ExxonMobil: How 3D simulations made with Unity and AI can transform complex industrial turnarounds
Shanti Zachariah - Unity Technologies
Senior content marketing manager

This Unity U/Day Houston presentation is by Jason Judice, principal machinery engineer at ExxonMobil.
Major industrial planned turnarounds are costly plant shutdowns that can cost millions of dollars, according to Jason Judice. Traditional 2D paper procedures often fall short. Teams work under high pressure and extended hours, which frequently leads to communication breakdowns, alignment issues between shifts, and delays during execution.
Jason explains in his talk how ExxonMobil is addressing these challenges with the help of Unity and moving away from static documentation toward dynamic, interactive 3D simulations that provide teams with clear spatial context and procedural clarity before field execution begins.
The visual planning process enables cross-functional teams to practice tasks ahead of time, exposing hidden spatial conflicts and logistical oversights early in the schedule. By discovering critical flaws during the simulation phase, like unworkable equipment paths or inadequate clearance, teams can adjust their plans pre-execution, helping to prevent multi-day downtime delays and potentially saving millions of dollars.
Key takeaways
- ExxonMobil is addressing costly turnarounds with Unity by moving away from static documentation toward dynamic, interactive 3D simulations that provide teams with clear spatial context and procedural clarity before field execution begins.
- Efficiency gains come from building and reviewing the simulation during the planning phase, which surfaces errors before equipment is physically touched.
- While digital twins excel at spatial scans and measuring static structures, 3D simulations are essential for visualizing step-by-step assembly sequences and testing dynamic clearance pathways.
- Visualizing complex equipment removals (such as massive turbines or internal components) reduces execution risks and avoids costly extended shutdowns, potentially saving days of downtime, in Jason Judice’s estimation.
- Key obstacles to scaling include high content generation costs, specialized animation skill gaps, legacy software licensing hurdles, and the need for zero-friction file sharing across global teams.
- Integrating AI agents speeds up content creation workflows, in some cases reducing development time from eight weeks to 27 minutes.

Full transcript
Speaker: Jason Judice, principal machinery engineer, ExxonMobil
Notes:
- This is a lightly edited transcript of a recorded presentation.
Introduction: The communication gap
[0:00] Hi, everybody. I'm Jason Judice. I've been a machinery engineer with ExxonMobil for the past 23 years. The last three years have been devoted to speeding up the way that we do turnarounds and field construction. I'm very much a user of “digital stuff” and not a creator of “digital stuff.”
[0:29] As I was preparing for this, I was thinking about where have we been successful and where have we missed the mark in the turnaround space. Almost every time we've struggled, it comes down to communication in one area or another. Then I was listening to Michael (Micheal Hotaling, executive advisor: Technology scouting, innovation and ventures, ExxonMobil) talk about his experience at home, how he goes to YouTube first (edit: to look up instructions, information). I feel exactly the same way. I go to YouTube and there's one extra thing that he didn't mention, is that those content creators are focusing on what exactly we need to know.
[1:03] Whenever I compare that back to what we see at work, it's very two-dimensional. We learn things, we write it down in text, maybe we have a couple of pictures, but we're building stacks of paper, stacks of things for people to read that people just don't experience at home.
[1:31] The whole thing that we're trying to do is focus on how we can change things by communicating the right context to the right people at the right time. When I say change things, I mean do something differently in the field.
Why turnarounds are hard
How many people in here have done a major turnaround or equipment outage?
[2:05] Okay, maybe it's brand new to a lot of you. This is how it works. We have these huge downtimes. It might be a $50 million, $100 million event. There's so many activities going on. There's so many interfaces with different companies, different people and different disciplines.
[2:29] It's very complicated. Every hour that this plant is shut down is real money being lost. There's pressure on top of the people. On top of that, all of the people are working extended hours compared to what they normally work because we have to do this as fast as possible to get the plant running again. Everybody's a little bit tired. This is the perfect place where you need enhanced communication,
[3:03] but we don't always succeed at that. Then, if you think about all of the planning and all of the effort that goes in ahead of time, the way it works is we have a planner. They're assigned a year ahead of time, maybe three years ahead of time, and they go deeply into the details. They understand with their eyes closed exactly what they're going to do at any moment. They use the digital twin to understand what they're going to do. They walk out the job.
[3:33] They understand it fully and completely. But if you think about everybody else around the table, those people don't understand the full context. They understand their small part. They may not understand how it fits into the whole situation. There are some people that see this long block of text and are reading through and really just space out. They're not able to bring their best selves to the job.
[4:03] The extra difficulty is that these things don't happen very often. Maybe it happens every five years. Maybe it happens every 10 years. People rotate through jobs so often that they don't have personal, hands-on touch experience with the activities. One of the things that we see, sometimes it comes out as conflict where
[4:29] day shift and night shift, get into disagreements about how to do the job. But I think that's really a representation of people that couldn't visualize the job beforehand, were not prepared, they were not bought into the job until they physically saw what they were supposed to do and they had a good idea. Once they had that good idea, that's great. But having a good idea whenever you're in the middle of execution, sometimes is worse than never having that good idea at all because you don't want two teams working together with two separate objectives.
Simulation as a planning tool
[5:04] The idea here is what if we could get everybody on board? What if we could communicate in a way that we unlocked every single person around that table ahead of time as we simulate the job? I'm going to pause here for a second. I want you to think about what I've just communicated and once I roll to the next slide, I want you to think about
[5:34] whether this is what you expected or not. I'm not talking about training. I'm talking about really giving people a chance to practice during the planning of the activity and go through the simulation to see how it should work. This particular task, the simulation on the right,
[6:02] was made ahead of time and it seems like, oh, you're just pulling this piece of equipment out. Maybe it's not that complicated. Well, for this one, you have to decide where are we going to cut this thing off into multiple pieces so that we can get it out through this tight space? Somebody has to decide how much of the piping and other equipment do we have to move to be able to get this thing out. Once you drop it to the ground, you have to have the right rigging that
[6:33] allows it to fall down into the right place. Then once you get it to the bottom, you need to know how to pull it out. This was one of the earlier ones that we did. It was super valuable for us. We thought the value was going to happen once we had the final product, but we were completely wrong. The process of developing the simulation was the thing that unlocked all
[7:01] of the value. The first thing we found as you dropped the equipment down, they initially planned to have a forklift in there in position to pull it out. It was easy to see from the visual as they went through that the forklift that they thought was going to pull it out could not get to the point that it could reach the equipment. We had to redesign this cart and rail system to be able to pull it out
[7:28] to get it out in the open area. I'm just imagining from other turnarounds that I've been through of what you must think whenever this thing came down and you realized we can't even get it off the hook. What are we going to do? I know it would have taken a couple of days of arguing on what to do, and then we would have eventually got it out.
[7:57] I think even this three-day duration savings is very conservative.
Barriers to simulating everything
This all sounds great. Maybe the next question is, why don't we do it for everything? That's a good question, but there are legitimate barriers. Content generation is very costly for us because we don't have anything that's the same as the next piece of equipment.
[8:27] Even if we have the same model number, it's going to have a different environmental context, or it might have different auxiliaries. If we build things that are almost correct, but lead people astray on one part of it, that's probably worse than if we did nothing at all. We could just keep it on paper if we do it that way. We never have CAD for our equipment, but we do have CAD modelers. That's a bonus there.
[8:59] We don't have animators. It's a new skill set that's not really inside of our organization. That's another one that makes it take longer for us to get things done. Then once we get to the turnaround teams, there's always the question of why do I want to spend the money to do this? What am I going to get back for the money that I spend? That part is not always obvious either, because maybe you have a critical path job that, yeah, it makes sense, but you really only capture value if you're able to improve the plan.
[9:34] But you don't know if you can improve the plan until you create the simulation. Then we get all of the other technology questions of why do we need to do this? We didn't need to do it last time. They may see it as extra work because it is extra work for people to sit
[9:56] down and think ahead of the turnaround about how we are going to do this. It is extra work, but you're hoping that it'll pay back at the end.
Digital twin and simulation
The next question might be, what about the digital twin? Shouldn't that be enough? We have visual assets. Yes, partially. I think this is something that can work together with the digital twin.
[10:30] The digital twin is good for a few things. It's good for a broad understanding of the space. It's good for measuring static items. It's really good for overlaying scans and CAD so you can see the models that you're about to build. But there are some things that it's not good at. We talked about the procedure we're using to focus people onto doing the right things.
[11:02] Digital twin is not great at that. It's so broad that it's difficult to pull people's attention to a certain spot. If you look at the two pictures on the right, I took these pictures from a procedure, a tie point procedure, where the planner had his written procedure, 2D procedure, took screenshots of the digital twin and pasted it into that 2D procedure. We took 3D and we moved back to 2D.
[11:33] It's a good workaround. It's a good way to represent something in a focused way, but you lose some resolution there. The other thing that it's not great at is with machinery, we have a lot of small parts that we're working on. We don't have scans of the small parts. We have scans of the equipment outside. You still have to build the ability to see that sequence inside the
[12:01] parts. It's not as good at dynamically moving something through the unit versus sitting there and statically measuring. As we do this, we really need to think about, what do we need? What do we need for each individual activity? On the left side, we need dimensional accuracy.
[12:28] We need to know that this thing is going to move through the unit in the way that we want it to. We're not going to have any clashes along the way. The right side is more about sequence and helping people to understand for quality and for optimizing the sequence. So far, we've typically partnered with third-party vendors to help us with the dimensional accuracy need. Because there's people, teams that are very good with building
[13:02] the models, doing the scans, capturing the scans that you need, and doing that clash detection. It's really critical because you don't want to mess this one up in the middle. We've started developing the sequence focused ones ourselves. This is one of the dimensional accuracy ones. This is about a 100,000 horsepower steam turbine that's being removed and replaced with a larger one.
[13:33] It's about 180,000 pounds with the kit that's being lifted with it. The ability to do this changes the amount of time that it takes. Because if you're not confident you can pull this thing down in one piece, you have to spend the time to take all of the parts individually apart. We're talking, you know, a week of time going out and a week of
[13:59] time going in at least to do this. We're talking about real savings, real game-changing things. As we did the simulation, we identified that we needed to change our bracket because we only have one foot of clearance on the shortest side. It needed to get to the end of that deck and turn it 90 degrees, drop it down, and you can see how it had to weave through the structure. I can't even imagine how you would get this data with a tape measure
[14:33] without having scans and models to compare to.
Our journey and platform criteria
I wanted to walk through a little of our journey to this point. We started mostly by making videos, either with third parties or in-house. Those videos, several of the third parties that we use are making the
[15:03] renders in Blender, and our in-house stuff was mostly made with SolidWorks inside the SolidWorks Animator. We did several proof of concepts for third-party platforms. A lot of those platforms were built on top of Unity. We started to see that as a common theme. As we got a little bit of experience with these things and understood
[15:31] what the different pieces looked like, we started to prioritize what we wanted to see. The first thing was, it has to be an easy, frictionless transfer to the end user. We didn't want to have large files being sent. We didn't want to have software being installed because that is a pain in our IT system.
[15:58] Some of the third-party platforms are license-driven on the user side, that also means managing those licenses just because we deal with people all over the world. It became a chore just to assign and unassign licenses for the end user. I think we don't mind having licenses on the developer side because that group of people is more consistent, but having a license for each user was painful. We wanted to make sure we had a platform that was going to be around.
[16:26] We don't want to do a bunch of work and then realize we have to do it again later. I think that's pretty obvious. Then, it needed to be really simple to take our existing CAD models, the things that we're already making, and import and build the animations very simply. That's our North Star. I would say we're not all the way there yet, but we're making progress towards it. Now we have a CAD team of modelers. They're able to take 2D cross-sections and make representative models.
[17:02] They're really good at that. It's days to a week to make models. We have tools that take the metadata from the SolidWorks models and automatically textures and get CAD files Unity ready. The modelers are working animations in Unity Studio. This is the part where we're still working on how to get more efficient at delivering and standardizing the look around it.
[17:33] Up to now, we've been able to convince people that this is worth it for the really high-end, things that we know are likely going to bring value. But that's like 1 or 2% of what we do. We know that there's other opportunities if it was so fast and so easy to get this done.
[18:03] We have people asking for it all the time. We have people that see the types of animations that we've put together and they want that exactly for their turnaround. The ability to do it easier, faster, cheaper is a way to capture the whole next layer of value.
AI-generated animation
[18:30] Recently, we talked a lot about AI. Maybe this isn't as impressive as some of the things. This was from about eight weeks ago. This is like ancient news in AI terms, I would say. The left side is a trapped bushing seal, compressor seal that we had humans animating and demonstrating their removal and reinstallation of the seal parts. The right side was AI.
[19:04] The human one was done in Unity. It took them about eight weeks to get to this point as they're learning and developing at the same time. The one on the right took 27 minutes to go through the animation and about $10 in credits. This particular one, I think, was right after the Fable 5 release.
[19:33] It was Fable 5 doing the AI. We handed it to the procedure, pictures, and the model. It took the procedures and went through. We had some agents that would go through and decide which steps would be animated.
[19:56] The animation was done in Blender and then uploaded into Unity as GLB. Not pictured here because it takes us a while to get these things approved for external distribution, but we have some recent ones with GPT-6, Astra, and Opus 5.5 that are kind of mind-blowing. Especially compared to where we started. The journey for us is not over, but we're looking forward to the things that come soon.


