[wg11] Fwd: RE: Step Standards for Additive Manufacturing

Thu Aug 28 11:03:51 EDT 2014

Hi Ed,

Thanks for the feedback below. I think it confirms there is an issue and 
opportunity.

We can both read P21 files. When the spec is finished we will both be 
able to read P21 e3 files.

However, if we do not agree on a common organization for the 
partitioning then the only way I will be able to read your e3 files will 
be to read all the partitions, assembly all the data in memory and 
create my internal data structures, and vice versa for your software. 
This is not going to work for incredibly large models.

If we want to be able process each others very large models then we need 
to agree on one or more organizations for the e3 files.

A simple example is a matrix. In the header of each P21 file we put a 
volume record describing the geometric limits of this model as a 
rectangular volume, then when I need to process your data to create a 
layer for some kind of machine processing I will be able to limit myself 
to those partitions that may contain material in this layer and vice versa.

Obviously this only solves one of the problems because what happens when 
you want to visualize the whole model before your print it? (The AVM 
program has a very hard time when it wanted to exchange the complete 
model of its vehicle. It was only 700M but some of the low end CAD 
systems took forever to read and convert the data).

Hence we need some kind of big manufacturing data project. The goal 
being to come up with one or more basic data organizations so that we 
can all cooperate to:

1. Build very big models
2. Visualize and check them
3. Machine them

Martin

On 8/28/2014 8:25 AM, Paff, Ed wrote:
>
> */Martin, I sent this to some of our guys who have been working with 
> AM in our cambridge office with our cadfix product.  Here are some 
> comments:/*
>
> *//*
>
> I'm not entirely clear on the context but only considering the AM 
> example I'd say that billions of triangles would be hard to justify 
> even if the software didn't buckle. The expectation is that some 
> designer will be sending his models off to a bureau via the internet 
> to manufacture -- Like Henry does with his snowflakes. If the file 
> size for a simple but highly curved object grows too quickly it 
> becomes a punishment to send them to the bureau. Henry's use case is 
> actually about the worst possible scenario because each of the 
> snowflakes he sends to Materialise is unique.
>
> Neither STL or AMF (fancy XML replacement for STL) really deal with 
> the issue of rapidly rising facet counts as the precision of additive 
> manufacturing machines increases. AMF at least uses a list of vertices 
> indexed by the facets but it's written in very idiomatic XML so it's 
> super-verbose in it's raw form has to be zipped and decompressed on 
> the fly while reading which can't do much for performance.
>
> For our current work in AM I would tend to favour plain old BREP 
> geometry with some kind of strict limit on the edge-face sloppiness 
> which is the cause of most of our woes. STEP is perfectly suitable for 
> this but things start to look a little tricky when we have to deal 
> with "organic <http://www.thingiverse.com/thing:11509>" looking models 
> that can't easily be converted into a BREP. Support for freeform 
> models would probably require a lot of planar triangles or a neutral 
> subdivision surface or high-order triangle representation like we use 
> in CADfix. However, I don't know if there's much urgency in industrial 
> AM applications to handle non-BREP data.
>
> Metadata is also fiddly and not really handled satisfactorily in AMF. 
> Multi-material printers are increasingly popular these days but 
> assigning a specific material or colour transition within a body just 
> isn't done. Usually it's supported by cutting the body up with planes 
> and assigning a single property each of the resulting bodies. 
> Similarly, decisions about layer positioning and models orientation 
> have a significant impact on the manufacturing results for precision 
> parts and there's a whole debate on whether this should be specified 
> as part of the model description or left up to the machine operator to 
> decide.
>
> So I guess I'm saying there's a lot of room for improvement over 
> current processes. A multi-file approach to massive models (or very 
> large build volumes in AM) sounds very sensible as a means of keeping 
> models manageable and any other moves to limit file size without 
> losing quality of data would probably be appreciated. For reference 
> CADfix can just about display 40 million triangles without dynamic 
> rotation falling apart but I imagine that 100x that number would start 
> causing all sorts of problems in the kernel -- I note, for example, 
> that the maximum 32 bit integer is about 2 billion so we certainly 
> can't have a mesh/frog with more than that many vertices without 
> switching a ton of code over to long integers.
>
> *//*
>
> *//*
>
> *From:*wg11 [mailto:wg11-bounces at steptools.com] *On Behalf Of *Martin 
> Hardwick
> *Sent:* Tuesday, August 26, 2014 11:55 AM
> *To:* wg11
> *Subject:* [wg11] Fwd: RE: Step Standards for Additive Manufacturing
>
>
> All,
>
> Here is the paper that raised questions about the ability of 
> traditional CAD systems to model very large structures for 
> applications such as additive manufacturing.
>
> This does not mean that they cannot be used for such modeling but it 
> does imply that new infrastructure is required to enable such modeling 
> and in my opinion the best hope for a near term solution is Part 21 
> Edition 3 combined with AP242, and the STEP-NC models for additive and 
> subtractive machining.
>
> Key advantages include:
> 1. Interfaces to read and write the geometry to and from nearly all 
> the CAD systems
> 2. Very efficient format for modeling facets
> 3. Information models both additive and subtractive manufacturing
> 4. International Standard
>
> What we need to do is create an infrastructure for processing this 
> data in very large volumes so that we can support applications like 
> additive manufacturing, 3D circuit design, very large buildings 
> construction and very large defense platforms.
>
> I would see this mostly as imposing an organization onto the division 
> of an AP242 file into billions of linked P21 e3 segments. Seemingly 
> trivial until you start to think about all the different ways that 
> applications will want to find and access the data.
>
> Here is the problem statement that I received from a well known 
> CAD/CAM vendor
>
> (1) Simple calculations. Suppose you have an object with a volume of 1 
> cubic meter, and the
>
> cells in the lattice have a volume of 1 cubic millimeter. Then there 
> will be 10^9 cells.
>
> If you use a curvy b-rep, there might only be 2 or 3 faces per cell; 
> with a facetted b-rep,
>
> maybe 10 or 20 facets per cell. But, either way, that's a huge number 
> of faces. If the cells are
>
> all identical, you can represent them using a repeated pattern 
> technique some of the time,
>
> but, for some operations, you will actually need explicit 
> representations of all those faces.
>
> No modeler (as far as I know) can handle that many faces, and no 
> graphics card can handle
>
> 10^9 triangles. Not even close.
>
> (2) These folks ... (in the attached paper)
>
> found that ACIS ground to a halt when they tried to model a lattice with
>
> around 2400 simple cells (see pages 9-10). This was 10 years ago, and 
> computers are
>
> bigger/faster nowadays, but I don't see that this will allow us to 
> make the leap from
>
> 2400 cells to 10^9. I'd love to be proven wrong. If plain old b-reps 
> could be made to
>
> work, that would certainly make life easier for us.
>
> Martin Hardwick
> President STEP Tools, Inc.
> Professor of Computer Science, RPI
> Team Leader ISO STEP-Manufacturing
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.steptools.com/pipermail/wg11/attachments/20140828/eab02a6c/attachment.html>