Artificial intelligence (AI) will radically transform manufacturing in ways that we can hardly imagine today. Manufacturers and technology providers with comprehensible data that AI systems can understand should benefit the most. In his “Industrial AI and the Factory of the Future” report on how AI could impact production, Ryan Kelly, the vice president of technology at AMT – The Association For Manufacturing Technology, argues that the technology will be at least as transformative as connectivity (Industry 4.0), computer controls (CNC), and electrification have been in the manufacturing space. “The AI-native factory is one where machines, software, and human operators share a continuously updated model of what is happening – sensors feeding data into models, models informing decisions, decisions producing new data,” Kelly writes in the report. “The loop tightens until the coordination overhead that once required armies of planners, expeditors, and analysts becomes something the factory manages largely on its own.” Getting to that stage demands manufacturing data that multiple systems can access and understand. Many early AI efforts are failing because of that lack of understanding between systems. In his report, Kelly describes the root of that problem and what manufacturers can do about it. Where We Are Right Now Manufacturers have been using AI-powered systems for decades. Intelligent toolpath planners that avoid collisions between CNC machine heads and workpieces rely on AI, as do quality and metrology systems that quickly identify flaws in products. What’s new are generative AI systems that can act somewhat autonomously… eventually. Most of what companies are offering right now are AI tools added onto existing systems, such as chatbots that allow tool users to ask for help. Kelly says those tools are valuable because they improve user experience and the performance of some systems, but they don’t radically reshape how factories operate. “This transformation is likely going to take a very long time,” Kelly writes. “Electrification was commercially available to factories by the 1890s; it took until the 1920s for most American factories to operate primarily on electric power. Digital systems in manufacturing date back to the 1960s; widespread adoption of MES and ERP in small and midsize manufacturers did not occur until the 1990s and 2000s. The AI transition in discrete manufacturing – especially the shift from AI as a feature to AI-native systems – is likely to unfold over a similar timeframe: not five years, but a few decades.” What the Future Could Hold Revolutionary technology changes in manufacturing have all fundamentally altered how people produce things in systemic ways. Systems powered by steam engines or water wheels brought more muscle into manufacturing, but the bigger changes came from moving manufacturing out of homes and shops into centralized factories. Electrical power introduced flexibility, allowing companies to build factories in new locations and place machines wherever they wanted them instead of using a centralized belt or shaft. That freedom led to greater complexity and modern measurement practices such as time studies and Gantt charts. Computer controls greatly expanded standardization in factories, allowing identical processes in Cleveland or Bangkok. The biggest change, Kelly argues, was not the systems but the visibility they offer. Engineers in multiple locations can track system performance anywhere they have operations. AI’s potential game-changing feature? The removal of latency. Today’s connected systems allow engineers across the globe to see the same problem, but vision doesn’t equal action. Choosing how to interpret that data and make changes requires time-consuming studies and countless meetings. “A manager had to read the dashboard. An engineer had to reconcile the numbers. An expediter had to call the floor,” Kelly writes. “The AI transition makes it possible for software to participate meaningfully in that step, collapsing the time between observation and response from hours or days to seconds.” Reflexive Production Kelly calls this potential AI-driven factory of the future reflexive production. Actions lead to data lead to reactions. As he described it earlier, “Sensors feed data into models. Models inform simulation. Simulation informs decisions. Decisions produce actions. Actions produce new data. The model updates. The loop closes.” That does not mean completely autonomous operations, however. Engineers and managers will continuously evaluate and update operations. Users of these new AI systems will set up parameters for what things they’ll let the system determine and what will require a human decision. “AI doesn’t replace human decision-making. However, the coordination cost of human decision-making collapses because everyone and everything is working from the same picture,” Kelly explains. “The plant manager, the process engineer, the scheduler, the machine tool, and the quality system all access the same operational ground truth. Each can act on it without waiting for translation projects, reconciliation meetings, and time-consuming data-cleaning exercises.” How To Prepare for an AI Future Technology thought leaders have written dozens of articles about how to adopt AI systems in existing operations. The general advice: Start small with a specific problem to solve, and use that project to get comfortable with the capabilities of new systems. As comfort grows, apply the technology to more operations. It’s good advice for the current tools that Kelly describes as having AI as a feature. But, to get to where AI manages significant portions of processes, manufacturers need to focus on some often-ignored parts of their systems – syntax, state definitions, and basic descriptions of what’s happening on the shop floor. “(Manufacturing) production data is typically siloed across multiple disconnected systems – ERP, MES, SCADA, QMS, and individual machine controllers,” Kelly writes. “Each system has its own definition of what a ‘part,’ an ‘operation,’ a ‘downtime event,’ or a ‘work order’ means. Joining data across systems requires a translation project every time.” And, he adds, those translation projects are larger and costlier than most companies expect. Many AI pilot projects fail or stall because engineers must spend too much time translating data before the new systems can learn from existing operations and make improvements. “While the AI works in the controlled pilot environment where it was deployed, the conditions required for it to work at scale do not exist in the broader factory – consistent data pipelines, semantic interoperability between systems, and organizational processes that trust and act on model outputs,” Kelly says. “The technology was not a constraint. The infrastructure beneath it was.” Technology standards, such as MT Connect, can provide a common language for software, machine tools, robots, material handling systems, and the other pieces of a manufacturing technology stack. Those who start early on standardizing their definitions will have an easier time adopting new AI systems as they develop, Kelly argues. Conclusion While manufacturers were early adopters of machine learning and other AI technologies, the new wave of generative systems promises a far higher level of disruption and opportunity. Those changes have barely begun and could take decades to fully develop. To prepare for the upcoming transformation, manufacturers should carefully study their current technology stacks and make sure they have data that can be interpreted by intelligent systems.Download the full “Industrial AI and the Factory of the Future” report here. Further Reading: Industrial AI Finds Its Niche at IMTS 20263 Ways Job Shops Can Tap Into the Data Center Boom4 Lessons for Scaling Industry 4.0 Across Multiple PlantsPredicting Manufacturing Technology’s Future “We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.” — Amara’s Law, a 1978 comment by the late Institute for the Future President Roy Charles Amara. Think back to the early ’90s when the internet broke free from a text-and-data archive for researchers and became a commercial force. Market enthusiasm was so intense that dial-up provider America Online could buy legacy media company Time Warner. Predictions that every device and process that everyone used worldwide would need an internet connection drove massive investments from data carriers and other digital infrastructure players. Investors lost billions as the build-it-and-they-will-come enthusiasm failed to produce a return on investment. Looking back 30 years later, however, most of those predictions came true. The industrial Internet of Things and Industry 4.0 systems are a direct offshoot of those overheated predictions. The internet truly is embedded in all aspects of daily life, and the miles of fiber-optic infrastructure that failed to produce financial returns in the ’90s are at full capacity today. We could be seeing the same thing with industrial artificial intelligence (AI). The technology promises to radically transform the art and science of making things, but that revolution could take decades to develop, and we’re at the earliest stages. Much of the world-changing hype around AI is likely wrong today, but don’t take too much pleasure in competitors’ failed pilots or the constant parade of AI embarrassments. Those big predictions could someday prove true. For manufacturers, it’s critical to remain highly skeptical about hard-to-believe claims from technology providers. Skeptical, however, not completely distrustful. Short-term claims may not pan out, but systemic change will come…eventually.
Generative artificial intelligence promises to eventually change nearly every aspect of manufacturing, but new systems could take decades to develop. To prepare for the change, companies should examine their existing technology stack and data practices to make sure all systems can understand each other.
Subscribe to Newsletters
Get the latest news from IMTS delivered to your inbox.