New study: the best way to get from point clouds to parametric solid models
NORMAN, Okla. – Last summer, Dr. Kuang-Hua Chang, Williams Companies Foundation Presidential Professor at the School of Aerospace and Mechanical Engineering at the University of Oklahoma, was given a task by Tinker Air Force Base, outside of Oklahoma City: figure out the best way to get a parametric solid model from a point cloud created by a laser scanner.
Tinker is one of the major repair and overhaul locations for the Air Force and has used Dr. Chang in the past as a consultant on engineering projects. Recently, reverse engineering has become a major theme. As planes like the F-14 now enter their third and fourth decades of operation, parts break, and there often are no modern CAD models for those parts. How do you make a copy in a hurry?
Enter Dr. Chang.
“They wanted to know what’s in the market and what can they do to utilize the current technology,” Chang said in an interview with SPAR 3D. “They already have several very nice scanners. They have Leica handheld, with a probe, and then an ATOS 3 – both are top of the line in my opinion.”
“But they didn’t have the software for when they get those points to put them together and come up with something that they can use in a short time frame,” he said. “So they wanted us to find out what their options were.”
That question, and the study that ensued, goes to the heart of questions that manufacturing firms and nascent reverse engineering firms are asking themselves every day. The answers they’re coming to ought to be informed by Chang’s conclusions, but SPAR investigation suggests the conclusion of Chang’s study and the conclusions reached by individual companies might be different, depending on the goals of the organization.
Charged by Tinker, Chang, along with graduate student Chienchih Che at U of O conducted a three-month investigation, looking at software packages in two main categories:
1. Mainstream solid modeling CAD systems, such as Dassault’s SolidWorks or CATIA or Siemens’ NX or PTC’s Creo, which have some point-cloud processing capabilities.
2. Purpose built reverse-engineering software developed to get from point cloud data to CAD geometry.
Chang quickly dismissed the first option, learning that the solid modeling solutions from the established CAD systems they looked at can’t handle multi-million-point clouds. The team then spent the rest of the summer investigating the software packages that constituted the second option.
This August, Chang published his findings in Computer-Aided Design and Applications in a paper titled, “3D Shape Engineering and DesignParameterization.”
The first ought to be good news for the industry: “After research and development in decades, technology that supports 3D shape engineering and design parameterization is matured enough to support general engineering applications.” This is independent, non-manufacturer-funded validation that scan-based reverse engineering ought to be mainstreamed and is no longer new or cutting-edge technology.
The second conclusion is more controversial: Chang’s team concluded, “Rapidform is the only viable choice for parametric solid modeling in support of 3D shape engineering and design parameterization.”
In our interviews, Chang provided assurances that Rapidform had no knowledge he was conducting these tests. And Tom Charron, VP of global marketing at Rapidform, told the same story: “We didn’t even know it was happening at all until it showed up in this journal. One of our developers saw it in the journal that he reads and obviously that got around real quick inside the company.”
How Chang and his organization came to this conclusion, and what it says about the way software in the 3D data capture marketplace is developing, points to a bigger question for software developers: Should new products be excellent in one piece of the overall workflow, then hand off to, or plug into, larger, more robust packages? Or should new software duplicate some of the capabilities of the more established packages in order make the workflow more fluid?
Simply put, model in CAD or model in the point cloud software – which is better?
On the two sides of this rift, and the two packages that Chang decided were worth serious consideration, are Rapidform and Geomagic.
Just about everyone involved in this study or who spoke for attribution for this article acknowledges that, since software development in the 3D data capture field is so rapid at this point, the conclusions have a short life. Geomagic has already addressed many of the shortcomings that Chang cites in their v. 12 release, noted Karl Matthews, VP of product management at Geomagic. Innovmetric hadn’t tackled parametric solid modeling at the time of the study with its Polyworks software, but now does in its newest release, said Innovmetric head Mark Soucy.
Looking at the bigger picture, though, “When I read the study, I can see at the front end of it there’s a lot of good review of the state of where things are,” Matthews said. “I think it takes an interesting change in the middle, from a sort of review of technology, which gets pretty detailed and techie, to a review of certain packages, largely Rapidform and Geomagic.
“Obviously he concluded in favor of Rapidform, and the thing that was important to him was the inclusion in Rapidform of a basic CAD engine, which does solid modeling inside of it,” he continued.
“Our approach is to do what we do very well: accurately represent the scan data, and then move the modeling task to the users’ CAD program. And the bridge is the ‘parametric exchange’ feature, which works in conjunction with how we identify and model the various entities inside a data sample, and give the user flexibility in transferring that data into their CAD program to reflect their design intent.”
“And that’s a key philosophic difference between the two of us.”
Rapidform agrees that the philosophic difference is stark.
“And we totally disagree with that approach,” Charron said. “We feel it needs to be integrated into one software.” Rapidform “combines the CAD system with all the scan-processing tools, all in one application. No one else does that.”
When time is your motivating factor, Rapidform argues (and Chang agreed in his study), it’s better to extract and develop parametic modeling features in the program that works best with your point clouds, then move these features to the robust CAD program.
In sharp contrast, Geomagic believes you want a robust CAD program for editing and managing features. “You change the bore diameter, and that changes the specs for more than one part,” Matthews said. “I think that needs to be managed in a full CAD environment. That speaks to why we did what we did – get the data into CAD so they can work as quickly as possibly in their own CAD environment.”
For his part, Soucy and Polyworks agree with Geomagic’s philosophy, and he feels strongly that CAD pros want to get the information into their robust CAD program as quickly as possible. Not only that, he feels Dr. Chang didn’t go far enough in his study. It’s not enough, he said, to simply evaluate the process of getting the parametric solid into the CAD program.
The important follow-up question is: “What’s the status of the model once it’s in CAD?”
As Dr. Chang cites the time savings of being able to do some of the CAD work in Rapidform, Soucy argued, Chang should also talk about what happens to that CAD work once it gets into something like Solidworks: “It’s a dumb model!” By that he means that all of the associativity and references between pieces of that model, what Soucy called the crucial intricacies of CAD, don’t carry through.
Chang addresses this in the paper by writing, “The dimensions and constraints added to the sketches and solid features in Rapidform are exported well, except again for those referenced to entities that are not available in SolidWorks.”
“Show that to any CAD expert, who really knows,” Soucy said. “Ask him if he’d make this in this way. You’d never create a shape this way. Never, never, never … You split your sketches into multiple sketches.”
Chang’s study, “I think, is a narrow study of parametric modeling in the software,” he concluded. “It doesn’t give an idea of the big picture.”
WHAT THE CAD GUYS SAY
So, we at SPAR talked with some CAD guys who use all three products: Rapidform, Geomagic, and Polyworks. Michael Raphael, Pete Kennedy, and Greg Chaprnka, the owner, operations manager, and lead engineer for Direct Dimensions, a firm that both provides scan-to-model services and resells modeling software (including all three of Rapidform, Geomagic, and Polyworks) agreed to discuss with this reporter Dr. Chang’s study and their own CAD workflow preferences.
When I posed Soucy’s quote above, Kennedy agreed that’s how some CAD users, especially hyper users of specific packages, would see the situation. “He’s right, you’d have to re-declare associations” after you’d brought the model in from Rapidform, and that might eat up much of the time savings that Dr. Chang found in his workflow.
Further, said Raphael, “if these same guys are CAD experts, they maybe just don’t want to learn a new way. CAD people can be really stubborn that way. They want full control in their own environments, so Geomagic and Polyworks are taking advantage of that.”
He also noted that the more versed you are with a CAD program like SolidWorks, the more you’re going to notice the limitations that come with importing a parametric solid from Rapidform.
“Rapidform is a very good workflow,” Raphael said, “and it can be the fastest, depending on the nature of the project, and the nature of the people performing the project. It’s not going to work for all cases and for all people and for all CAD platforms.
“As much as it might work for other people, some find the Rapidform constraints a problem.”
“The basic reasoning behind Geomagic’s philosophy is that they believe CAD guys are experts in their CAD system and they don’t want to learn a new one,” Raphael continued. “If they’re very good in their own systems, they want to use that one. That’s an argument that holds water. It’s a viable approach to doing things.”
In order to tackle all the new reverse-engineering projects that are coming down the pipeline, and to get creative with how to solve increasingly complex scan-to-model problems, the Direct Dimensions crew said it’s unlikely that a large engineering firm, or manufacturing outfit with a large internal engineering department, will be able to continue to be happy with just a single CAD platform and a single way of doing things.
“I think the [Chang] study serves one important purpose,” echoed Geomagic’s Mathews, “which is to increase awareness that this [point cloud to parametric solid model]technology exists, and I think everything that helps potential users understand their options, and understand what the state of technology is today, that helps the market.”
Indeed, while Chang stands by his findings re: one software package vs. another, he’s also keen on people knowing that he found scan-to-model reverse engineering is “now low cost and a lot of people can afford it and start doing reverse-engineering work. It’s gaining momentum because the technology is real, in our opinion … I still recall when we started [10 years ago], the best equipment that was available at that point was the 3D camera, and there was almost no software that offered viable capabilities to take point clouds and generate any useful form in a short turnaround time.”
Ultimately, said Geomagic’s Matthews, “you have options. You should find out what works for you. That’s the key thing. There are different workflows; there are different tools. Some tools are going to suit some users best. That’s a natural state of the market. What’s really important is that users should get involved and try these things out for themselves. Get the demos; get the trial software; look at vendor websites and all the videos, and try to figure out what is important to their respective workflows and then evaluate the tools against their needs.
“I don’t think any generic study can make user decisions for them.”
WHAT THIS SAYS ABOUT THE FUTURE AND THE SCAN-TO-BIM MARKET
It shouldn’t be surprising that this same philosophical difference is beginning to manifest itself in medium- and long-range scanning as well, where the workflow, some people believe, is beginning to come more in line with the short-range, reverse-engineering workflow.
In the scan-to-BIM world, “what it really is underneath is a parametric solid modeling program,” Charron said. “It’s a very similar concept at the core.”
Matthews agreed, saying that the long-range and short-range workflows have developed separately for a couple of reasons – most boiling down to the sheer number of points (maybe tens of billions of points) needed for a complete scan of a large building vs. a small part of an airplane (usually not topping a billion) but that advances in processing power and storage have begun to make those workflows more fundamentally similar.
“Traditionally, there have been different ways of handling the amount of data you get with a reverse engineering scan vs. an AEC scan,” Matthews said. “The amount of data has just caused software to be written differently for those two different domains. It will be interesting, as algorithms advance, how they merge together.”
Direct Dimensions’ Raphael is surprised the two workflows haven’t come together quicker. “These three big boys that have been so deep in point cloud processing have ignored [the AEC]space, and I don’t understand it,” he said. “I get that they’re not architects, but neither am I and we’re scanning and modeling buildings on a regular basis, and it’s not that different [from reverse engineering]. It’s not as complex as surfacing, but it’s still millions and millions of points.
“I don’t understand why these three players in the mechanical space haven’t added more building-oriented tools to their platforms,” Raphael continued. “I think the building space is even bigger than the mechanical space.”
Innovmetric’s Soucy didn’t completely discount Raphael’s reasoning, but noted some major differences. “I’ve been in the BIM market since 2001,” he said. “I see some parallels, and in a way the close-range market is ahead of the long-range, because of the long experience using scan data. But in reverse engineering the key difference is that when we do close range, we only do one part. Typically it’s one fender, one this, one that.
“While if you do BIM or a long-range application, it’s extremely complex … So it’s almost impossible to have a reverse engineering system … These models are so complex, and only programs like AutoCAD or Microstation can handle that. The people who are successful sell plug-ins, like kubit or Pointools or ClearEdge.”
And that’s where the similarities start (and maybe end). “They help extract the models and the CAD is done by Microstation.” Likewise, with Polyworks, Soucy said, “Our role is quickly extract the data and get to CAD, and what Geomagic does is very similar.”
However, “if you look at the BIM world,” Soucy said, “a lot of the shapes are square-ish. They can be described by libraries of parts that are already labeled. You can reverse engineer a BIM environment using a library of parts that you scale down.”
Which is where that different architecture thing rears its head again. Rather than create a parametric solid, or even a surface model, in programs like Rapidform, Geomagic, and Polyworks, an attack like ClearEdge’s, where many of the simple shapes are quickly identified and extracted, may make much more sense for the AEC market.
One parallel that may make sense between the long-range and short-range markets, though, is in how the big CAD softwares will work with the point cloud processors and plug-ins. As large-scale engineering software firm Aveva last week bought long-range point-cloud processing suite LFM, many people SPAR spoke with for this article believe that the mechanical CAD software packages are likely in the near future to either buy point cloud processors like Geomagic, Polyworks, and Rapidform, or simply develop the technology themselves.
“I do think that will eventually happen,” said Direct Dimensions’ Raphael. “On the mechanical side, there’s a lot of tools there, and as the success stories come out, it piques the interest of the CAD guys. And it’s starting to happen and they’re getting involved. I do think there’s an acquisition that could happen.”
Why don’t these big firms already have this point cloud processing capability built in? Well, they were created largely before scanning became part of the process, reasoned Direct Dimensions’ Chaprnka. “They just don’t expect people to start from a point cloud. It’s a computer drawing instead of a paper drawing, as opposed to starting with a scan of something. I wouldn’t be surprised if they start adding it later.”
“Their engines are built differently,” Raphael agreed. “They’ve always thought reverse engineering was a very slim need in the world. And none of them really adopted it.”
As the demand for point cloud data to be used in engineering grows, both at the small and large scale, grows, it’s likely the large engineering CAD packages will continue to either work with smaller software companies so that the capabilities are better utilized in the CAD packages (much like Pointools works with Bentley, for example, or kubit with AutoCAD) or they will purchase the software company outright and hardcode it into the basic CAD offering.
As the market develops, it’s likely this philosophical divide about when to get the data into the CAD package, and how to process the point cloud into different types of models, will continue to shift. Considering the speed of the market’s current development pace, it’s unlikely any study, such as Dr. Chang’s, will definitively settle that philosophical divide, but studies like his have, and will, certainly help to shape the discussion and move the market forward.