This article is part of a VB Lab Insights series on AI sponsored by Microsoft and Nvidia. Don’t miss additional articles in this series providing new insights, trends and analysis on how AI is transforming organizations and industries. Find them all here.
BMW Group plans to open a new electric vehicle plant in Debrecen, Hungary, in 2025. By the time the factory goes online, the facility’s layout, robotics, logistics systems and other key functions will already have been finely tuned, thanks to real-time simulations using digital twins.
It’s the world’s first “digital-first” factory and a striking example of the ongoing and growing strategic pursuit of digitalization by manufacturers worldwide. AI is a key part of many efforts. Advances in intelligent technologies and products are enabling new or improved use cases across the manufacturing lifecycle, from product design to engineering to fabrication, testing and assembly.
Digital-first factories represent a leading edge of the global boom underway in “Industrial AI”. Before COVID-19, “Industry 4.0” gained momentum as a vision to accelerate and transform manufacturing. The approach seeks to harness a powerful combination of advanced analytics, AI, cloud technology, robotics, the Industrial Internet of Things (IIoT), human-machine interaction, renewable energy and advanced engineering, among others.
Faced with economic uncertainty and ongoing supply and labor shortages, manufacturers today continue to invest in intelligent technology and infrastructure as key foundations of “smart manufacturing” in this so-called Fourth Industrial Revolution.
In 2023, combined investments by manufacturers are forecast by IDC to account for a hefty 16.6% of $154 in billion global AI sales.
Naturally, goals for AI differ by company. Broadly speaking, manufacturers are deploying smart technologies to help improve current efficiencies and future competitiveness. And, of course, to keep pace with fast-changing market trends and customer needs. Most seek benefits in three key areas:
- Greater intelligence to help increase manufacturing precision, throughput and yields at lower costs
- Improved agility to enable faster product design and prototyping, better performance analysis and a more flexible, resilient supply chain
- Improved sustainability to reduce energy costs and environmental impact
The latter is of growing importance. Many firms face complex, rapidly evolving ESG (environmental, social and governance) requirements. By using less power and material resources, smart factories and manufacturers can reduce consumption, emissions and waste, while increasing materials recycling. AI can help optimize logistics and transportation routes. Generative techniques can simplify the design of more sustainable materials.
New and advanced use cases
How are firms planning to achieve these benefits? Current and planned implementations show heavy investment in maintenance and quality analytics. (See Figure 4.)
Predictive maintenance
An alternative to routine or time-based approaches, predictive maintenance driven by AI can help prevent problems before they happen. GPU-accelerated computing applied here lets manufacturers analyze huge amounts of sensor and operational data faster, with greater accuracy, in real-time, so they can predict failures and schedule repairs. Proactive, AI-driven maintenance can significantly reduce false positives and negatives. What’s more, engineers can use the information to pinpoint root causes of potential problems and take corrective action to prevent future quality issues.
Quality assurance and inspection
QA/QI are top AI priorities for many companies. No wonder; defects cost manufacturers nearly 20% of overall sales revenue, according to The American Society of Quality (ASQ). Sub-par products increase product recalls and warranty costs, and eventually damage brand image, sometimes fatally.
To help detect defects faster and more reliably, many manufacturers have turned to AI-based computer vision applications. Current automated optical inspection (AOI) machines, however, require intensive human involvement and capital. New methods promise to use AI and ML more effectively to improve the quality of manufactured components. They can spot defects like cracks, paint flaws, misassembly, bad joints, foreign bodies like dust, hair and more.
Credit: NVIDIA
A leading-edge approach under development uses object perception and synthetic data to bootstrap training models that can detect specific defects faster and more accurately.
Supply chain resilience and efficiency
The COVID-19 pandemic painfully exposed the inability of many companies to adapt to unforeseen challenges in production and distribution. Worldwide shortages of finished products and parts, from toilet paper to semiconductors, persist today. In a recent survey of manufacturers, 72% of respondents identified disruptions in supply chains and parts shortages as the biggest uncertainty for 2023. Shipment delays remain a top concern, with lead times often twice as long as usual.
In response, nearly 90% of supply chain professionals plan to invest in ways to make their supply chains more resilient, especially with cloud. Many manufacturers are deploying data analytics and AI/ML to better forecast demand and inventory levels, optimize logistics and transportation routes, and coordinate suppliers and distributors. The goal is to prevent and minimize disruptions with improved efficiency and agility.
Manufacturers best positioned to succeed in this new normal will leverage AI and secure, scalable cloud technology and infrastructure. Improved planning and optimization can increase service levels and reduce costs while offering the flexibility to execute in the cloud and at the edge. Better end-to-end visibility lets manufacturers use supply and demand signals to help minimize risk and capitalize on future opportunities.
Slowed by complexity and far-flung data
While manufacturers’ investments in digital and data foundations are booming, the sector’s implementation of operational AI continues to lag other industries.
Difficulties moving AI into production at scale undoubtedly is one big reason why only 10% of 700 companies worldwide surveyed by PwC had completed or were in the late stages of their digital factory implementations. Nearly two-thirds could show only partial results or were stuck at the start of their digital journey.
According to researchers, major culprits include complex system environments and highly diverse and distributed machine landscapes. Many organizations struggle with scaling individual solutions across their entire production network. High implementation costs frequently inhibit progress, too. High costs often are rooted in the need for a specialized technology stack — hardware, software, skills and infrastructure — that must be integrated and optimized for maximum impact.
And then there’s data. Over the last two decades, many discreet and process manufacturers invested heavily in building the digital foundations for a smart factory. New technologies and instrumentation gathered vast amounts of unstructured and structured operational data from machine control systems, videos/surveillance, IoT and other disparate sources for streaming into analytic and AI platforms.
But more data is not always better. Many manufacturers continue to struggle to derive and deliver actionable insights from far-flung seas of OT and IT data. Widespread issues with data quality, availability and centralization often compound the challenge.
Technology advances promise progress
Smart use of new and field-proven foundational cloud technologies, however, promises to help manufacturers overcome these many challenges.
“AI-first” environments
Conventional IT infrastructure – processing, storage, networks, development environment, frameworks, software, virtualization – is woefully inadequate to handle the exponential growth in data sets, complexity, parallelism and the overall needs of manufacturing AI workloads.
“AI-first” infrastructure and toolchains are purpose-built for AI. These bring manufacturers pre-integrated platforms and models that can simplify and accelerate training and deployment, from edge to cloud, while keeping scarce resources focused on impactful data science. A full-stack, end-to-end environment makes it much easier to unify data from many sources. It provides a platform to make data digestible and usable for real-time decisions and model training across the AI production process. Consultancy PWC considers a standardized digital backbone a key building block for factory transformation.
For many manufacturers, smart industrial operations at scale will require cloud-based AI infrastructure. Besides flexibility and scalability, this approach lets companies seize benefits from cost reductions and new capabilities without heavy capital expenses. According to Accenture, shifting or building AI infrastructure using flexible, pay-by-the-use cloud services can yield a 20-40% cost reduction compared to on-premise deployment on underutilized systems. That savings doesn’t include additional savings from power reduction and space consolidation. Further, Accenture says the ability to easily move development QA and training outside of production environments reduces manufacturers’ operational risk.
Supercomputing
Lack of required computing speed throws sand in the gears of many AI efforts. Slow processing extends training, delaying time-to-value. Advanced large language models (LLMs) and real-time requirements further worsen the problem. Applying high-speed computing helps accelerate AI delivery across every stage of manufacturing and can yield a 20x improvement in time needed for training. (See Figure 7).
Cloud-based delivery makes supercomputing more widely available to manufacturers. It provides immediate, flexible access to supercomputing infrastructure and software needed to train models for generative AI and other data-intensive applications.
A new offering from Microsoft and NVIDIA delivers supercomputing as an on-demand service, billed monthly and available globally. It will give enterprises immediate access to the infrastructure, software and computational power to needed to train, build and deploy advanced AI models and applications, from cloud to edge.
Industrial metaverse
“Digital-first” factories like BMW’s and other smart manufacturing applications depend on bridging the physical and virtual worlds. Linking real-time data from physical sensors to their digital replicas in the emerging “Industrial Metaverse” makes it possible to automate, simulate, adjust and predict AI-driven business processes in real-time. Manufacturers no strangers to blended worlds; one in five are experimenting or developing a metaverse platform or solution for their own products, Deloitte says.
New services make it easier for enterprises to leverage the metaverse for smart manufacturing. NVIDIA Omniverse Cloud, a platform-as-a-service (PaaS), gives developers instant access to a full-stack, native and agnostic environment. Connecting with Azure Digital Twins and Internet of Things cloud services lets manufacturers build and operate industrial metaverse applications and accurate, dynamic, fully functional 3D digital twins. As with supercomputing services, Azure provides the cloud infrastructure and capabilities needed to deploy these enterprise services at scale, including security, identity and storage.
These new capabilities can improve manufacturers’ ability to digitally monitor, simulate, control and operate physical assets. That translates into better, faster visibility into operational performance, along with improved ability to predict issues early and course-correct more quickly.
Collaborative development
Integrating 3D platforms with Microsoft 365 Teams, OneDrive and SharePoint lets far-flung groups collaborate in real-time via video, voice and simulations. Accenture recently demoed an impressive early effort designed to shorten the time between decision-making, action and feedback. (See Figure 8).
As approaches mature, technicians in service centers could, for instance, use AR glasses to do complex repairs in a virtual environment, connecting with other experts to work on the problem using digital twins.
A German company has introduced a new technology that lets manufacturers transform 3D data into scalable applications and interactive experiences. Instant3DHub enables developers to collaboratively build, deploy, run and automate applications with “any data, any device, any size.”
And generative AI is emerging as a way to enhance factory automation and operations through software development, problem reporting and visual quality inspection. A new proof of concept by Siemens and Microsoft shows how plant workers and others can use natural speech on mobile devices to document and report manufacturing, quality or product design issues.
Bottom lines: Smarter is smarter
Not every manufacturer will need or want to pioneer state-of-the-art AI. But all can benefit greatly from implementing AI and simulation. For manufacturers and others, improved quality, greater efficiencies, stronger supply chains, and accelerated time-to-value and innovation are the very definition of smart.
Microsoft Azure and NVIDIA are partnering to accelerate AI through GPU-powered Azure cloud infrastructure and solutions that bring manufacturers real-time speed, predictability, resilience and sustainability.
Go deeper:
Microsoft Azure and NVIDIA gives BMW the computing power for automated quality control – YouTube
Transforming Computational Engineering in Manufacturing and CPG
# MakeAIYourReality
VB Lab Insights content is created in collaboration with a company that is either paying for the post or has a business relationship with VentureBeat, and they’re always clearly marked. For more information, contact [email protected].
