckyTran]

Quarterly statements of publicly traded companies are public if you know where to look. You won't find how much money Gambit or Fluent burned during their startup phase, but you can get a quick grasp of how much these companies spend today to maintain their software.

In the case of Ansys, they have a dedicated page for their recent reports.

[andy_]

Originally Posted by Kvothe28

Hello dear community,
I'm currently in my last Master's semester (mechanical engineering) and I'm taking an entrepreneurship course in which we have to build a theoretical business as an assignment. Since I specialized in CFD simulations during my studies, I chose CAE software as my business model. For this I now have to work out a business-finance plan.

The good times for commercial CFD business plans was in the 80s and 90s. Today the software is mature with, I suspect, poor prospects for a new general purpose CFD code but there are likely to be opportunities in specialised areas if you can find the right niche and then grow that.

Originally Posted by Kvothe28

Do any of you have any knowledge about how much money has been invested in the development of CFD software before it went live for the first time? Are there any estimates for this? I would appreciate any advice or guidance on what to look for.
For CAE software I was thinking of something like Star-CCM+ from Siemens or Ansyl's Fluent. So user friendly software with a graphical user interface (GUI).

Much of the methodology in the STAR CFD code was developed for a decade or two in Gosman's group at Imperial College in collaboration with others like Raad Issa's. This continued after the creation of CD. The code would have been written externally but the cost of writing software is minor compared to the cost of gaining CFD knowledge. It is also minor compared to the cost of supporting clients, maintaining software and running a commercial CFD company. Not sure how one would value the prior activity at Imperial College but it was significant.

The Fluent CFD software initially came out of Swithenbank's group in Sheffield and unlike Gosman's group they were not a well established CFD software group. I believe they started by adapting Imperial College CFD software and their GUI was written by Creare rather than themselves. The GUI seemed to play a leading role in their early success which surprised me and probably some of the other CFD companies with better CFD modelling software. They also got into some trouble for using university resources for commercial activity so initial investment was likely a bit blurry at the start.