Issue #1,099 | Inside the Business of CAD | 7 June 2021
Nearly all analysis software works the same way: break up 3D models into many small pieces, and then do an analysis on each small element to find out when the model might fail. The concept goes back Boris Galerkin, who mused that just because something can be proven physically doesn’t mean it shouldn’t be proven mathematically.
And so in 1915 Mr Galerkin developed an approximation method using differential equations, following which Richard Courant developed the finite element method as a special case of Galerkin’s Method. That was 1943; today, FEA [finite element analysis] is overwhelmingly the most common form of model analysis.
IGA [iso-geometric analysis] is a different way of splitting 3D solids into those chunks before performing analyses on models. Mike Scott, Derek Thomas, and Kevin Tew pioneered R&D on IGA under the name of Isogeometrx LLC. IGA is based on U-splines (unstructured splines, patented by Mr Thomas). The company rebranded as Coreform when Matt Sederberg came on board in 2016 to commercialize IGA.
Then about a year ago, Coreform developed a new IGA formulation, generalizing the finite cell method. The idea is to adjust simulation speed and accuracy, coarser or finer, according to the needs of each stage of the product development process. By decreasing meshing time and eliminating steps, Coreform speeds up design and analysis. (See figure below.)
Coreform IGA handles both material and geometric nonlinearities, as well as nonlinear constraints, including contacts. For a given degree p, the bandwidth is the same but the fill-in is greater for smooth functions. Most customers define their own subroutines and material models through Julia, an open-source programming language. See julialang.org.
The following figure shows how Coreform fits into the panoply of FEA programs:
Before U-Splines
Before co-founding Coreform, Mr Sederberg commercialized the T-splines technology developed by his father, Tom Sederberg. T-splines were an improvement over meshing, which usually limits meshes to rectangles; T-splines meant that meshes could come to a stop before covering the entire surface. (See figure below.)
One catch is that it takes a lot more computing power, but today’s computer can handle it. Another, more serious catch is that unlike U-splines, T-splines were never invented to be used for analysis.
The stand-alone T-Splines software was bought by Autodesk in 2011, which integrated the technology into its Inventor and Fusion 360 programs – much to the disappointment of customers who had been using the CAD-agnostic version of T-Splines.
Cubit and IGA
Coreform offers two software products:
Coreform Cubit is the mesher using hexagonal and square meshes (see figure below). It cleans up poorly-made CAD models, and everything can be automated with Python scripting. With coarse bounding-box analysis, the result from Cubit is nearly instant.
The free version Coreform Cubit Learn for non-commercial use limits exports to 50,000 elements.
Coreform IGA implements IGA with linear and nonlinear solvers to solve static and dynamic physics made of nonlinear materials and contacts. Input files are in human-readable JSON [JavaScript object notation] format. Subroutines are implemented in Julia.
Coreform is especially useful for complex assemblies. The company is interested in working with early adopters to understand their needs.
Q&A
Q: Traditional FEA is many decades old and is very well known. How would one go about explaining IGA to management that is comfortable with FEA, but has no idea what a spline is?
A: Here’s what I tell them: Every traditional FEA mesh is a spline. However, it has historically been so intractably difficult to build “high-order, smooth splines on unstructured meshes” that it hasn’t been worth encoding traditional FEA meshes as a spline.
Any practical, commercially-viable simulation tool needs to have unstructured meshes; real world parts are unstructured and complex.
Academics have known for years the benefits that smooth splines provide in FEA. See the section titled “The piecewise polynomial spaces” at the bottom of page 8 of doi.org/10.1090/S0002-9904-1973-13351-8. The last sentence in the paper reads, “But the actual construction of the psi^i is genuinely an open problem,” and refers to the problem of creating smooth, high-order splines on unstructured meshes. U-splines are a solution to this problem.
Smoothness is often referred to as “continuity.” We would say that traditional finite elements are “continuous (C^0)” while IGA is “smooth (C^1, C^2, C^3...).”
In fact, for special cases, we already use smooth splines in FEA, we just give the elements different names like “Euler-Bernoulli beam element” or “Hermite shell element.” However, even these haven’t been recognized or implemented as splines.
What IGA does is implement the finite element method upon the recognition that all meshes are splines. This means that certain critical parts of the finite element method are generalized, such as integration and assembly. Critically, IGA is FEA, but with a whole new set of rich mathematical theory and tools that are more effective at solving FEA problems.
Q: How much of Coreform is based on T-splines?
A: T-splines are not part of the Coreform IGA solution. Our U-splines are a general form of spline designed to support simulation.
T-splines technology was a stepping stone to U-splines. Coreform’s technical founders learned from the shortcomings and positives of T-splines when they began developing our U-spline technology.
Q: Can we use user-defined materials (UMAT) in Coreform?
A: Yes, we have a fully defined user interface that leverages Julia. It can handle user-defined materials and other subroutines for assembly, element definition, and so on.
Q: Is there a discontinuous Galerkin function in Coreform?
A: We do leverage some techniques that are closely related to the discontinuous Galerkin method [which converts continuous functions to discrete ones].
Q: Solving time?
A: We use the open source PETSc [portable, extensible toolkit for scientific computation] parallel solver library. We have 15x fewer elements and 20x fewer degrees of freedom for a typical part, so it will solve much faster. However, with the same number of elements, we are slower because of the setup time.
Q: Why is accuracy per degree-of-freedom [DOF] an important quantity to measure/report? As the stiffness matrix gets denser for spline basis, wouldn’t the solve time also increase, despite the smaller number of DOFs?
A: This is highly problem dependent, but as the complexity increases (such as due to nonlinear materials, geometry, contacts, and dynamics) the disparity between Coreform IGA solve times and traditional FEA increases, in that Coreform IGA is much, much faster.
Q: Do you have the ability to report the uncertainty in the calculation of the results?
A: We do not yet support uncertainties in computed quantities.
Q: Does Coreform IGA model [analyze] contacts between parts?
A: We are actively developing contact capabilities. New capabilities will be rolled out with every release. We have a fully-defined (self-) contact algorithm that can be deployed over both body-fitted and immersed contact surfaces.
Q: Can Coreform simulate any kind of partial differential equations set by the user? Not only stress mechanics or Navier-Stokes fluid dynamics, but also flow fields of electrolyzer systems, for example?
A: You can add additional physics through our user-defined interface. Our code is designed to be highly customizable, as this has been a requirement by all our customers.
Q: Is MPI [message passing interface] solver (for parallel computing) supported in Coreform?
A: Yes, it is fully parallelized.
Q: Geometry is obviously very important, but I am sure you are aware that materials’ parameters have been calibrated based upon element technology for traditional solvers, so that certainly applies to IGA. Users will have to look carefully at their material properties. Any comment about this?
A: We have quite a bit of experience porting advanced (often proprietary) material models from commercial FEA codes to Coreform IGA. We have never encountered a situation where special techniques were used in addition to what is commonly used in FEA.
Q: Is there Bezier extraction in IGA?
A: We use Bezier extraction extensively in Coreform IGA. In fact, the U-spline technology is defined in terms of Bezier extraction.
Q: BEM [boundary element method] can have even smaller DOFs for isotropic homogeneous models. However, the resulting stiffness matrix isn’t sparse anymore. How about IGA and sparsity patterns?
A: The sparsity patterns for IGA are exactly the same as p-FEM except there may be more fill-in due to smoothness. This means that the matrices are still sparse, just less sparse than traditional FEA. The matrix is the same size as the number of degrees of freedom in the model. For an IGA solution that has 1000x fewer DOFs it’s okay that this smaller matrix has a higher relative density.
Q: Can we directly use CAD files (such as Solidworks) in Coreform?
A: Coreform Cubit includes data importers that support most CAD inputs. Some of these importers are free with a standard Cubit license, while others are licensed separately. See coreform.com/products/translators for details.
Q: Is there a way to improve the quality of the display of the results, such as smoothing out the results?
A: Yes. The ability to see the results directly on the underlying CAD model is literally days away.
Q: Is the process during decomposition using multiple adjacent subdomains, and using same or different material— Is it a single mesh or multiple meshes?
A: Both scenarios are fully supported.
Q: Can we solve fluid-structure interaction problems (FSI) in Coreform?
A: We don’t currently have fluids/FSI capabilities, but it is on our roadmap.
Q: Heat transfer?
A: That is the next simulation we plan to implement. We never implement a feature without a customer request.
Q: Any plans to support a “Coreform IGA Learn” version?
A: Yes! It will be very similar to Coreform Cubit Learn. Stay tuned.
Q: May I know the immersed technology behind Coreform IGA? Is that adaptive quadrature or reparameterization?
A: We accommodate both workflows. One can be preferred over the other depending on the application. The immersed technology is a generalization of the finite cell method.
Q: What advantages do you see using Coreform IGA over known open-source IGA solvers considering an academic environment? Sadly, we have to take care of possible fees, etc. at academia.
A: Contact us and we can discuss fees.
We’re committed to supporting the open-source community:
-
We added Bezier elements to ParaView/VTK for direct visualization of IGA results. See blog.kitware.com/implementation-of-rational-bezier-cells-into-vtk.
-
We’ve been adding IGA capabilities into the open-source MOOSE solver via its libMesh FEM library. See github.com/libMesh/libmesh/issues/2179.
-
We added Bezier extraction to the open-source Exodus mesh format so that Cubit can communicate IGA’s spline meshes between open-source solvers, such as MOOSE. See github.com/gsjaardema/seacas/issues/189.
We believe there will always be a space for open-source IGA solvers, and that it is mutually beneficial for Coreform to support open-source solvers.
Q: How do you suggest forming an IGA structural model based on point clouds of geometry (based on measurements of parts from something like LIDAR)? What would be the workflow in the context of Cubit?
A: That is of great interest to us. It is not a clean way to build a model. For a point cloud, is a point inside or outside? You still need to build triangulations but it does not need to be clean.
And in Other News
A solution for CAD vendors stuck on Amazon’s cloud-rental hamster wheel: “Analysis of 50 public software companies finds that repatriating workloads from public cloud infrastructure could help them reduce their cloud spend by 50%.” (Sarah Wang and Martin Casado)
- - -
After burning through $2+ billion in funding from SoftBank and other venture capital firms, construction startup Katerra last week told its 8,000 employees that after six years in operation it is shutting down. The company wanted to revolutionize building construction through off-site modular-panel factories and lots of technology. katerra.com/vision
- - -
NCG CAM Solutions releases NCG CAM v18.0 with new features like
-
Create a patch from two 3D curves
-
3D tool guide
-
Horizontal area passes from outside to inside
-
Waterline offset passes (see figure below)
Another new function, Trim Surfaces to a Boundary, will be added later. As the company says, “Perpetual Licensing (This will NOT change).” Download the demo version from www.ncgcam.com/demo-download (after registration).
- - -
Precast Software Engineering is now a subsidiary of Allplan, operating under the name Allplan Software Engineering. www.allplan.com/us_en
Letters to the Editor
Re: A 2021 View of the PLM Market
Here are 171 programs that claim to be PLM at capterra.com/product-lifecycle-management-software: “Find the best Product Lifecycle Management Software for your business. Compare product reviews and features to build your list.”
Can you imagine giving an employee this task from this list?
PLM was never to be this. Those that created it, American Motors, started with a drawing archive. Today, they are trying to manage engineering documentation. The major programs are associative and create a nightmare for document control. The Info-techs have devised many unworkable processes to solve this problem.
Boeing’s engineering was taken over by Dassault’s Catia 5 PLM, the result was the failed MBE [model-based engineering]. They eliminated the drafting group and document control without preparing the staff or doing a test run, all cold turkey.
In the past, all documentation was controlled by drawings (part and assembly drawings) The drawing package defined the complete project and stood alone. PLM’s efforts to take over document control has made a mess of engineering. But it was just part of a perfect storm!
- Joe Brouwer
The editor replies: As PLM is meant to do all things, then all things belong to PLM:
-
P = product = everything that ever was made, is made, and will be made
-
L = lifecyle = the duration of every product, from before it was designed until after it is torn apart by the recycling biz
-
M = Management = managing it all with software and, more recently, using IoT for feedback.
- - -
Part of the problem with the PLM market seems to be that it is compelled to make up new names for the same thing. I’m sure that PLM, PDM, MRP, ERP, and MES have some differences, but all appear to do the same thing.
Or maybe they don’t all do the same thing, but they all — no matter what they started as — now claim to be the total solution to running businesses.
My direct experience of all of them is limited to the ways they impact the engineering world, but it’s always interesting to listen to a CAD company explain that their engineering software is now ready to take care of your accounting system, and vice-versa.
- Jess Davis
The editor replies: That is the ulterior motive behind CAD and other software vendors pushing their PLM: one software to rule your whole business, reaping subscription fees until the day your business ends.
Notable Quotable
“Apple is second to none in this field, perpetuating an obscene throwaway culture, candy-coated with neat videos and communication extolling their green posture. (A note in passing: a lack of further investigation into this is a major journalistic failure.)”
- Frederic Filloux
Thank You, Readers
Thank you to our readers who donate towards the operation of upFront.eZine:
-
Jim Wright: “Thanks for your efforts!”
-
Anthony Mirante
To support upFront.eZine through PayPal.me, I suggest the following amounts:
-
$25 for individuals > paypal.me/upfrontezine/25
-
$150 for small companies > paypal.me/upfrontezine/150
-
$750 for large companies > paypal.me/upfrontezine/750
Should Paypal.me not operate in your country, then please use www.paypal.com and use the account of [email protected].
Or mail a cheque (US$ or CDN$ only, please) to upFront.eZine Publishing, Ltd., 34486 Donlyn Avenue, Abbotsford BC, V2S 4W7, Canada.
*7422
Comments