Creo MBD Powers GE Appliance’s Digital Transformation

00:00 All right. Well, thanks, everybody. Thank you for joining. We'll go ahead and get started with the talk today. My name is Daniel Campbell. I'm with Nvidia. And I'm joined today by Trent Stone from GE Appliances. And we're going to talk a little bit about how the Creo tools were used to help power GE Appliances transition on their digital transition on their digital transformation. And we'll talk a little bit about the journey that they've had and some of the processes they've implemented, some of the tools they've used.

00:33 Before we start, let me give a little bit of background about myself. So my name is Daniel Campbell. I'm the vice president of the model based definition business with Nvidia. So my company, Nvidia, we are a software tool provider, PTC partner. And what we really focus on is using the model based definition in your CAD model and finding ways to utilize that downstream. And as you'll see in this talk, we we really focus a lot on a lot of the machine consumption use cases

01:06 as opposed to just kind of more human readable. And generally the companies that we work with are the larger OEMs because it's typically the larger OEMs that are the ones that are kind of pioneering the use of model based definition and really starting to sort of design their processes using MD and flow it out to their supply chain. And, and it's been a lot of fun working on this journey of implementing MBD with our customers, especially like,

01:37 like what we've had with GE Appliances. Hello, everyone. I'm Trent Stone. I'm a senior design engineer at GE Appliances. I work in the digital engineering group. We're responsible for the CAD CAD and basically all technical documentation documentation of our product. So I'm going to put in a shameless plug for our brands that we produce. They range from the uncomplicated Hotpoint to the ultra premium monogram. So if you're in the market, take a take a look at us.

02:10 So our headquarters are in Louisville, Kentucky. We're in the appliance park location, which is a 750 acre industrial park. We have five manufacturing facilities there on site. One interesting fact, it does have have its own zip code. So it's a pretty large location, kind of iconic, as Dan Daniel would say, in the appliance industry. More importantly, we have 1600 design and manufacturing engineers there.

02:41 So lots of people to consume this need that we're creating. So Daniel is going to kind of give you an overview of what is NBD to us. So I can just kind of start out like that. So we'll just kind of lay out, I guess, some definitions and some basic ideas about what we're talking about when we talk about NBD and what the point of NBD is. And then we'll get then Trent will get a little bit into the specifics of of the GE, GE Appliances journey there.

03:12 So starting about with model based definition, what is the point of model based definition? I think the main thing is not to think of MD as that. It's the same thing as a drawing, but you can rotate it. I think it's important to understand that model based definition really has a scope that should be much larger than that. And I think one of the best ways to think about how I guess how to think about model based definition is to compare it to or to take a kind of zoom out and think about how we're doing 2D drawing based workflows right

03:43 now. So the way a 2D drawing workflow works is you generate the shape in Creo, then you generate a drawing, you put some projection views of the model onto the drawing, you add annotations onto the drawing, then you print it to a PDF and you send that PDF downstream to quality or to manufacturing. And then somebody looks at that, they interpret it and then they type it into some software somewhere. Like that's the way things are done right now. And if you really take a step back and think about that, it makes no sense because that data initially is being digitized by

04:16 the design engineer when they're entering, when they're creating their drawing. So like where industry is going is that people are getting away from the idea of entering that information initially into the drawing for the purpose of a PDF and instead enter it directly onto your Creo 3D model. That way you can have software downstream, consume those annotations directly so that you don't have an engineer reinterpreting that data, reentering it. The software can consume those 3D annotations directly,

04:49 just like software is doing right now with the shape information. And so that's where things are going. And so in other words, it's important to think about what we call the machine readable nature of this data. You know, how can you make sure that the annotations are entered in such a way that they could be read by software processes? And so Trent's going to talk a little bit about some of the the software tools that they have there at GE Appliances. So I'm going to kind of expand on the machine readable because that's that's really what's important about this.

05:21 We want to be human readable and machine readable. So really the true benefit of it is to automate things downstream. And if you're not machine readable, you're not able to automate. So internal softwares that we use our dimensional variation analysis, team tolerance team, they can consume this. These annotations, the PMI that are in our models, they don't have to recreate it. So time savings there. We have a cost modeling software. We use aPriori. They're able to to also consume

05:51 the data inside our models to help cost it even better. So softwares are a big one, but also your internal processes. So our process, we can auto generate reports from the data through the tool that we get from Nvidia, the capability reports, dimensional reports, all of that. So it's all about automation and downstream to automate these things, that's where you get the true value. And at GE, we had the conversation just in the last meeting with the lilacs

06:23 or whatever lixil about the renewables we use, the renewables created in Windchill and we use Creo view to consume the data. There's complications with that, with, with anything, but we require our suppliers to go get Creo View Express so that they can also consume this data. And we don't have the breakage in the digital thread. So how did we get to this point? We were pretty lucky in that our our now CEO, Kevin Nolan, who was vice president,

06:54 president back in 2015, came back from a GE corporate summit and said we're going to do need. So in three months we had a human readable form of MBD. We kind of just took our 2D drawing, annotated it in 3D and just kind of replicated it. So by quarter one, 2018, we had already teamed up with Nvidia and we started exporting Schiff That kind of turned the page on us being more of

07:27 a machine readable type company, and we were there by Q4 of 2018. So I think I've heard it in lots of lots of these sessions already. Brian Thompson said it in his session MBD is real and we are implementing it, so it's not something pie in the sky anymore. So how did how did we get there? It's good standards, right? We have to have good standards and processes because this is new to new to everybody.

07:58 Engineering, manufacturing, supply base, everybody. It's new to them. So it's crucial to have good standards. So I'm going to give a quick demo of how you have to change some of your thinking and some of your standards. So in the past, we might have put a general note on our drawings and said, you know, all holes and slots are a tolerance of plus 001 -0 zero three. And the human can interpret that. They see the slot, they see the size dimension.

08:29 They know to apply the tolerance that's in the note. So it comes out just perfect, right? The machine looks at it, doesn't see a tolerance, might go apply the Creo parametric tolerance and it comes up with a different outcome. So they're toe to totally different things. So the standards now that we're saying are so important, say we don't do that anymore. We either put the tolerance directly on the dimension or we use the Creo parametric tolerances they provide.

09:01 So now the human sees it, interprets it, comes, gets the outcome machine sees it, they get the same outcome. So again, it's very important because you're not going to create dimensions and annotations like you did in the past. You're going to have to change your processes. So that's why standards are so important. I'll talk a little bit about some of the tools side of things, so I'll show you a little bit about one of the tools that GE Appliances is using to help check the quality of their PMI to try to again enforce that PMI is machine readable

09:36 and to put tools in the hands of the designers to make sure that PMI is machine readable. So inside of our software Nvidia, we have a tool that we call ready check and we kind of try to think of this as kind of like a spell checker in word that basically will point out what we identify as potential issues with the machine readability of your PMI and make suggestions when we can about how maybe it needs to be fixed. And then it's the the designer's job to go in and review that

10:06 and make sure that that that they agree with, with whatever changes are made and go and repair that model. He's being a little modest too. It's more like grammar check. It's a step above spell check in my opinion. So yeah, it really does try to get into like the details of the standard and yeah, that's right. And so to give kind of a little bit of an example of what what we're talking about, we have a little video and I can show you a little video about how it works,

10:36 and it kind of goes through a very basic test case or a couple of basic use cases that you're probably familiar with in terms of Non-semantic PMI or issues that you could run into when you're authoring your PMI . And actually there's a this is on this little video. We put it on YouTube for later if you want to download the slides and view this. So basically our starting point is that we have this, we have our model, it's loaded up in Nvidia and we do an analysis of the PMI

11:09 and in this case we find two annotations that we are highlighting problems with. So they're highlighted in red and we'll go ahead and we'll look at the first one, which is a cylindricity tolerance. Okay, So we'll look at that cylindricity tolerance. The red highlighted section is what it's currently attached to and we are highlighting in green what we think it probably should be attached to. So it's a pretty common error. You have an annotation pointing to a model edge instead of actually referencing

11:39 the surface. So we'll point that out. You can go through right click repair and then we correct that reference. So now it's correctly referencing that surface. Another classic case is we have for X size and position, but it only references one object. It does not reference all four objects in the pattern. And so we identify that it only references that red thing and we're saying, yep, probably should reference those three green holes as well. So you can go ahead and right click say repair and we'll update those references

12:11 so that it references correctly all four holes in that, in that, in that pattern. So there's a couple like very sort of standard, very basic cases. Just to kind of give you an idea of how the tool works and how it looks and everything like that. So yeah. I've got another example. Those are those were both semantic reference type errors, right? We didn't have the correct surfaces or not enough surfaces, but this tool is smart enough to know some some design problems in our models. Maybe. So maybe we didn't constrain the model correctly.

12:43 And three of those holes are 0.225, but the fourth. 0.226. For some reason we didn't make them equal. We didn't constrain them correctly. MTV is smart enough. Ritecheck is smart enough to say you've got four buy on here, but I only see three at 0.225. So it flagged that as an error too. So not only semantic references can be caught , it's kind of design problems too. Yeah, and actually that's an interesting one because it's like that's even almost beyond the scope of just even pure MVP, because if you're doing some type of machining to the to the geometry, I mean,

13:16 you could have Yeah, that could be you're catching manufacturing errors actually type of thing. So yeah, no, there are a lot of very interesting cases that I could detect here and it's usually a big, a big part of, of a successful rollout. And actually, I think Trent's going to talk a little bit about some of the success that they've seen with some of their downstream processes. Yeah, So. As I mentioned earlier, we use this to automate our reports. And what I'm showing here is not ready check errors

13:46 but errors when they got to our helpdesk. This is over the past year. You can see our error count was pretty high once the reports got to our helpdesk. They vary in time to fix too. So some of these fixes might take the team a half an hour. Some of them may be ten minutes, but it's all time and wasted time and money at that point. So as you can see, we implemented ready check in February of 22. It took us a couple of months to get everybody on board with how to use the tool,

14:17 how to understand the errors, how to fix them, because we go back into our Creo and fix it. We're not using the auto fix that Daniel showed earlier, but as you can see, our errors are way down, so that's saving us time and money again. It's it's not just the app we're using downstream or using it in the Diva and places like that so that they can reuse the annotations we're creating. So the lessons that we have learned at GE are obviously robust standards.

14:47 As I mentioned in the beginning, this is new to most people and most users, so you have to have a robust set of standards. And we've created a what we call a best practices playbook or a playbook of sorts to support the ready check tool. So we have the playbook has all the errors that we see through their tool and how to fix those errors. So here's the air, here's how to fix so that robust standards

15:20 and this playbook has really helped us implement this and make it successful. Another thing is, is we engage with people. We went to Cap Vidya. They're experts in the field, tech committees , user groups, you know, form relationships with other companies if you can, that are already doing this. And that helped us a lot. And and I heard this in the in the the last session too, is this is not an engineering only activity. You have to get everybody in

15:52 the product life cycle involved manufacturing supply base. It's crucial to to have everybody on board with this or it won't be successful. Yeah. No, I think that's right. And I'll I'll add a couple of points there to the end. And these points really is, you know, we work with a lot of different users and so we see a lot of the lessons learned from a lot of different types of users and different industries. And I think a couple that I would pull out is one that

16:23 the transition to model based definition is a journey. It's not something that you're going to plan and then implement in a matter of even a year. I mean, it's a journey that takes a few years and you want to try to sort of start small, start with certain pilots, identify certain business areas that you want to start transitioning over to and and do it in a piecemeal way. And the next thing I think is just again, the final point, I guess, is to to think about that machine readable use case.

16:53 You know, don't go to bed just so that you have a drawing that you can rotate. You want to think about what are the downstream use cases for this? Okay, that's great. I'm going to go to bed. Who's going to consume this data know Should I go and talk to quality? Should I go and talk to manufacturing? You know, find stakeholders downstream so that when you generate the data, you know that there's someone who can utilize that and end up showing some some value from that. And with that, yeah, that's, that's what we had today.

17:24 So we'd like to open it up for questions. And anyone who has questions, if we could use the, the microphones. Just is there a set of rules like layers and stuff like that that we have to put. The notes and other stuff that we've put on a regular drawing. Did that come across? No. We use True MVD. They have an annotation option on the model on

17:54 the combo views. And you do these assign to states and remove from states. So they're not we don't have to put them on a layer to to do anything with them. So I think you showed the examples of how it's applicable for the design as an advantage of design. But as you were saying, it's more applicable in the downstream processes. And I think for programing, a CMM might be one of them. But if you could elaborate more on those downstream processes and how you can grab value out of that in machining or whatever it is, that'd be appreciated. Yeah. Yeah. So I think probably the biggest one is generation of

18:27 the reports like what GE Appliances is doing, and that's usually the starting point because that sort of starts to lay the foundation of, I mean, I guess you could think of it as transcribing your PMI into a table format where you're putting your PMI on the part. Why? What are you doing that for? Well, we'll create a table and that establishes some traceability for us to do some planning for every item on that, on that, on that table that you have that we call the bill of characteristics. A lot of that's going to be quality criteria. So like GPA and so it ends up going to quality workflows like generation of reports,

19:02 CMM, automation, things like that. But some of those annotations might be like, for example, a Weld symbol. And so that will go out to a welding process. Maybe it's a set of sort of work instructions for welders, maybe it's the robotics software for generating a Weld program. Maybe it's a quality aspect of the robot of the weld. There are cases for visual inspection and sort of non dimensional inspection. In other words, stuff that's not done with the CMM.

19:34 And I think the other big one is probably like results gathering. So, you know, you've now you've specified your quality criteria on that part with the PMI and now you've manufactured 20 articles, 20 serialized parts, and you've gathered data from every one of those. Now you could store basically a digital twin of each one of those and associate that as built data back to the original model. And that's kind of what we talked a little bit about. Schiff That's that's that was the that's the intent with. Schiff.

20:04 So we're using in our metrology, we have poly works, we do a lot of scanning. So it's valuable in that realm, too. Yeah, It's so there are there are quite a few use cases, but it is usually that that ballooning that really is like the starting point for everything. Yeah. Yeah. Uh, sure. My question is about the file format of gif versus file. I understand you conclude model based annotations in both right now.

20:36 GIF has the the bill of characteristics, but it can't necessarily be used for Cam right now. Is that correct? And what are the next steps to kind of close this gap? So you don't have two different files, one for manufacturing and one for inspection? Yeah. So I would say my biggest so the biggest question I think with model based definition in Cam is

21:08 what cam tools are able to do with the PMI and there's not as much there. So a lot of the cam tools, when they import models with PMI on them, they'll visualize the PMI and that's basically it because a lot of what Cam does is is looking at the shape basically to understand a cutting path. There is some stuff that they could do with PMI in certain cases, but most softwares don't really read a lot of it. So even those tools that read step, it's usually what they're focusing on is,

21:38 is like step 8 to 14 or the flavors of step that don't read or that don't do anything with PMI. So yeah, I mean, so there is yeah. So there is the step API 42, which is the PMI and then the chef which has PMI, we work with both of them and we have customers that use both of them and that's fine. And yeah, I think really the biggest question is just what, what in general can cam do with PMI? Not not even necessarily about data formats. Yeah.

22:09 Yeah. I guess just to follow up at GE, when you currently release a product, are you including I guess the viewable but also a step file for someone who can use non parametric information? Yeah, Fishbowl will actually gave us a solution. It's a it creates a technical data package for us. So we we send out a step and just the PVS from Windchill that's generated we also collect related documents from Windchill and add those into the TDP also.

22:42 Perfect. Thank you. So the timeline you showed earlier, was that a was it three year or was it four year timeline that as you all implemented? MVD Well, it was a really quick, hey, we can do a human readable, you know, in three months we could could at least get to that point. But I mean, it was a pretty quick turnaround in 2018, right? Devin Yeah. So the big question on that is you all have a lot of CAD users there and so do we.

23:15 How do you train? How did you all what process did you use to train those different CAD users? You had to do baby steps, I would. Think lots of videos. We became very good at producing videos for sure. And the other struggle is, is most of our CAD designers are offshore. We use tech so they're not even on site. So videos were crucial to support the standards that we have.

23:45 So that's that was the big one. Thank you. Actually, I got one other thing I think that I've noticed about GE Appliances versus some of our other users is there's a huge focus on GTA like the amount of because you guys have multiple GTA trainers in house, right? We have a dimensional control team that prescribes they're the experts and they prescribe the the GTA for us and then the, the designers then take that and apply it. Yeah. But there just seems to be an expectation there that that everyone will understand Y14 five

24:16 and understand how to really apply it correctly. And I think that makes a big difference actually in terms of absolutely adoption. Yeah. Hi, you guys are speaking. I know this is too you guys are speaking about the downstream impact in the connections, but where I'm at in med Device, we've been thinking more about how to use it with requirements management and risk to do. Think about that. And can you speak a little bit more about the kind of upstream side of this? Yeah, so it definitely fits in very well.

24:47 Like I was saying, the, again, sort of the beginning of a good implementation starts with generating that bill of characteristics, just sort of ballooning the PMI and applying IDs to all the criteria on your part, all the PMI, what you're talking about that can get tied in when you can start to tie in your fmea and your fmea info to that model and to the balloons. And so we do that with a lot of our users where we have workflows, where we'll tie in that fmea information with the balloon model,

25:18 and then we can start to automate some of the stuff like calculation of Criticalities calculation of control. Plans and things like that based on sort of a much more uniform, software based process. That's basically the software manifestation of your business process, rather than leaving it up to a committee of five different people that get together and try to figure it out. I mean, so the end result of running through kind of a requirements management process like that using MD is much,

25:48 much more consistent every time you do it rather than being subjective, you know, depending on who's involved and what type of mood they're in that day and everything like that. So that's actually a huge, huge opportunity, I think, for model based definition. A couple of questions. One was the check tool that looked like it was editing the data natively. What about her back in Creo? What happens with that? Yeah. So that that is the rendering that you're seeing was in Nvidia.

26:19 We have some capability that some of our users use where they can push those changes back into the native model as well. So it could push that back into Creo as well. That would be a question of if some people like doing that, they want the changes to be propagated back, others they want to view it almost more like a 3D report and then they'll manually make the corrections in Creo so both modes would be possible. Yeah. Okay. And then overall for GE, what did you change in like on the Windchill side from a structure management standpoint, are you still managing structure with the gear parts

26:52 as kind of Windchill is implemented and meant to be done with? We are, yes. We I wouldn't call it a true engineering bomb yet. And the CAD does not drive our structure at this point. We're working on a project now to to start that. But yeah, we're managing the parts. We're doing the associations to the from the CAD to the, the GE parts for visualization and things like that. But not to drive the, the bill of materials from, from the CAD side.

27:22 No, not yet. We're heading in that direction. We actually have a project running right now that that that's that's our next step. And then with that in mind as well from a high level. Like top level assembly type of thing. How are you? I'm assuming that's not this is not for every deliverable. There's got to be some, like, high level drawing. I'm still thinking. No, no, it's all MVP. So. Assemblies. Large assemblies? Yep.

27:52 We. We like to think we're only doing those with with dimensional requirements on them. But a lot of times it ends up being work instructions. So that's part of that project I'm telling you about is starting to maybe use work instructions within Windchill to, to take the load off of those large, large assemblies that we're basically, you know, put the tape here, you know, those kinds of instructions, the dimensional ones, though. So, you know, a good example of that is our foam case for our refrigerator. Those obviously have a lot of annotations on them.

28:22 So those are definitely model based and we're doing assemblies and parts both. And then lastly, how are you managing the the exported file? I would imagine something like a step file within Windchill. Yeah. So upon release we use their MVP server tool. So on the back end after we release it automatically creates a chef file and then we store a document in Windchill and then we relate that through. We're also using navigate to so we're.

28:54 What type of relation. It's just a related document. Actually. I don't even think it's a related document yet. I'm sorry. It's just we name it a certain way so that it it works out that way. Yeah. So, so it, it understands it enough in navigate that it goes on the well we have a quality tab so if you go look at the part you know you can go look at the cad, then you go to the quality tab, you can see that chef document that they can download and then start generating reports from it. Okay. Thank you.

29:26 Did you find you had to change your fundamental modeling practices like your part modeling, foundational approaches? Not not a whole lot, no. I mean, it would be nice to have, you know, Creo nine functionality of divide surface and things like that. But even the Nvidia tool is smart enough to understand, even though I've created it as one surface and cut it, it's Creo still sees that as one surface. You could split it if you want,

29:56 but Nvidia is smart enough to to understand those kinds of things. So we've not had to alter our modeling. Just the annotation side, all the PMI that that has really shifted. So how we do the work. The machine readable annotations. Can they be added annotations or do they have to come from the modeling of the geometry? For it to work downstream. I didn't understand that one. So can you.

30:26 Oh, you can use them both. We create all ours. Okay, so. So your modeling practices don't have to be as strict because you don't need all your. I understand where you're going with that now. Yes, We're not. You can use the driven dimensions, as Mark said, but we're creating everything. It's just simpler that way for us. Thank you. And we're using the standalone style, not the annotation feature. So there's even two ways to create the annotations. So that's another decision you have to make. And we decided on the standalone style much easier to manage and and manipulate.

30:57 You don't have to go into the feature be part of it to, to manipulate it. How critical was to the implementation? Would would it have worked without Garr being in play? It would work, but it just doesn't work as well. It's NBD is really, really set up for the all over profiles, the profile of a surface and positioning and things like that. So it allows you to be limited in how you dimension your part.

31:31 Truly, the limited dimension outlook is how I look at it. It just makes things simpler. And we had a push. I mean, our company had a push to start going to Garr and that's why they formed the Dimensional Control team and that made it even easier for us to have that team prescribing things for us. But definitely Garr is I would call a must. For me. I would say part of why it's important actually is so if you if you take a non GAAP approach to dimensioning a part, you're leaving in a lot of ambiguity.

32:03 And so now you take ambiguity and feed it into a software workflow and like it just doesn't make any sense. And so it does sort of like really encourage a less ambiguous approach to controlling the shape of your part, which is what geometric tolerancing is really about.