Tata Daewoo: validating vehicle noise, vibration and harshness in virtual reality

In this recorded presentation, Jinyoung Park, NVH (noise, vibration, harshness) engineer at Tata Daewoo Mobility, demonstrates how his team validates vehicle NVH in a virtual environment created with Unity Industry, instead of a physical prototype. The virtual NVH validation is a simulation tool for monitoring and measuring noise, vibration and harshness. In the presentation, Park explains the pressures reshaping automotive development, the limits of simulation-only prediction, and how he combined measured test data with Unity Industry to build a standalone headset tool that lets anyone hear and inspect a truck's noise sources at full scale.

What you will learn

• Why high-mix, low-volume manufacturers cannot prototype their way to NVH sign-off
• How real test data turns simulation into a validated virtual environment
• What Asset Transformer and Asset Manager solved for a non-developer
• How the tool helps decision-makers judge an NVH investment before they commit
• How Tata Daewoo achieved an NVH validation accuracy level of ~ 95%.
• How Tata Daewoo used Asset Transformer to optimize CAD design data from 29 million triangles to ~2.7 million triangles

This presentation was recorded at the Unite Seoul conference in July 2026.

Video transcript

Speaker: Jinyoung Park, Tata Daewoo Mobility

Runtime: approximately 33 minutes

Editor's note: This is a lightly edited transcript of a recorded presentation.

Introduction

[00:05] Hello, everyone. Today I'd like to share my vision through a case of developing automotive NVH packages using virtual environments and real test data. My name is Jinyoung Park.

[00:37] I began my NVH career in Italy and worked extensively on vehicle NVH development for a number of global OEMs. I'm currently working at Tata Daewoo Mobility, a commercial vehicle company.

Today I'll first talk about the challenges the automotive industry faces, and then about the trends we're seeing now. Then I'll talk about how we applied digital innovation to automotive NVH. That's what I'd like to walk you through. NVH here stands for Noise, Vibration, Harshness, which is basically what we commonly refer to as noise and vibration. I'd also like to explain why we used Unity Industry.

Who Tata Daewoo is, and why truck variants matter

[01:06] Before we begin, let me briefly introduce Tata Daewoo. The vehicle models you see here are Tata Daewoo's core products. We sell everything from heavy-duty trucks to medium-duty and light-duty trucks.

[01:40] However, there is one thing you should keep in mind here. Unlike regular passenger cars, trucks have quite a lot of variants. What you see here shows only three classes: heavy, medium, and light duty. But even within heavy-duty trucks, there are many variants, such as mixers, cargo trucks, wing bodies, military trucks, and recovery trucks. And depending on that, different engines may be installed, or the number of tires may vary, so there are many complex variants like that.

The pressures reshaping automotive development

[02:11] So first, I'll talk about the current challenges facing the automotive industry and then the trends. As many of you know, starting about 10 years ago, as electrification and carbon emission regulations tightened, the industry has been facing a number of issues.

[02:38] In particular, as you know, vehicle development in China is moving incredibly fast, and traditional OEMs simply cannot keep up with that pace. On top of that, carbon regulations are also coming into play, so they can't even build the internal combustion engines they used to excel at, and instead have to chase the electrification trend. So there is real pressure to minimize the cost, time, and resources that development consumes. At the same time, development has to move faster, while also keeping sustainable management in mind.

[03:05] In this situation, even the traditional industry leaders, the German OEMs, are struggling to keep up with the electrification trend. So for a company like us, Tata Daewoo, which does high-mix, low-volume production, we had a lot of concerns about how we could keep up with those development trends.

From physical mockups to digital twins

[03:34] Many OEMs have found their answers in the technologies that are advancing today. In the past, development was often done like this, with mockups or real test data. Up until now, as shown in the middle, vehicle development has largely been carried out through computer-based simulation. And some leading OEMs are already using digital twin models in virtual environments to validate vehicle development, as the industry moves toward what we call software-defined vehicles.

[04:08] Now, moving on, how do we apply this digital innovation to NVH? Since I'm an NVH engineer, I thought about how I could use these technologies in my own field.

[04:39] I took the test data we already had plenty of and, using a virtual environment like this, set myself the goal of building a development paradigm that goes beyond prototyping.

What NVH actually means

Before getting into that, let me first explain what NVH, that is, noise and vibration, actually means. First, there is environmental noise generated by automobiles. This includes the noise made as vehicles pass by, or the discomfort you feel as a passenger inside the car, the kind of discomfort that noise creates.

[05:11] As you can see, noise generated by automobiles is always present in our daily lives, and it's a problem that NVH engineers like me need to improve and solve.

Structure-borne and airborne noise

So let me talk about the main noise sources in a vehicle. These can be broadly divided into two categories.

[05:42] There is structure-borne noise, and there is airborne noise. Structure-borne noise comes from the engine, or from the road surface while driving, and enters the cabin as structural vibration. Airborne noise, on the other hand, includes the wind noise generated by the airflow as we drive, or noise generated by the tire pattern that comes in through the wheel housing.

NVH as brand identity

[06:13] But when we talk about NVH, the goal is not simply to make a car quiet. NVH also reflects a brand's identity and the identity of the vehicle itself. Here, you can see photos of two cars. They are exactly the same car, a Porsche GT2.

[06:42] One has the full package fitted, while the one on the right has the bare steel exposed. Looking at them, I'd like to ask what impression you get right away. The one on the left seems quieter and more expensive. Doesn't it just look more comfortable overall, too? On the other hand, the one on the right somehow looks uncomfortable and like it would be noisy.

[07:11] In this way, the NVH package is also an important factor that determines a vehicle's identity. It's not very apparent in this example, but if these two cars were different brands from different companies, the one on the left would feel like a more premium brand, while the one on the right would feel like a slightly cheaper brand. Of course, this GT2 RS is such an expensive car that it's not a perfect example, but in any case, NVH also expresses a brand's identity in that way.

What sits inside an NVH package

[07:43] So NVH engineers need to think carefully about the brand philosophy and the direction the vehicle is aiming for. Now, about the NVH package itself, you may be wondering what exactly is included in an NVH package, so I've shown it here in a simple diagram. On the left, I'll explain it based on commercial vehicles, and on the right, I'll talk about passenger cars.

[08:12] With commercial vehicles, obviously, rather than the noise comfort of actual passengers, they're seen more as cargo carriers, basically as equipment that makes money, so compared with passenger cars, the NVH package itself isn't as extensive. Still, if we divide them broadly, there are interior trims and exterior trims. For interior trims, they filter out the noise coming from outside the vehicle into the cabin.

[08:42] And they're also used to make noise generated inside the vehicle easier on the ear. And they serve an aesthetic function as well. For the exterior, there are parts like undercovers or engine bay side covers. These are parts that address noise radiated outward from the vehicle engine or other noise sources, especially with the goal I mentioned earlier, which is reducing environmental noise.

Why NVH matters more in electric passenger cars

[09:13] On the other hand, to explain passenger cars a bit more, maybe it wasn't the case in the past, but as vehicles have become electrified, they've become much quieter compared with internal combustion vehicles. So NVH performance in passenger cars has become very important. Especially these days, if you drive or ride in a Tesla a lot, you'll notice that the infotainment system inside is now bigger than any other component in the car. So while the inside of a vehicle used to be simply a means of transportation, now the concept has expanded into a space where people consume content.

[09:45] Because of that, the importance of interior trim has grown considerably. And especially as regulations like carbon neutrality become stricter, we also have to watch fuel efficiency closely. For fuel efficiency, what matters in passenger cars is residual heat management.

[10:15] Especially with diesel engines, or today's internal combustion engines like gasoline engines, when the vehicle is started from a completely cold state, it emits much more carbon. So the question is how to retain that residual heat a little longer, so that when you start the car the next morning for work, carbon emissions are reduced as much as possible. Because of that, inside the vehicle, engine encapsulation functions that wrap around the engine are widely used.

[10:42] As for exterior trim, for commercial vehicles, as I said earlier, they're used simply to treat noise radiated to the outside. But in passenger cars, they're also used for aerodynamic performance, to improve fuel efficiency and reduce aerodynamic drag, so they are applied quite widely.

Two approaches: simulation and physical testing

[11:12] How do we go about NVH development? Generally, there are two main approaches. One is to run simulations and make predictions through computer engineering like this. In general, once the design of the NVH package I showed earlier is decided, the material composition and so on are all determined along with it. Using the material properties, we get sound absorption data, and there's a sound insulation graph shown here. These data are used as the input, and then, in a full vehicle model like this, through FRFs, or frequency response functions,

[11:46] and natural frequency analysis, we feed that data in and predict the interior noise inside the actual vehicle. But the most important point here is that in simulations, reliability varies dramatically depending on the quality of the input data. It would be ideal if accurately measured data were entered, but usually, when we run simulations, the data we need are simulation data obtained through the design process,

[12:20] so depending on how accurate those simulation values are, the results may be accurate, or they may be inaccurate. On the other hand, there's also a way to acquire data through actual vehicle testing. That's another development method. Once a prototype vehicle or something like that is developed, we can use a dyno bench or real road driving, or drive the vehicle on a proving ground and measure dynamic loads, or measure the structure's response functions to real inputs like this,

[12:59] or obtain data through natural frequency analysis. Then, through transfer path analysis using that data, we can assess the vehicle's noise contribution and identify through which structural paths the noise is being transmitted more. These are the kinds of test methods we have. As I mentioned earlier, actual vehicle test data are somewhat more accurate, but there is a gap between them and simulation.

The variant problem, and the cost of prototypes

[13:27] With simulation, since vehicle NVH performance is predicted entirely through simulation, there can be a fairly large gap from actual vehicle test data, and to overcome that, we need to conduct validation tests for the various variants I mentioned earlier. So at vehicle development companies like ours, for nearly all variants, or at least for the representative ones, we produce prototype vehicles,

[13:58] run actual vehicle tests, and feed the verified data back into the analysis. That is how we validate. But when the number of variants increases like this, the problem that arises is complexity. And now we're not just dealing with internal combustion engines. There are hybrids, and we also have to build electric powertrains. And for passenger cars, there are hatchbacks, sedans, and wagon types as well. These days, there are SUVs, compact sedans, and so on.

[14:28] When there are too many variants, things become less predictable and therefore less accurate. Along with that, the effort required for validation also increases, and there's the question of how far we can go in producing all those prototypes. As I mentioned at the beginning, for a company like Tata Daewoo, which has a very large number of variants, we simply can't build prototypes for every single vehicle.

[14:58] In general, we only produce prototypes for the best-selling models, and evaluate those. But for the other models, we have to verify them through non-test methods, such as simulation. And in that process, when we change something, there is some uncertainty about how the NVH performance will change. And of course, there are costs, time, and waste that occur from that as well.

Hybrid simulation: validating NVH in a virtual environment

[15:31] So in a situation where the development paradigm has changed, we've given a great deal of thought to verifying NVH performance while also reducing time and cost. The idea is to create a virtual validation environment based on test data. And I'd like to call that hybrid simulation. Generally, this approach uses the data obtained through the actual tests I mentioned earlier as input data to implement a virtual-environment validation model.

[16:09] And further, by introducing virtual environment technology, the concept is to evaluate the vehicle in the same environment in which the end user evaluates the car. The idea is to evaluate it right from the design and development stage, in an environment just like the real vehicle.

Swapping components in the virtual environment

This is about changing vehicle components in the virtual environment. I've shown the V-model here as an example.

[16:38] For example, if we swap out the suspension or the engine, we bring in input data, the test data we've accumulated so far, place it in the virtual environment, and then run the test again through that. The data tested in the virtual environment can be accumulated again as test data through the equipment typically used by NVH engineers for evaluation.

[17:09] And the data accumulated through those tests can later be used for optimization through AI algorithms, and also as training data for purposes like that. That is the function we want to build. From there, we can test various scenarios, then identify the solution with optimal NVH performance. Whether it's one, a hundred, a thousand, or eventually even ten thousand cases, we test them all, and from among them, select the one or two most suitable solutions and deploy them into actual mass production.

[17:38] That is the development goal we set. As you know, since many of you have seen Avengers, just like Doctor Strange predicted one possibility out of tens of thousands, we hope this can do something similar.

How it works, part one: acquiring the test data

So now I'll explain how we're going to implement this. It can be broadly divided into two parts.

[18:07] First, since it's built based on test data, there is a process of acquiring test data. Here we have a large volume of NVH data measured during existing vehicle development stages. We can use that and go through a post-processing step to convert it into a model tailored to the car we want to develop.

[18:36] But if the components of this vehicle are new products that have not been developed before, then inevitably, we have to conduct at least one actual measurement. For that measurement, as I mentioned earlier, we need to analyze the contribution of noise through transfer path analysis and so on. Then, using frequency response functions, we extract quantified audio sources from the data.

How it works, part two: building the virtual environment

[19:06] Next comes the stage of implementing the virtual environment. Generally, when the design department releases the CAD data, we need to take that data and bring it into the Unity environment. I'll briefly explain this again a little later, but using Unity Asset Transformer, we convert the CAD data into a Unity-compatible format and import it. Then we match the extracted noise data

[19:37] and material property data to the virtual environment. Here, there are also parts where we need to use the SDKs provided by these devices, along with those complex algorithms and sensor fusion algorithms, to handle audio rendering and graphics rendering. And when we export it through Unity to the device, it functions as a standalone tool.

Results

[20:12] As for what this technique achieved, I can't give the exact numbers, but at least when we look at the spectrum in our NVH validation, it has an accuracy level of about 95%. From our experience in vehicle development, improving wind noise on an EV means running the vehicle for dozens of laps at the very least. By verifying that in a virtual environment,

[20:43] we greatly reduced the cost of building vehicle samples and prototypes. And usually, when we build handmade prototypes and things like that, a significant amount of time and waste is generated. By minimizing that waste, I believe we achieved sustainable management.

Why we chose Unity

[21:12] Then, to briefly explain why we used Unity, first of all, Unity is already a widely proven and well-known tool in the entertainment field. But Unity is also expanding its ecosystem into industry and simulation, and existing simulation companies are also pushing into the territory of graphics rendering.

[21:41] From that perspective, I believe Unity Industry clearly presents the direction it needs to move in. Among those, what I found especially useful were Asset Transformer and the asset management tools. I also saw tremendous potential in the ecosystem.

Unity Asset Transformer and CAD optimization

[22:09] As you all know, and as I mentioned at the beginning, I'm an NVH engineer, not a software developer. So when developing a program, my biggest problem was this. Especially since I'm not a designer either, I had absolutely no knowledge of handling CAD data like this. The solution that showed me how to turn this CAD data into a format usable in Unity was Unity Asset Transformer. With Asset Transformer, almost all graphics formats used in the automotive industry, such as CATIA, UG, or Pro/E formats,

[22:42] are handled automatically without needing a separate tool, and converted straight into Unity-ready formats like FBX. And most important of all, there is optimization. In our NVH simulation, the most important thing is a seamless simulation environment, in other words, one with no delay. But the design files we generally use now

[23:15] run to tens of millions of triangles in the mesh alone. Now, if I were to take that as is and display it on a headset device like Meta Quest, it would crash outright, basically 100 times out of 100. To show you a quick real-world example, this is the CAD design data from one of our own vehicles. And even this is data we've already minimized quite a lot.

[23:45] Even so, as you can see here, it still has about 29 million triangles. But using the optimize and decimate functions of the Asset Transformer I mentioned earlier, you can see it cut down by roughly a factor of ten with a single click. Right now, this has been reduced to about 2.7 million. So what's the advantage of this? Especially in NVH performance simulation, the visuals being shown are also extremely important.

[24:14] That's because if people sense a gap from the real environment, it starts to feel off, like, "Wait, is this a virtual environment?" With this feature, without sacrificing quality, you can optimize the mesh to a level that runs smoothly on a headset device without overloading it.

Unity Asset Manager and shared materials

[24:44] Next, I found Unity Asset Manager extremely useful. As I mentioned earlier, in other graphics tools or game engines, making materials through things like Blueprints used to make my head spin. Unity is comparatively simple and straightforward, but when it comes to creating realistic textures and materials, I don't really know what light maps are, or how to make height maps, or even where to find that data.

[25:16] From a non-expert's point of view, it's great if you can easily use data that someone more specialized has already made, and this does exactly that. Through Asset Manager in Unity Cloud, I can take high-quality asset data created by experts and bring it straight into the build project I'm working on. Or within the same company, assets created by people who specialize in graphics or rendering can be shared across teams and imported into my own project.

[25:48] That has been incredibly useful. To show you a quick example, you can see the texture clearly, right? This texture has the feel of a real vehicle dashboard. If I tried to make this myself, I'd have to learn how to create light maps and height maps. But without needing to do that, I just went into Asset Manager in Unity Cloud.

[26:16] It was a preset leather material made by an expert, and it turned out Unity provided it by default. I imported it into my project, dragged and dropped it, and just like that, this high-quality texture and material were complete.

The Unity ecosystem and device SDKs

Next, there's the most important part. The ecosystem is extremely important. One thing I really appreciate about Unity goes back to this.

[26:48] A while back, there was a device called the Oculus Rift. It was around 2013, and it came out around the same time as HoloLens. From what I remember, Unity was already supporting virtual environments back then. Since then, a great deal of reference material has built up around that support, especially tutorials and hands-on manuals written for Unity.

[27:20] That created an environment where even people like me, with no background knowledge at all, could develop something using Unity. As a result, major device companies like Meta are developing and distributing their own SDKs based on Unity. So what's the advantage here? People like me, who don't have development knowledge, can run into difficulties when working with hardware interactions.

[27:53] But if you use the SDK, all the prebuilt building blocks for that specific device are already provided. So for things like implementing a user interface, I can build it using the SDK that's already provided, and when it comes to what I want to build, it creates an environment where I can focus more on that. That's the biggest advantage, and it's why I chose Unity.

Demo: evaluating vehicle NVH in a headset

[28:29] So now I'll briefly show you, through a video, what kind of program I developed using this. It's a standalone system where you can simply take a headset anywhere, anytime, and evaluate vehicle NVH. Since it's built from real CAD data,

[29:06] it matches the real vehicle at a one-to-one scale. The dimensions match the real vehicle exactly. As I mentioned earlier, through the Meta SDK, hand tracking and head tracking are implemented directly. The user interface here also uses the Meta SDK.

Making invisible noise interactive

[29:38] This is based on our own test data, and what you're hearing is simulated noise. So as I'm showing you now, noise isn't normally something you can see or interact with as data. So how do we take that noise, something invisible, and interact with it in a way that people can easily understand? That was the idea behind the initial approach I took.

[30:15] And when we evaluate a real vehicle, the hardest part is that it feels like a noise is coming from somewhere in the back of the car, but while I'm doing a test drive, I can't exactly go look for it. But this is played back based on data that I measured on the track. So I can move to the section where the noise occurs and track where the noise is coming from. Here, the feature lets you look at each individual noise source

[30:45] and see what level of contribution it has. You can select them one by one and listen to them. And here, to make the noise data easier to understand, we introduced a frequency analysis feature. It lets you see what kind of frequency characteristics that noise source has, so it's easier to understand. So even if you don't have a deep understanding of noise, you can interact with it, view it numerically, and understand it more intuitively by seeing and hearing it.

Supporting investment decisions

[31:21] So with noise data handled like this, if we reduce costs or change materials, we can simulate how the vehicle's NVH performance would change and provide ideas and insights. For example, say we invest a certain amount of money. If, based on this evaluation, the vehicle doesn't change much at all, then we may decide not to make that investment.

[31:50] So it was developed as a feature that helps decision-makers narrow down their options. So the focus here was to evaluate the vehicle's NVH performance before making a prototype and use that to suggest a direction. Through this, we aim to reach zero emissions, cut carbon emissions, and contribute to sustainable management.

Closing thoughts

[32:24] To give a quick wrap-up, first of all, digital validation is no longer optional. In vehicle development today, it's the direction we need to move in, and it has become one of the ways to get there. And on a different level, let me put it this way. An idea is just an idea if you don't act on it or bring it to life. If you have an idea, bringing it to life is what gives it value,

[32:55] and Unity is a real source of inspiration for making that idea a reality. Thank you.