Welcome to the Onshape forum! Ask questions and join in the discussions about everything Onshape.
First time visiting? Here are some places to start:- Looking for a certain topic? Check out the categories filter or use Search (upper right).
- Need support? Ask a question to our Community Support category.
- Please submit support tickets for bugs but you can request improvements in the Product Feedback category.
- Be respectful, on topic and if you see a problem, Flag it.
If you would like to contact our Community Manager personally, feel free to send a private message or an email.
Onshape and simulation
caradon
OS Professional, Mentor Posts: 300 PRO
I have to admit it...
One of the most tantalizing aspects of Onshape to me is the prospect of cloud-based, on-demand simulation.
In my office, I haven't been able to push (traditional) simulation tools (yet).
There are three reasons for this:
This begs a few questions:
SimScale is a very interesting cloud-based simulation platform. However, it costs €200/month (1 year commitment). I imagine this is a whole lot more than what the base subscription for Onshape will be.
So I'm very interested in how Onshape will be pricing simulation (and for that matter other computationally expensive tools like rendering as well...).
I think for simulation, rendering... paying for actual use (as measured in hours, minutes...) makes most sense. This is the 'render farm' business model, really.
What do you think?
Dries
One of the most tantalizing aspects of Onshape to me is the prospect of cloud-based, on-demand simulation.
In my office, I haven't been able to push (traditional) simulation tools (yet).
There are three reasons for this:
- High cost (initial + maintenance)
- Lack of expertise
- Relatively low use: we might only need simulations a couple times a month.
This begs a few questions:
- Is Onshape envisioning offering CAE/FEA tools?
- What kind of tools would that be? Linear analysis? Nonlinear analysis (Yes please!)? Thermal?...
- What shape will it take? Fully integrated? Partnering with an external cloud simulation service?
- And most importantly: What will it cost? Or what will the pricing model be?
SimScale is a very interesting cloud-based simulation platform. However, it costs €200/month (1 year commitment). I imagine this is a whole lot more than what the base subscription for Onshape will be.
So I'm very interested in how Onshape will be pricing simulation (and for that matter other computationally expensive tools like rendering as well...).
I think for simulation, rendering... paying for actual use (as measured in hours, minutes...) makes most sense. This is the 'render farm' business model, really.
What do you think?
Dries
2
Comments
It's not practical (pronounced "affordable") to keep a big compute farm running all the time for occasional usage. Would you be willing to wait 10 minutes for a large compute job to start? If you had to optimize for cost or time, where would you compromise?
-John
Now it gets interesting...
I'm interested to see what other people think...
Dries
I think what makes a lot of sense is for Analysis to lead design as much as possible. These guys seem to have the makings of something interesting (http://www.solidthinking.com/ProductOverview.aspx?item=Inspire Overview&category=Products)... If the analysis could be done quickly, then It could more lead the geometry decisions rather than lag it. In the current set of tools you have to imagine/conceive the geometry first - then analyze. You end up only analyzing what you conceived of - even if its part of an iterative process where you conceive, analyze, revise the concept, analyze, etc... If you could give loose endpoints and parameters, then the analysis could yield the most efficient load paths and create geometry only where needed. This might lead to some radical geometry that we would not initially conceive of and shapes that maybe only 3D printing could produce - but I think this is where we are going. I'm sure we'll find that designs will then more replicate nature as nature is intricately more efficient than what we have been able to conceive of from scratch.
-John
Ideally, this tool would let you do:
- Simple linear static sims - no large displacements, etc.
- Simple fluid flow
- Low-part-count motion sims
I don't know if this would actually be cost-effective, but potentially you could offer a delivery date for free simulation results (for example, 24 hours out) instead of having the sim start on demand. Then, the simulation gets run when the servers are idle/not working on paid jobs (unless the delivery date gets close, in which case the free simulation runs regardless)I hope Onshape will finally allow me to do simulations.
Dries
There are a lot of cloud CAE options right now, but none are as quick as a locally hosted solution that a dedicated analyst can, and will, justify spending today. Until that changes, it needs to be designed for the casual analyst...
For them, it needs to be easy to use, exist in an embedded environment, and be fast. Stupid fast.
Linked[in]
Also... Keep it simple, stupid.
Dries
1. High capital and on costs - we would need non linear high deflection analysis for medical plastic parts. We are looking at say £6k plus £2k a year.
2. Lack of use. We want this functionality for INTERNAL use only. I would not sell it as a service as we are not analysts and would not wish to sell such a service. This is purely for our own internal quality assurance, so what we design will be as right as it can be. Incidentally we do the same for 3D printed parts on our U Print. We don't charge for this, they are printed to prove out our designs.
3, Lack of expertise. I am a degreed mechanical engineer. I did an FEA course. My major project was in FEA related areas, yet I don;t feel confident enough to get results from an FEA system. FEA is all about assumptions. get them wrong and your results are crap. CAD vendors like to pretend it is all easy but maybe they should be on the reciving end of a £6k report from a leading UK design consultancy on the structural integrity of a product done in solidWorks Premium (linear static, low deflection, limited mesh options). Like I said, crap in, crap out.
4. all FEA systems are for specialists. Prove me wrong. Show me a FEA system where I can say "sit on the seat, tell me how much it deflects"...yes they do this...after you make all the assumptions, restraints, etc. In other words, you need to know what you are doing.
So what do we need?
Easy. Full multi physics FEA/CFD (non linear, high deflection etc) in the monthly charge in all packages. Like rendering. One price. If FEA and so on is to be used, it needs to just be there at all times. I don't want to be worrying about paying per minute.
Better interface. MUCH better interface. I have tried most FEA systems on the market and they share one feature. They are all archaic, difficult to use and require assumptions. FEA needs to represent real life scenarios and offer real time feedback to be of real use in a design phase. we are a long way of that yet.
So if Pnshape said to me we are offering one monthly cost, for all features, no hidden extras, then I'd say, yes. If it is trying to modularise things then what is different from the existing options?
One monthly cost would be very clear pricing, but how sustainable is such model?
Take f.i. LAGOA. They offer a free plan. You get 100GB storage and unlimited rendering time. How can they do that?? (Well, sort of a redundant question now they're under the Autodesk umbrella...).
A fixed cost for computationally expensive tools would seem unbalanced to me. Occasional users will pay too much. Power users will pay 'too little'.
Dries
The winners in this game will be the companies that have the tech in house, who can bolt on FEA, CFD, DFM etc as well as CAD. Autodesk can. Dassault can. Seimens can. PTC can..everyone else licenses the tech. What Onshape have is a delivery platform for licensed technology. Either they sell that to others or they find the new start ups who offer new ideas in FEA etc who they can persuade to join with them in their journey.
I think if we are being honest here we have to recognise that Onshape must be looking for the "out" at some point. The question is who will provide the out. Not Dassault!
I think Dassault won't get themselves into another Parasolid debacle.
I agree about the challenge of competing with 'ulbrella corporations'.
Yet I also believe there are still lots of opportunities to go 'solo slim'. There are still smaller (yet very agile) companies that won't sell their soul to the devil. Luxion (KeyShot) is one such example. (Although it doesn't qualify as small anymore...)
Dries
I see two movements in this day and age:
- Consolidation (huge getting huger) in an attempt to snatch up technologies. This has been going on for a few decades with increased momentum in the last few years.
- More recently a dynamic/explosive growth of cloud services in all markets.
For me, the excitement lies in interconnecting many of those cloud solutions. I think extraordinary stuff can happen, but you need a proper architecture/platform. I think Onshape could be that platform (if they do everything right...). At least it's the best candidate I've seen so far.Who cares if the technology is licensed or not? I think on the developer side the great appeal of SaaS/cloud lies in how easily functionalities can be integrated (API) AND interchanged.
Not wanting to sound like a fanboy, I'm just a believer.
Dries
At the moment I would like to see Onshape flying with drawings, surfacing, document management, document properties, expanded feature tools, fastner library's before getting a FEA, but hope it will be added in down the track.
Twitter: @onshapetricks & @babart1977
Also, rarely do companies start with non-linear FEA, let alone multiphysics. Remember, implementation costs include training, and you gotta crawl before you walk, let alone run.
@babart77 - I couldn't agree with your last sentence more. Spot on. Once OS has proven itself a CAD service provider, then it must get into FEA. The back-end solvers are cheap (free?) nowadays, so as they continue down their current path it will be an obvious next step. The CAD needs to be dialed in first, though.
I agree with @DriesV that the focus should be on how the pieces are all put together, who cares if the back end is built from scratch or licensed. As long as it is structured to always grow and be updated frequently, it will be able to adapt with the changing needs of the designers. This is one massive shortcoming of the current CAD providers (as David Corcoran noted in his webinar last Friday).
So... Right now, a designer still needs a couple of apps running concurrently: Email, Web Browser, Word Processing, Spreadsheet, CAD/CAE, MRP. Fast forward a couple more years, and the designer can do all of that with just a web browser open with a couple tabs: Google Docs & OS. Boom. Done.
Linked[in]
Dries
We have changed the way we are doing Solid works FEA, now we use a separate model in a structured FEA directory, sometime even export the model form Solidworks as a step then drop back in and director edit before running the FEA, obviously there is no link back to the original parts, so any design change have to be manually updated but do not think this a big issue as you generally have your head around what need to be done to the production part base on the FEA and have an understanding of the reason the geometry is slight different.
I don't see that Onshape needs a fully integrated FEA but a link into the document management for FEA and ability to set up and save instants of geometry for FEA with out interfering with production models, I can see Onshape branching and merging and direct editing working well for this. It would be good to have a one stop shop for pricing eg. add FEA on to your monthly subscription I think there would have to be different levels, hopefully you can just add as needed and not be locked if not required.
Twitter: @onshapetricks & @babart1977
I am happy to invest in anything that will yield returns. For my business FEA will be nothing more than an expensive luxury. We would only ever use it for in house verification. I would not contemplate selling it as a service due to PI issues with our work type. I have seen too many FEA reports from so called experts that I can see right through, that are based on poor assumptions and/or plain bad analysis techniques. FEA software is still very specialist and too costly.
@DriesV Odd turn of phrase there Dries about selling your soul. It depends very much on the financing structure of the company. We have all seen family run businesses who successfully carve out a niche and make a good living. But if you have VC or private equity financing a business then generally they are after a return at some point. How do you attract stars to a start up? By offering equity, and a plan to cash in that equity, whether that be a floatation or a buy out. Given the backing Onshape has I expect to see a sell off at some point. I don't know what the backing for Luxion is, but Im quite sure if Dassault offered them $100m they would be happy!
@KevinQuigley - I believe you. My radar just goes off when somebody says that cost is the #1 reason they don't use FEA...
Linked[in]
I would like to OS focus on geometry and getting this right.
I think 90% of the solidworks data out there in the world is crap and hope that we can fix this with organized, well structured data that anyone can work with. I hope that OS data lives throughout the life of a product for all departments in an organization. I'm tired of manufacturing destroying my assemblies and watching all the different phases my data takes to complete a product. I hope this gets cleaned up and OS comes up with a solution to this problem.
Let's get back to the simple concept: engineering layout->assemblies->parts which has never existed in parametric modelers and it would be nice to get back to this simple premise. This is where the focus needs to be.
I've done 1000's of analysis and never on customer data. I always rebuild the geometry from scratch. Customer data never meshes properly and time spent trying to mesh it is a waste. You're better off starting over with something simple and building on it.
As far as analysis is concerned, the meshing needs to be free. I don't want to have to pay for someone's solvers to understand my geometry.
Analysis has to be cheap because it doesn't pencil out if it doubles the cost of a project. Building models using classical engineering (pin and paper) and breaking 5 prototypes is still a more cost effective design approach for most products.
Calculators like what I'm imagining already exist in a number of different forms. Some examples include:
- Excel worksheets that I'm sure a lot of us have written in the past for personal use
- Diagramr, a simple beam-bending simulation applet that generates load, shear and moment diagrams (disclosure: written by a close friend of mine)
- The MITCalc series of Excel-based calculators. I've never actually used their calculators, but I do reference their support pages frequently (they have pretty good explanations of how to run various first-order-estimate engineering calculations): I'm guessing that the level of detail in their tools' calculations would be just about right.
In OS, the ideal implementation would be calculators like these that reference 3D geometry for input parameters. For example, in the beam-bending simulation example, you would select an extrude feature as your beam. The calculator would check to make sure that the aspect ratio of the "beam" was appropriate for the beam-bending calculation method it uses; let you select sketch geometry or faces to generate the beam section; let you create sketch geometry to apply loads and restraints to the beam; and then dynamically update the calculations and results as the geometry changes.I think SW tries to do something like this with the Beam Calculator, but I've never used it (looks pretty rudimentary). I can see how this could easily become a really buggy, frustrating feature, but if it's executed simply and robustly, it would be very helpful. Thoughts?
Meshing: The trend in CAE is that mesh creation/manipulation by the user is going away. It is on the way to becoming nothing but a background process. This is driven by the rising popularity of dynamic meshes (morphing, re-meshing, result-driven optimization), smarter "smart meshes", and of course the advancement of hardware performance.
Despite historic amounts of press, the current cloud CAE offering is still pretty bad. It's either VPN access or command-line based, both of which fall short in post processing.
Bottom line = There's massive opportunity here.
Linked[in]
However, I'd argue that broadly, current FEA tools are significantly harder to use than current CAD tools. Consequently, I'd imagine the development effort require to create an FEA package that meets the usability standards that OS seems to be striving to set will be commensurately greater. There is a massive opportunity here, like you say, but it's also a massive challenge. In the meantime, if OS can offer some other type of CAE tool - whether it's calculators like I described, or something else - that might be a selling point. Particularly for concept development, I've generally found that basic hand calcs with approximations are faster and more accurate than mucking around trying to do quick-and-dirty FEA.
I see there being a larger problem and it's about the geometry. Let's get this right. Do you think pro/e and solidworks are good solutions to bring products to market? Can it be better? With limited resources, we really need to get the geometry part correct.
I see a lot of reports/studies of shaded images and never see the mesh. I stopped using h-adaptive & p-element because I could mesh better and faster using my own controls. Now, meshing has probably gotten better, but I still want to see the mesh in all reports before I trust the results. Hopefully newer solvers still show how they are breaking the geometry up into elements.
Maybe newer solvers don't break the geometry into elements, is this true? Do you know of any good articles on revolutionary CAE techniques? Maybe there's a reason they changed the name from FEA to CAE.
@Bill - If a software has automation tools that are slower than doing it manually, that's a failure on their part in my mind. To me, being "powerful" in CAE is having both the capability and the understanding to manage your assumptions. Cheap CAE tools have limited capabilities, and the expensive ones have 1,000 page help files. There are several tools that sit in that awkward space in-between.
There is a general belief that meshing should be taken out of the user's hands. We humans just can't be trusted... The only reason a "coarse mesh" has been even offered was so you didn't wait forever just to find out you made a mistake somewhere else. As compute power continues to increase in performance and availability, default meshes will get finer and finer, until mesh-independent results are displayed on the screen as fast as the screen can display it. Or... "as fast as your internet connection can transfer it."
There is one slight hiccup though - and that is the human nature of curiosity. If a prudent designer is given 30 man-hours to "test" their design in FEA, and the software gave them an answer in 10, chances are they would think of 2 more "tests" to put it through...
Linked[in]
I'm really hope OS gets the geometry right for all departments in an organization. This should be a different thread. I'm really interested to hear stories from design engineers about what happens to their models as they get released. Most people strip all assy ref out, circle interpolate splines and some release by exporting parasolids to control manufacturing. I haven't seen a good process yet. I'm going to start another thread "What's the worst thing manufacturings done to your models".