ROUND TABLE: How AI is reshaping engineering in the automotive industry


Marking the first of our series of round table discussions, we gathered three experts to learn more about the current and potential use of AI in the automotive industry.

AI can address specific original equipment manufacturer challenges, enabling greater product development efficiency and faster time to market, which are crucial in the highly competitive automotive industry. Marking the first of our series of roundtable discussions with automotive industry specialists, we gathered three experts to learn more about the current and potential use of AI in the automotive industry, covering sales efficiency, potential job creation, regulatory aspects and its role in addressing specific OEM challenges such as electric powertrains and battery testing. The roundtable also discusses the UK's AI ambitions, the potential benefits of AI in product engineering and manufacturing, and its impact on decision-making processes. We also consider the future vision of AI in the automotive industry by 2030, including its application in real-time data analysis, supply chain management and waste reduction. Contributions from Richard Ahlfeld, CEO and founder of Monolith; Johan Sundstrand, CEO and founder of Phyron; and Teruyoshi Adachi, CTO of Helixx.

Monolith’s no-code AI software is trusted by the world’s top engineering teams including BMW, Rolls-Royce, Honeywell, Kautex-Textron, Siemens and Mercedes-Benz to develop better quality products in half the time. Its Next Test Recommender (NTR) can provide recommendations for validation tests during the development of batteries and fuel cells. NTR offers a ranked list of critical tests, thereby reducing testing time. Phyron’s AI-powered software creates studio-quality video ads from still images and data, mimicking how a salesperson sells a car offline. It identifies vehicle attributes from images, overlaying relevant sales information to enhance marketing efforts. Helixx’s computational manufacturing approach leverages AI and machine learning (ML) to deliver licensed EV production microfactories within 180 days globally. Its AI-powered Mission Control program instantly detects quality and production issues in real time, reducing the need for costly batch testing and enabling prompt efficiency updates.

The following is an edited transcript of the conversation:

S&P Global Mobility: How can AI help address specific OEM challenges, particularly as many shift to electric powertrains?

Richard Ahlfeld (Monolith): The promise of AI is simple: greater product development efficiency and faster time to market. Products are becoming more complex and yet engineering teams do not have more time. In the 2023 State of AI in Engineering study we commissioned with Forrester Consulting, 70% of US and European automotive engineering leaders believe they need to find new ways to speed up the ideation and launch of complex new products like battery-electric vehicles in order to stay competitive.

By using historical or current engineering test data to train self-learning models, a growing number of visionary engineering leaders are reducing testing time and simulations for products with even the most intractable physics, increasing competitiveness and speeding time to market.

Batteries are a great example of a product that could benefit from the use of self-learning models, reduced testing time, simulations, etc. Battery testing is extremely complex with thousands of design scenarios that all require extensive testing. Additionally, as the physics of complex products like batteries become more and more intractable to understand, engineers find themselves in a dilemma, either conducting excessive tests to cover all possible operating conditions or running insufficient tests that risk the omission of critical performance parameters.

One of our OEM customers has 4,000 test stands running at any one time, generating up to two terabytes of data per week. They do not know what optimal tests to run and do not have the ability to process this vast amount of data. 

On that point, how are you helping your customers address battery testing problems?

Ahlfeld (Monolith): The biggest pain point we see is that product validation teams at vehicle manufacturers do not know which tests are the most critical to run, especially on new technologies.

We recently released a major product update called Next Test Recommender (NTR) that gives active recommendations on the validation tests to run during the development of hard-to-model, nonlinear products such as batteries and fuel cells.

In one fuel cell use case, an engineer trying to configure a fan to provide optimal cooling for all driving conditions had a test plan for this highly complex, intractable application that included running a series of 129 tests. When this test plan was inserted into our software, it returned a ranked list of what tests should be carried out first. Out of 129 tests, the platform recommended the last test — number 129 — should actually be among the first five to run and that 60 tests are sufficient to characterize the full performance of the fan, a 53% reduction in testing. This is game-changing technology.

Can AI help in automating decision-making processes in manufacturing?

Teruyoshi Adachi (Helixx): When making a decision, AI will not only compare the benefits of inputs and outcomes, but also quantitatively evaluate the cost and inherent risks associated across the board. The difference comes through identifying what can be immediately automated and what takes a longer time to automate, but in many cases, each decision can be automated.

How does AI assist in circularity and waste reduction in manufacturing?

Adachi (Helixx): In today’s physical-first industrial systems, efficiency of production is achieved by force through economies of scale. Throughout industrial manufacturing, we are increasingly seeing the destructive and wasteful effects this production method leads to, from fast fashion to industrial farming. To use an automotive analogy, it is the equivalent of entering the corner with full acceleration and then slamming on the brakes with full force, entering the corner at high speed, leading to inevitable oversteer and wasted energy with screeching tires, scattering noise and brake dust every time. It is a non-circular method. However, in Helixx’s computational manufacturing methodology, AI gives us the ability to take each turn precisely, like a ballerina as bright and smooth as if dancing on swan lake; there is no wasted energy, so we fundamentally reduce our energy demand and material waste in step with each other.

On the retail side, how can AI help dealers and OEMs sell cars more efficiently?

Johan Sundstrand (Phyron): Phyron’s AI-powered software makes studio-quality video ads from still images and data, removing the need for dealers to generate video themselves. Machine learning algorithms have been taught to recognize hundreds of thousands of vehicle attributes from every angle, so they can identify exactly what is in an image and overlay the corresponding sales information.

The AI can process an unlimited number of videos concurrently, taking about 10 minutes to render them all, posting them live on a single retailer site, or scaling them across an entire network if needed. If a detail changes, such as price, or promotion, the data feed updates, and the video is re-rendered automatically as part of the service.

With this kind of technology, we believe there’ll come a point where AI can get very close to how a salesperson sells a car offline.

The UK Prime Minister Rishi Sunak has made AI a key priority and wants the country to become a global hub for the sector. Is this realistic?

Ahlfeld (Monolith): The head of the UK’s artificial intelligence taskforce has played up the country’s ability to take a seat at the center of the AI revolution. The UK has some catching up to do, but I’m optimistic the country can claim a prominent position.

There are obviously many different things that fall into the bucket of AI and intelligent computation – with most attention recently focusing on large language models. But beyond the headlines, there are a multitude of different model architectures that can be even more impactful. This is where Monolith fits in, with what we call the ‘most sensible’ form of AI. It is the ability to learn from data, allowing you to understand something so complex that normally you would be unable to understand. Machine learning from data is a fantastic shortcut through a jungle of complex things.

These smaller, “classical” models focused on specific questions can bring about profound advances, and where the UK has an opportunity to grow its expertise.

How should AI in industry be regulated?

Sundstrand (Phyron): Regulating AI is a challenging, probably even impossible task, because the technology is rapidly evolving and can be applied in different ways. AI is still in its early days; regulation could harm progress and innovation in the field.

We understand that the accuracy of ChatGPT can sometimes be called into question, but that is where you need a trained specialist working alongside the technology.

Can product engineering benefit from what we have learned by using large language models like ChatGPT?

Ahlfeld (Monolith): ChatGPT nicely visualizes through text how much more you can get out of data. Essentially, the software is taking existing data and delivering an output that the end user finds interesting or useful. That is exactly what Monolith does with AI in engineering product development. However, unlike ChatGPT, engineers do not need that much data to train a self-learning model. We leverage the data that exists, and is often going unused, to deliver new engineering insights and accelerate product development.

With this outcome, it is clear that self-learning models will become a standard tool for engineering. As AI becomes a trusted part of the product development process, we expect engineers across automotive and other industries to significantly reduce verification and validation steps that currently take weeks or months. Of course, there are areas in which AI is more suited than others, but the wheelhouse for our AI software is firmly located in deeply complex engineering problems where the physics are intractable, and the number of parameters are extensive.

What kind of jobs can be created by AI?

Sundstrand (Phyron): An important consideration for using AI software is knowing how to ‘prompt’ effectively. The more specific details you give on exactly what you want, the better, and be prepared to work with it a bit to get the content you want. We absolutely see a need for new jobs working with technology like ChatGPT — specifically ‘prompt engineers.’ At the moment, the majority of people are treating engagement with ChatGPT as a bit of fun, but we are rapidly approaching a moment where the interaction can and should be professionalized. There will be exponential growth in consultancy services where prompt engineers and ChatGPT specialists will be in high demand.

Is AI a risk to engineering jobs?   

Ahlfeld (Monolith): It is a fair question. There is understandable anxiety among knowledge workers that AI could eventually take work away from humans. However, we see much more upside than potential risk of downside. Where AI might replace jobs at some point down the line, I believe this technology will not only foster greater engineering creativity but also create many more new jobs. If we are going to have an economy that grows, we need to reinvent how we do things. We cannot keep doing things the same way and expect progress.

The aim of implementing AI technology is to reduce monotonous tasks and enhance engineering productivity and creativity. In the areas where our AI technology is applied, we are seeing the incredible innovations our customers are realizing in new product development. With Monolith, engineers at Kautex-Textron are solving previously intractable vehicle acoustics challenges, and at BMW Group, by optimizing crash test performance earlier in the design process, the team is reducing dependence on costly, time-intensive testing.

Do you think AI can replace humans in auto retail?

Sundstrand (Phyron): It is only a matter of time before AI is selling cars as effectively as a human salesperson, and this could happen as soon as 2025. The speed at which self-learning software is developing and being embraced by retailers means that a fully competent AI-powered sales bot is as close as 18 months away.  

This prediction comes at a time when we have made massive leaps with our own AI-powered software recently. 

Our “always on” paid ads platform helps dealers and OEMs reach new audiences with full AI video and media buying automation. This all-in-one ecosystem automatically creates studio-quality videos and takes care of media spend for every car in stock, with zero human input.

We have also started to enhance our videos with bespoke, fully automated voiceover. This functionality is initially available in over 22 languages with options for four male and four female voices in each language. Every characteristic of the voiceover can be adjusted, including the pitch, tone, speed, voice equalizer and the music gain.

We know the automotive retail market is more competitive now than ever before. In this environment, anything that gives a company a competitive edge and increases efficiency should be welcomed.

How does AI optimize supply chain management in manufacturing?

Adachi (Helixx): Supply chain risk occurs because systems are interconnected in the supply chain, and these systems are only as strong as the connection that holds them together. The Helixx system weaves a smart and consistent thread through the entire fabric of supply chain production and usage. The functional area of cyber risk management in the supply chain is exactly what we are developing at the moment. Our goal is to eliminate excess and shortages in the supply chain as much as possible through ultra-efficient inventory management, as described earlier..

What is the significance of a machine learning go-no-go program?

Adachi (Helixx): The fact that the entire manufacturing process evolves through machine learning is a defining feature of what we call computational manufacturing. Helixx is not simply a digital layer on top of a physical process like most modern ‘smart factory’ concepts, it is digital from the first principles. Go-no-go is one of the primary functional roles of Helixx’s application of ML; its capabilities extend across cost, quality and supply chain to instantly detect and analyze the real-time twin against the digital twin and resolve inefficiencies or defective processes. In other words, you could take a process like Darwin's theory of evolution and repeat it at ultra-high speed to rapidly learn and evolve from the manufacturing process to the end product. Darwin said that smart things survive, not strong ones. AI and ML have enabled us to evolve smarter, rather than by sheer power or size.

Can machine learning help OEMs improve technology through better testing?

Ahlfeld (Monolith): Absolutely. Every OEM has the challenge of developing or picking the best battery system for their technology, and they need to go to market with total confidence in how it performs — particularly around safety. We know that engineering a battery pack requires a huge amount of testing because it is a complex new technology and in many areas is unproven in the field. We also understand that engineers at major OEMs are feeling increased pressure when it comes to guaranteeing the thermal performance of battery-electric vehicles in real-world scenarios.

AI can give added peace of mind to OEMs, because ML models are typically more accurate when predicting the lifetime performance of batteries. ML models typically register 1%-3% inaccuracies, whereas most standard models, like equivalent circuit models (ECMs), register as much as 10%. ECMs are currently still used because they are easier to embed into operating systems, but they have a higher degree of inaccuracy, and the industry knows it. Most battery labs are trying to find ways of incorporating ML tools into their durability models because they are more accurate and result in better-performing batteries.

What are the cost implications of detecting quality and production issues in real time?

Adachi (Helixx): Our computational manufacturing ecosystem allows us to instantly spot any quality or production issues in real time, which in traditional automotive manufacturing are only spotted in ‘batch testing’ when it is effectively already too late to resolve, leading to significant cost implications. In short, we can fundamentally reduce the frequency of defective products. Take Toyota's Just In Time model: as long as the production timeline is accurate, there is no need to have excess inventory, and you can make logistics more efficient. Helixx uses 85% of parts sourced locally in some way, leading to improved productivity and reduced environmental impact on the local economy. In other words, partnering with Helixx is an enterprise development opportunity for local businesses and suppliers to scale their own businesses in tandem with Helixx licensees.

What will AI in manufacturing look like by 2030?

Adachi (Helixx): Because we have already started to implement some of what we can today call AI, in the coming years, we can use this AI to create the next AI. Therefore, we can replicate the same level of advancement from the past 10 years in only 2 years, and then what we make in the next 2 years should be achievable in only a matter of months. The pace is extraordinary. Therefore, the world in 7 years is not an extension of 7 years of our creation based on current capabilities. Our goal at Helixx is to use this pace of advancement to ultimately accelerate and empower sustainable development, boost local economies and tackle the air pollution and mobility issues that densely populated megacities in developing regions are grappling with. Only time will tell, but we will not be waiting as long as we have, that is for sure.

Ahlfeld (Monolith): ML is becoming an increasingly important part of our personal and business lives, either as a conscious decision by the user or subtly through the basic tools we use on a day-to-day basis.

As we see with the teams at BMW and Kautex-Textron, as just two examples, AI software is transforming how automotive engineers develop complex products. By 2030 we think every engineer will be an AI engineer. On our way to this aspirational goal, we have defined a near-term vision that by 2026 we will empower 100,000 visionary engineers to use ML to halve their product development cycle. Because the Monolith platform was built by engineers, you do not need to be a Python coder or data scientist, just a domain expert in your field. Our no-code AI software was built by engineers specifically for engineering domain experts to quickly understand and instantly predict complex physics where simulation tools and traditional research and development (R&D) methods fall short and have a slow time to market.

We are entering an era where AI usage is a hybrid of consumer applications and domain-specific tools for a targeted set of use cases. For Monolith, our solution is squarely the latter, designed for R&D leaders seeking to make the verification and validation of new products more efficient and their business more profitable.

Sundstrand (Phyron): We believe that the use of digital immersive video will play a crucial role in the fully automated car-buying process. By leveraging the latest advancements in digital video production, customers will be able to experience the cars in an interactive and emotional way, allowing them to make an informed decision about their purchase.

Whether it is through virtual reality or 3D visualization, digital immersive video will become an essential element in the car buying process of the future. It will enable customers to have a more realistic and engaging experience and will be an important step in convincing them to make the purchase. We expect to see a significant increase in the use of digital immersive video in the car industry as more companies adopt this technology to enhance their sales process.

S&P Global Mobility comment:

Graham Evans, Director Electrification Technology Research in S&P Global’s Supply Chain, Technology & Aftermarket Group noted, “With the advent of AI, ML and big data processing capabilities, it is evident that society will be able to iterate quicker, more powerfully and more effectively than ever before to solve problems. For the automotive industry, this evidently means solving complex engineering problems more robustly, in a shorter period of time, with less costly physical testing. As a former NVH Engineer, seeing complex multidimensional acoustic phenomena being tackled in the manner of the example cited by Monolith and Kautex-Textron is very illuminating. Issues like fuel slosh, brake squeal and gas rush, which are incredibly difficult to measure repeatably, will be characterized much more effectively and solutions homed in on more rapidly with this technique. Buzzwords aside, fixing problems faster benefits vehicle quality and development times, liberating engineers to perform more creative development-focused tasks. It is an exciting time for engineers embracing and harnessing the capabilities of such tools.” 

preload preload preload preload preload preload