The BMW IT Technology Innovation and Research Center in the US is both a research lab and closely aligned to the carmaker’s production in Spartanburg, South Carolina. Bennie Vorster, who leads special projects there, talks to automotiveIT International about industrializing new tech, and making sure AI assists workers. 

BMW model identification

BMW is uses AI to verify built models with customer orders

Outside Germany, one of BMW’s key research organizations for business and manufacturing technology is the IT Innovation and Research Center. With bases both in Silicon Valley, as well as Greenville, South Carolina, the IT center carries out research for systems and tools across the enterprise, including financial services, sales and marketing, engineering, quality, HR, production and logistics. It is part of the carmaker’s central BMW Group IT department led from Munich, which coordinates the company’s enterprise and manufacturing IT backbone.

Similar to other laboratory locations across BMW, the IT center operates to a large extent in research mode. It has a strong connection, for example, to Clemson University, with whom it shares a campus at the International Center for Automotive Research, working closely with engineering and software professors and students. It also collaborates with other academics in the US in areas like artificial intelligence, sensor connectivity and cyber security.

But the center also does projects independent of the university and much closer to BMW’s operational IT. Its west coast base in Mountain View, California works on data analysis projects and aligns closely with tech companies and startups. Meanwhile, Greenville’s proximity and affiliation to BMW’s plant in Spartanburg, South Carolina – the carmaker’s largest globally, building its X series SUVs – allows for close integration with production. The center also links closely to BMW’s sales and marketing across the US and the Americas.

This integration is part of what makes the IT Innovation and Research Center something of a hybrid lab: research-led, but with a strong objective to industrialize new technology and bring it into day-to-day business.

“We obtain difficult to solve challenges from the business and do research and innovative solutions back to the business, mature it towards industrialization and hand that back over after stabilization to operational teams,” says Bennie Vorster, vice-president of information management and special projects at the center.

A similar focus on implementation characterizes the center’s collaboration with startups. Rather than acting as an investor or incubator, the IT center works with these companies on a direct path towards developing them into BMW vendors.

Although there is a close alignment to business need and industrialization, the research center works on highly complex and advanced technology. It has already played an important role, for example, in developing AI-based quality inspection systems, including inspection at “near-metrology levels” – in other words, extremely high accuracy – to compare newly built vehicles with orders, ensuring there are no faults and that the vehicle was built to customer specifications.

The center is furthermore working on neuromorphic – or very low energy – computing and sensors, which could play a key role in production and supply chain visibility.

Bennie Vorster is a South African native and BMW IT veteran who has worked extensively at BMW in IT build, solutions and operations across Germany, the US and the Americas. His background and expertise include electronics engineering and business administration in production, vehicle product parts testing.

Bennie Vorster, BMW

Making AI work for workers

Source: BMW

Bennie Vorster has a long career in group and business IT at BMW, and is now leading cutting edge AI research in business processes and production

Mr. Vorster, what are your top priorities right now at the IT Research and Innovation Center, whether in terms of technology, research or implementation of projects?

Bennie Vorster:  Research and innovation that will enhance customer satisfaction, protect our customer, and deliver high quality products. That remains our priority using technology and research in various fields of AI, security, and other tools at our disposition. We emphasize the human customer as our utmost priority to protect and deliver exciting new solutions in a high quality drive in everything we do.

How do you work with startups and disruptors in the automotive industry? Do you take on an incubator role, or do you do this in coordination with BMW’s corporate venturing division, iVentures?

The BMW iVentures and BMW Startup Garage together through our central IT organization would be the primary entry point where [incubation] would be applicable. Should we find interesting startups and new technologies for investments, we channel those in that direction.

For startups that want to do projects that we deem feasible for research and innovation, which have the potential value and, most importantly, can become industrialized – we would do joint work directly. In this way, a startup can grow and become a longer-term vendor and develop a supplier relationship with BMW.

Artificial Intelligence is expected to revolutionize many business and industrial practices in the automotive industry. In your view, where, if at all, is AI already making a difference in current business processes?

Yes, and in my view it’s not only the automotive industry. Dr. Viktor Mayer-Schoenberger [a professor of internet regulation at Oxford University], echoing many others, has written that AI is “a revolution that will transform how we live, work, and think.”

At BMW, we have made progress on data visibility, prediction and costly repair work prevention, process optimization and vision inspection to name a few.

What are the biggest opportunities for AI in the automotive industry’s processes and manufacturing? And what are the biggest risks?

Let me start by addressing the risks. Any technology, not just AI, can be used for good or bad. In my view, AI and all its subsets – machine learning, deep learning, etc. – should be used within the organization to educate and improve processes and systems. This is a key objective to obtain buy-in and acceptance [from users and workers]. Make it a ‘companion’ or ‘helper’ to the human. The moment your intention is to use it as a ‘stick’, humans will work against the value it is able to provide. 

Now, for opportunities: AI systems will improve business quality of product, process effectiveness and efficiency, protect investment and resources in maintenance. If we take the specific example of supply chain complexity and product market offerings to customers, [AI will] take care of BMW’s valuable customer base by delivering the right products at improved quality and higher value.

It is important to stress that AI – I prefer this to stand for ‘augmented intelligence’ – is there to support employees, relieving them from repetitive tasks, not to replace them.

I understand that you have worked on an AI-based, computer-vision quality inspection system. Can you tell us more about this project?

We are working on near-metrology level inspection on the full vehicle exterior to have a final end-of-production line inspection for quality and conformity to production order. We use multiple AI algorithms that run on GPU-based systems to do real-time inspection within a moving production line, without interruption. We have further made it GDPR (Global Data Protection Regulations) compliant by automatically removing humans working on the vehicle from the AI-inspection process.

Go Innovate! Live: This December 3-5th, Dr. Martin Ziolkowski, data scientist at the BMW IT Technology Research Center, will discuss applying AI in production and business processes. Find out more here.

What are its benefits and advantages?

The main goal is to be able to have near-metrology level inspection on the quality and conformity on every single vehicle, which is difficult and complex for human inspection. It also frees up resources to focus on complex human tasks that require consistent process adherence.

Have you implemented this already? What is the plan to roll it out further?

This is already in production in one location at our BMW plant in Spartanburg, South Carolina. Our challenges were full integration in processes of existing systems. We found the willingness to learn about AI technology and use it by our high-quality production workers exciting and encouraging. And, this two-way street taught us valuable lessons to understand what our production workforce expects AI to do for them – to be a valuable companion and helper.

BMW Spartanburg

Bringing innovation to life

The IT centre’s proximity and alignment to BMW’s massive production plant in South Carolina allows it to test and apply innovation in production and logistics

What is your next objective in rolling out or developing this technology?

We will continue to further improve the system in monthly agile sprint cycles and expand this into the other manufacturing areas. The modular approach of the building blocks we have used allows easy integration in upstream and downstream processes of the BMW Group.

Carmakers including BMW are establishing clouds to connect their plants and supply chains. To what extent will it be possible in the short term to achieve this level of connectivity?

The BMW Group launched its own BMW Group IoT Platform [based on Microsoft Azure] in 2016. Currently there are already more than 3,000 machines, robots and autonomous transport systems interconnected via the BMW Group IoT Platform throughout various plants. 

Are you working on cloud and edge technology that might enable more widespread interlinking of industrial and business processes?

Yes, new edge technologies will enable faster adoption to a cloud direction.

Looking forward, we intend to do research in low-power neuromorphic computing as use cases for edge devices in AI. The future of AI likely will become a digital and analog merge towards a brain interface as well. These very large-scale, integrated circuits can perform massive computation in AI at low power and be scaled down. We have started research in use cases and will evaluate results and industrialization potential in the near future.

Your center has also been involved with demand planning for extremely complex and unpredictable supply chains. Do you envision creating digital twins that could more effectively predict demand – including its impacts across the lower tier supply chain?

I can agree on the complexity and unpredictability [of the supply chain] from experience. The modeling towards a digital twin in process, supply and manufacturing are ongoing research projects in our team to improve the business.

The partnership between the OEM and lower tier supply chain can be improved in using digital twins that hand off to each other in the same manner. Currently we have data interfaces to place orders, logistic, supply information and so on. The goal is AI communicating with AI, each optimizing their business goals, but in harmony to the improvement of the whole ecosystem.

So much of the automotive enterprise runs on legacy systems that cannot be replaced or phased out quickly. Does your research play a role in how BMW might integrate new technology into these large systems?

Legacy systems are following a slower pace of change. However, they can create leapfrog benefits for the business. In my opinion, we should not view ‘old’ systems as a hindrance, but rather make them work with modern technologies and systems. Many parts of legacy systems can branch out to RPA [robotic process automation] processes, interface with AI solutions for tasks and continue keeping systems of record required by the business and compliance as well as legal matters.

In our AI vision project, we did exactly that by actively interfacing with legacy systems. The AI system is also subject to a system of record on decisions it made and which tasks it executed. Therefore, we should not see such systems as obstacles, nor should we use them as an excuse against implementing AI solutions. 

AI camera BMW

Capture and analyse

BMW builds up images in the plant and uses AI to assess whether there might be potential faults

The growing connectivity of industrial processes raises questions around cyber security. You have previously spoken about the risk of subtle, largely undetectable industrial sabotage, which might include an increased fault rate for machines. Does AI play role in detecting and defending against such attacks?

Cyber security remains a constant task. We must also modernize and provide layers of different security systems by humans and AI, as well as process and task segregation.

For example, if a machine fails to constantly deliver a designed quality in operation and expected MTBF (mean time between failures), the question is whether that is due to age, random damage, or injected poor performance behavior for ‘slow economic disadvantages’ towards a competitor?

Using AI algorithms in various layers in your cyber security team will assist in finding potential slow damage algorithms in your ecosystem.

Where do you see the biggest cyber security vulnerabilities in the supply chain today?

We constantly review and audit our processes and systems. In my opinion, smaller suppliers that are not at a level to use cyber security teams should include that as a service with their cloud providers. To be ‘off the grid’ is not possible today.

“Make [AI] a ‘companion’ or ’helper’ to the human. The moment your intention is to use it as a ‘stick’, humans will work against the value it is able to provide.”

Bennie Vorster, BMW IT Research and Innovation Center

AI on the production line

BMW has already implemented a number artificial intelligence – or ‘augmented intelligence’, as Bennie Vorster prefers to describe it – applications into serial production and logistics operations at several plants. Many of these applications have been developed and supported by data analytics teams and researchers, including from the BMW IT Innovation and Research Center.

In many cases, these applications relieve employees of repetitive, monotonous tasks such as checking for warning lights; in other cases, they detect errors with more accuracy. And often, it is the employees who are helping to build the images and information databases that form the basis of deep-learning, neural networks.

AI container recognition

Boxing clever

BMW uses AI to recognize whether containers need additional securing

Automated image recognition

An important focus is automated image recognition. BMW uses an AI-driven process to scan component images in production, comparing them with hundreds of other images to find deviations in real time. The process helps ensure that parts were correctly fitted, for example.

At BMW’s plant in Dingolfing, in Bavaria,AI is also being used to detect dust particles or oil residues that might remain on components from the stamping and body shop, and which workers and cameras have sometimes mistaken for very fine cracks. With the new AI application, the neural network accesses around 100 real images of each possibility (of unblemished components, those with dust and those with oil droplets) and can thus determine if there is in fact a crack or not.

Image scan analysis can also detect and correct problems occurring earlier on, such as where weld metal has sprayed out in several car bodies, or where dust may be on the body before entering the paint shop.

At BMW’s engine and component plant in Steyr, Austria, AI has been implemented to prevent the unnecessary transport of empty containers on conveyor belts. Previously, all containers were transported by conveyor to the removal station for large containers, including smaller ones which then had to be forward to the lashing station for further securing, before moving to their removal station. Using stored image data, the application recognizes whether the container requires lashing or not, and determines those containers that move directly to the lashing station.

As well as parts installation, AI is also in use for aspects of final vehicle inspection. At BMW’s plant in Dingolfing, in Bavaria, an AI application compares the vehicle order data with a live image of the new car’s model designation, including color and drive options, all of which is stored in the image database. If the live image and order data don’t correspond, the final inspection team receives a notification.