Machine learning comes to machinery: AI meets manufacturing
19 11 2024 | Dan Nima
AI is coming for manufacturing, but this won’t resemble the rise of ChatGPT. In the last two years, with the advent of large language models (LLMs) and multi-modal models such as ChatGPT, Bard and Claude, AI has taken centre-stage in public discourse. In fact, until Meta’s Threads recently took the crown, ChatGPT was the fastest digital application ever to acquire 100 million users, just two months after being released in November 2022. As generative AI increasingly edges into more domains, including the creation of written text, computer code, music, video and podcasts, one space on the AI front seems to be comparably quiet; manufacturing.
Manufacturing today is immense and is expected to grow significantly. The total value of globally-manufactured outputs was an estimated $16.2 trillion in 2023 and has been growing almost 3% per year, meaning it could reach $20 trillion by 2030. Hardly surprising, as the world’s populations continue to demand more goods and a higher standard of living. Meanwhile, 20% of global GHG emissions arise from industrial energy use and industrial processes, whilst industry is the second largest consumer of global freshwater, at 20%, second only to agriculture (70%). Individual discussions with engineers only help drive these issues home. One major manufacturer spoke of having to mill (carve out) and waste up to 80% of each large aluminium block purchased to create the components they required. Another mentioned discarding around half of every batch of high-performance materials produced due to microscopic production defects. When it comes to reducing its environmental impacts, it is clear manufacturing needs all the help it can muster.
AI thrives on data and manufacturing produces more data than any other major sector. Manufacturing today’s increasingly sophisticated products, repeatably and reliably, both produces and requires vast amounts of data. From 3D product designs, to bills of materials, machining toolpaths, process control parameters like temperature and mixing rate, quality control data such as colour or tolerances for fittings, the list is endless. The shift towards industry 4.0, characterised by increased automation, sensors, constant digital connectivity and machine learning is likely to bring to light orders of magnitude more data than these organisations have historically had to manage. However, unlike in the case of today’s most popular generative AI tools, this data does not exist on the internet at petabyte-scale for all to find, scrape and train on, generally for free. Instead, it is often closely guarded within the walls of each of the enterprises that manufacture the world’s goods. And rightly-so, since this data underpins these companies’ competitive edge and speaks to their commercial journey of success or failure.
As data threatens to overwhelm manufacturers, AI can help make better decisions, faster. This is where AI thrives; analysing vast amounts of data in real time, identifying patterns, informing decision-making and, increasingly, generating solutions across complex systems. By some estimates, AI could add $13tn worth of total economic activity by 2030, of which ~$1tn could be within the industrial sector alone. This may be conservative, as some leading manufacturers, such as LONGi and Haier, are already achieving 30-40% productivity gains with AI-driven defect root cause analysis and assembly co-pilot assistance, respectively. If these kinds of improvements scale broadly, they could add $6-8 trillion in global manufacturing output by 2030. Key bottlenecks to realising these productivity gains include the pace of industrial digitalisation, largely driven by the adoption of cloud computing, sensors, and connectivity.
Manufacturing has a long tradition of improving the efficiency of resource use. Key milestones in this journey included the broader adoption of the lean manufacturing philosophy, pioneered by Toyota in post-WW2 Japan, which focuses on eliminating the ‘7 wastes’, such as excess inventory, unnecessary movement, and defective products. Alongside this there has also been an increasing shift towards circular economy principles, such as designing products for reuse, repair and recycling. Examples include Phillips setting targets of generating 25% of revenues from circular solutions by 2025, and members of the Consumer Goods Forum, representing 10% of global plastic packaging, voluntarily committing to design for recyclability guidelines. More recently, industry is increasingly electrifying processes, such as for heat or power, that historically depended on relatively-inefficient fossil-fuel combustion.
Building on this tradition, most companies are initially using AI to help do more with less. For example, ToffeeX is using physics-informed generative design to help companies design components, such as aerospace cooling plates, that they claim can be almost 40% lighter than through traditional methods. Guidewheel uses AI to help manufacturers with real-time process monitoring, enabling defect detection and root cause analysis, and have reportedly helped a plastic manufacturer save 52MWh/year through identifying periods where machinery was idling and non-productive. Finally, in cement production, Carbon Re’s AI-driven process control platform claims to have helped achieve reductions in fuel-derived emissions of 5-20% while delivering energy savings of up to 10%.
The age of AI in manufacturing is just beginning, but the potential for climate benefits remains uncertain. Over time, a greater share of the market will turn to AI for more sophisticated applications, such as helping to find higher-performance, lower-impact materials or redesigning inefficient processes entirely. Meanwhile, advancements in both robotic hardware, software and falling costs are expected to substantially lower labour costs as increasingly sophisticated tasks can be automated. This comes with the added benefit of opening up rich new data streams of in-situ feedback on processes and product quality, further driving the flywheel of continuous improvement in factory automation. However, with all the technological potential AI holds, the broader challenge lies in how manufacturers will ultimately utilise these gains in efficiency and productivity. Will they simply redeploy the savings into increased production, potentially offsetting any environmental benefits? Or will they instead ensure these savings are conserved to produce meaningful sustainability improvements?
Many startups are already seizing upon the opportunity to use AI to innovate across the entire production chain and create both commercial and environmental benefits, for example:
Examples of startups using AI within the manufacturing space
In the rest of this series, we will explore how AI is impacting each stage of manufacturing and the potential climate benefits this could generate. This begins with a deep-dive on manufacturing design and simulation, looking at how AI is being applied through material innovation, generative design tools and multiphysics analysis, before moving to later stages of the production process, including execution, quality control, logistics and finally, long-term asset management.
AI is set to transform how we design, produce and distribute products. And given the scale and trajectory of manufacturing’s impacts, this cannot come too soon.
If you’re building a start-up at the intersection between AI and manufacturing with the potential for positive climate impacts, please get in touch.
Special thanks to co-contributors George Darrah, Irena Spazzapan and Louis Millon
Photo credit: Recraft.ai
contact@
systemiqcapital.earth