Best of LinkedIn: Smart Manufacturing & Industrial AI CW 31/ 32

00:00:00: Provided by Thomas Allgeier and Frennis based on the most relevant LinkedIn posts about smart manufacturing, an industrial AI in calendar weeks thirty one and thirty two.

00:00:09: Frennis is a B to B market research company that supports smart manufacturing providers with building feature-by-feature competitive intelligence That shows exactly how their product stacks up against the competition.

00:00:21: you can find more info In the description.

00:00:23: imagine Running a massive gas production plant And suddenly your control systems just go completely blind.

00:00:29: Oh, wow.

00:00:30: Yeah You're forced to operate entirely on manual controls for six solid hours.

00:00:35: No dashboards no predictive alerts

00:00:37: just

00:00:38: raw operator intuition keeping the whole facility from a catastrophic failure.

00:00:42: That's terrifying right.

00:00:44: but today we're cutting through this endless hype To look at the physical high stakes reality of smart manufacturing.

00:00:49: We certainly are and you know if you are tuning into this deep dive?

00:00:53: You were likely navigating in just an absolute flood of information out there.

00:00:56: oh yeah so much noise

00:00:58: Exactly about digital twins, industrial artificial intelligence.

00:01:01: The factory of the future.

00:01:03: So our mission today is to filter out that noise.

00:01:06: We've reviewed the most relevant conversations happening among manufacturing and Industrial AI professionals on LinkedIn And we are distilling it down To the critical mechanics That you actually need to understand.

00:01:18: Absolutely no fluff just focused insights On the trends that Are fundamentally rewiring the industry.

00:01:24: Let's

00:01:25: get right into it

00:01:26: then.

00:01:26: Yeah so Yeah.

00:01:28: Because before any organization can deploy these advanced AI models or autonomous robots, there's a strict prerequisite that has to be met and that is the data foundation

00:01:39: Right?

00:01:39: The plumbing

00:01:40: Exactly!

00:01:40: The plumbing Data readiness isn't something just magically occurs as a byproduct of buying an AI tool.

00:01:46: It is deeply intentional structural work.

00:01:49: Okay let us unpack this because it really seems like classic trap facing C-suite today right?

00:01:54: Yeah

00:01:54: big time

00:01:54: Leadership teams, they want to talk about the flashy AI dashboards and automated workflows.

00:02:01: But very few people wanna address that frustrating unglamorous work of standardizing raw machine data first.

00:02:08: Nobody wants do boring stuff Right

00:02:10: but AI isn't magic.

00:02:12: It is just advanced mathematics.

00:02:14: applied.

00:02:14: your data If you're data is a mess well... The math fails

00:02:18: Which why organizations actually succeeding are embracing that unglamorous plumbing?

00:02:24: Peter B recently highlighted a fantastic case study about CDE Group.

00:02:28: They're a global leader in wet processing equipment.

00:02:31: Okay, what did they do?

00:02:32: Well their leadership didn't just jump straight into artificial intelligence... ...they focused entirely on moving from basic disjointed data collection to generating structured industrial IoT insights Makes sense And then used Siemens Insights Hub To Do It.

00:02:46: What's fascinating here is the sheer scale that you achieved.

00:02:49: Right because I actually have to push back a little bit Here On Their Timeline.

00:02:52: Yeah, because going from just two connected assets to over eighty plants globally in only eighteen months.

00:03:00: I mean In the industrial world traditional IT or PLM rollouts can take years.

00:03:05: Yeah usual truth.

00:03:06: yeah.

00:03:06: So how did they bypass that massive deployment friction?

00:03:10: Well The mechanism They used to bypass That Friction was strict standardization.

00:03:14: So instead of treating every new facility like a custom project, you know building bespoke API connectors for every single machine They created a unified data model.

00:03:23: Got it

00:03:24: Think about defining universal socket.

00:03:27: when plant number eighty came online its datastreams simply snapped into the existing architecture Like

00:03:33: a Lego brick.

00:03:33: that standardization is really The only way to reach the velocity You need to eventually deploy AI across the global enterprise

00:03:40: and that structural discipline.

00:03:42: It really applies across every vertical.

00:03:45: Cedric Nike recently shared some comparative insights on two drastically different companies, you've got Gravathy which is building a next generation plant for low carbon iron and then Sanofi the global pharmaceutical giant.

00:03:58: I mean You cannot get much further apart than smelting iron in manufacturing vaccines completely

00:04:04: different regulatory and operational universes.

00:04:06: yet they are executing their digital transformations using They're actively getting pilot purgatory.

00:04:15: Oh, pilot purgotory is the

00:04:17: worst!

00:04:17: Right that phase where a company builds this shiny AI proof-of-concept That works perfectly on one machine but it completely breaks when applied to the rest of the factory Yeah

00:04:27: just doesn't scale.

00:04:28: Exactly So.

00:04:29: Sanofi is scaling their digital foundation across more than fifty manufacturing sites.

00:04:34: first they are standardizing how their batch data is structured which finally gives their AI models the trusted, uniform data they need to drastically speed up batch reviews and release times.

00:04:45: And Gravity is actually taking that logic a step further – They are building their future plant as digital twin before physical concrete has even poured!

00:04:53: Wait really?

00:04:54: Before it's

00:04:54: built?!

00:04:55: Yeah….

00:04:56: By defining this data architecture in virtual world first …they guarantee when the physical plant goes live ...it acts like learning factory from day one.

00:05:04: It automatically optimizes complex variables such as hydrogen use.

00:05:08: That is wild, but looking at these massive structural investments I want to ask you a broader maybe more challenging question.

00:05:15: Sure go for it.

00:05:16: When we see companies scrambling to build these foundations Is this just a localized technology trend or are we looking at a global infrastructure boom?

00:05:24: because It feels similar too like the race to lay the transcontinental railroad Or build the early internet backbone

00:05:32: that infrastructure comparison is highly accurate.

00:05:35: It is fundamentally a geopolitical and economic race.

00:05:38: Yeah, Erdem Uren actually shared a really encyclical analysis on this.

00:05:42: he ranked the world's leading regions On their industrial AI ecosystems breaking down exactly how The resources are distributed.

00:05:49: right.

00:05:49: so the United States clearly leads in AI computing power Foundation models and cloud infrastructure.

00:05:55: Meanwhile China possesses unmatched manufacturing scale And just the sheer volume of industrial data that comes with it

00:06:01: Because of all the factories.

00:06:02: Yeah, exactly.

00:06:03: and Europe sits in this unique position leading in deep engineering expertise automation And specialized industrial software

00:06:12: which means no single region holds all the cards.

00:06:15: I mean you also have South Korea dominating the semiconductor supply chain and Japan leading in precision robotics hardware.

00:06:22: Exactly the point, but the most critical takeaway from that analysis and how this connects back to our data foundations is That the ultimate winner of this economic race Will not be The entity that trains the smartest Most sophisticated AI model in a vacuum.

00:06:38: Okay then who wins?

00:06:39: The

00:06:39: winner will Be the organization they can physically integrate those models into millions Of active machines and production lines.

00:06:45: first.

00:06:46: Flawless execution and data readiness matter far more than the algorithm itself.

00:06:50: But here is the wall that companies hit, right?

00:06:53: Even with perfectly structured raw data if the mechanical engineering team calls a specific valve Component A And the software team labels it sensor-twelve.

00:07:02: The AI is completely paralyzed.

00:07:04: It has no idea what its looking at

00:07:06: Exactly!

00:07:06: The AI doesn't know they are same physical object.

00:07:09: you have to contextualize the data which forces us look at our next major theme today, The Digital Thread.

00:07:15: Yes because manufacturing complexity has vastly outgrown the era of isolated software applications.

00:07:21: You just cannot build efficiently when your mechanical electrical and software domains operate in silos.

00:07:26: It's impossible now Right.

00:07:28: So establishing a digital thread Which is continuous seamless stream Of data connecting engineering, manufacturing purchasing And downstream operations it no longer optionally.

00:07:39: It's a strict business imperative.

00:07:41: Here's where it gets really interesting, because the digital thread is what forces these theoretical data models to map onto physical reality.

00:07:49: Ornevan Neuwenhoijs discussed MoVoo Robotics as major warehouse automation player over in Belgium.

00:07:54: They fundamentally changed how they use product lifecycle management or PLM software.

00:07:58: Right!

00:07:59: Because historically A lot of organizations merely used PLM As a glorified digital filing cabinet Just for storing CAD files

00:08:06: Exactly.

00:08:07: Instead, MoVu Robotics treated it as the central nervous system for their entire operation.

00:08:12: I was looking into mechanics of how this works and by establishing a single source-of truth if a mechanical engineer updates physical dimension on robot arm.

00:08:22: The software team immediately sees how that constraint changes their control logic.

00:08:26: That's huge!

00:08:27: There is no waiting for a weekly status meeting or, you know hunting down an email attachment labeled version three final everyone operates on the exact same real-time data which drastically reduces the risk of expensive rework on the manufacturing floor.

00:08:42: In that concept of a single source of truth becomes exponentially more powerful when you start layering generative AI on top it.

00:08:50: Oh absolutely

00:08:51: Bart Hubenach detailed how this integration is currently playing out in the pharmaceutical sector.

00:08:56: By establishing a robust digital thread that connects early engineering, tech transfer and active plant operations these companies are achieving things that would have sounded impossible just a few years ago.

00:09:07: So let's break down what actually looks like on say Tuesday morning at an engineer department.

00:09:12: Okay so they're using generative AI to automatically generate highly complex process flow diagrams and instrumentation diagrams.

00:09:22: Engineers can use simple natural language prompts to create structured connected plant information.

00:09:27: That sounds almost too easy.

00:09:28: Right, if an AI was just guessing that would be incredibly dangerous.

00:09:32: But because the large-language model is anchored into a governed digital thread it only pulls verified approved engineering specifications.

00:09:40: And for anyone outside of pharmaceutical space That anchoring is vital for GMP good manufacturing practice compliance.

00:09:47: Yes, absolutely critical.

00:09:48: the AI isn't hallucinating some random pipe fitting.

00:09:52: it Is placing a specific compliant component based on strict regulatory standards that are locked inside that digital thread.

00:09:59: It automates the heavy lifting of manual documentation freeing up The engineering teams to focus on actual process optimization which

00:10:06: what they should be doing anyway

00:10:07: Exactly, and that perfectly illustrates a concept shared by John Nixon regarding a major brewery utilizing digital twins and virtual controllers.

00:10:18: Think about it like this... Trying to apply AI TO your factory without a digital thread is like trying to bake a cake but you're ingredients are locked in five different kitchens across town.

00:10:29: That's perfect analogy.

00:10:31: The intelligence of the baker is irrelevant if they can't get the ingredients together.

00:10:36: Right,

00:10:36: you cannot unleash artificial intelligence on operational data that hasn't been deliberately connected and contextualized first Absolutely.

00:10:44: By the way If your finding this deep dive helpful as you navigate your own operations and strategy make sure to subscribe so don't miss our future additions.

00:10:52: We want ensure getting this actionable intel delivered straight into your feed.

00:10:56: So we've explored the underlying data foundation And woven a digital thread through engineering domains.

00:11:02: Now we move from the virtual world to the physical shop floor

00:11:05: and The shop floor itself is undergoing a radical evolution for decades.

00:11:10: The focus was just on automating physical tasks.

00:11:13: But today, We are actively virtualizing the automation itself.

00:11:17: Yes, the industry is undergoing this massive shift From hardware-defined automation To software defined automation.

00:11:25: Jacob Abel broke down the technical reality of Virtual PLCs or VPLCs For a very long time, I'm a programmable logic controller.

00:11:33: You know the physical brain controlling machine was proprietary hardwired box.

00:11:37: right if you bought into specific vendors hardware ecosystem your locked in.

00:11:41: So to clarify the mechanics here, what does virtualizing that PLC actually entail?

00:11:46: Does it mean we are sending a factory's critical control logic up-to-the-cloud and back?

00:11:50: because I mean from a latency perspective relying on an internet connection stop at high speed robotic arm sounds incredibly dangerous.

00:11:56: Oh well

00:11:56: That is most common hesitation.

00:11:58: And Jacob directly addressed you absolutely do not run real time Control Logic From The Cloud.

00:12:02: Okay good

00:12:03: Yeah they would introduce unacceptable Latency and massive single points of failure.

00:12:07: Instead A Virtual PLC Is A Containerized Application

00:12:10: Meaning the software is bundled up in its own secure, isolated digital environment.

00:12:16: Correct.

00:12:16: and that container runs on a standard industrial PC located right there on the edge physically next to machine.

00:12:23: it stays completely local.

00:12:25: but because of control logic as now just software running into container It removes that hardware vendor lock-in.

00:12:31: That's

00:12:32: game changer!

00:12:32: It really is...it delivers near native performance incredibly low latency And high data throughput required to feed AI models onto floor.

00:12:41: This raises an important question, though regarding cybersecurity at the edge which we absolutely must address.

00:12:47: We definitely will but before we dive into those security vulnerabilities The operational upside of this Edge intelligence is just undeniable.

00:12:55: Mark Campbell shared a striking example involving Blue Scope-the Australian steel manufacturer.

00:13:00: They implemented Siemens SunSci Predictive Maintenance Technology across their operations

00:13:04: Which again relies entirely on processing high volumes connected Edge data

00:13:07: Exactly and results speak for themselves.

00:13:09: Since twenty-twenty two, Blue Scope has saved roughly two thousand hours of unplanned downtime.

00:13:14: Two thousand hours?

00:13:15: Yes!

00:13:16: To understand the mechanism there The EDGE AI is analyzing micro vibrations and temperature anomalies in real time.

00:13:23: It's detecting that a critical bearing is going to fail days before it actually snaps.

00:13:29: Saving two thousand hours in heavy steel manufacturing isn't just an efficiency metric, it prevents massive process interruptions lowers emissions and pays enormous tangible

00:13:39: dividends.".

00:13:45: Speaking from the perspective of an operational technology chief information security officer issued a very sobering warning.

00:13:52: What did he say?

00:13:53: He pointed out that while edge computing and predictive AI offer these incredible benefits, The way the industry assesses OT cybersecurity is fundamentally flawed.

00:14:02: Organizations often treat it identically to IT security.

00:14:05: they run a network scan identify exposed assets And patch software vulnerabilities

00:14:10: which completely ignores the laws of physics governing a manufacturing plant

00:14:13: precisely.

00:14:15: And this ties back to the scenario I mentioned at the very beginning of our discussion.

00:14:19: Muhammad shared a story about watching process engineers keep a gas production plant operating manually for six hours.

00:14:25: All right!

00:14:26: Yeah, This was after losing all visibility their control systems due to failure.

00:14:31: Those human operators had understand every single fluid interaction Every pressure limit and cascading physical consequences.

00:14:40: That is a phenomenal reality check.

00:14:42: I mean, if a virtual PLC or an edge server goes down due to a cyber attack the instruction to switch to manual on an incident response plan Is not just in administrative checkbox.

00:14:52: No Not at all.

00:14:53: It requires profound deeply specialized level of operator judgment.

00:14:58: Exactly no AI dashboard.

00:15:00: our predictive algorithm can replace that human intuition.

00:15:03: In crisis when we evaluate cybersecurity and manufacturing context We cant look data loss evaluate the severe physical impact of losing a system.

00:15:12: We have to brutally assess whether human workforce on the floor actually retains knowledge and runs the plant safely without automated systems holding their

00:15:20: hands.".

00:15:22: The ultimate culmination of all these elements is physical AI.

00:15:42: This is the frontier where artificial intelligence actually inhabits a physical body and manipulates the real

00:15:48: world.".

00:15:49: Yes, we are witnessing a monumental strategic shift here.

00:15:53: for the past few years... ...the focus has been on AI that analyzes vast datasets or generates text in images but now we're seeing the emergence of foundation models specifically fine-tuned.

00:16:06: So what does this all mean in a practical industrial setting?

00:16:09: Victor Martinez provided some excellent analysis on Hyundai's strategy with Boston Dynamics, and he drew a direct comparison to Tesla as approach.

00:16:17: You know, Tesla didn't just design an electric vehicle.

00:16:19: they engineered a closed loop fleet learning platform.

00:16:22: Right let's break down that mechanism.

00:16:24: Every time A single tesla encounters new obstacle or edge case On the road It gathers data That updates The central model Better at autonomous driving overnight.

00:16:37: Exactly, and Hyundai is applying that exact same closed-loop logic to industrial robotics.

00:16:44: They are not focused on programming one robot To do one specific task in a cage right.

00:16:49: they are deploying these systems into real variable operations learning from every physical interaction And transferring that learned experience across their entire robotic fleet.

00:17:00: now the physical hardware like battery density actuator strength Is still limiting factor today?

00:17:05: But as the hardware matures, The Ultimate Competitive Advantage won't belong to the company with strongest Roba

00:17:10: arm.

00:17:10: No it wont!

00:17:11: It will belongs to an organization that owns a platform capable of learning fastest from physical world.

00:17:16: If we connect this to bigger picture you can start see how Physical AI will completely reshape entire layout.

00:17:25: Frank Brothan discussed this shift in the context of Siemens collaborating with NVIDIA to integrate advanced AI into humanoid robots, and he made a crucial distinction that honestly often gets lost.

00:17:37: What's that?

00:17:38: Humanoid robots are not being developed to replace highly specialized dedicated production equipment.

00:17:43: Right, I mean a humanoid robot is never going to outpace the dedicated CNC machine milling aluminum or high-speed bottling line packaging beverages.

00:17:51: Exactly!

00:17:52: Specialized machines will always dominate in speed precision and efficiency for rigid, high volume tasks.

00:17:59: However there's massive operational gap in modern manufacturing.

00:18:03: There are countless tasks that're too complex Too variable Or just require too much adaptability For traditional hard coded automation.

00:18:10: And thats where the humanoids step.

00:18:12: That is the exact gap where humanoids powered by physical AI will be deployed.

00:18:18: They're designed to complement the existing production lines, they step in and handle the flexible human-like dexterity required when a customized product run changes

00:18:27: over.".

00:18:27: And we have to be brutally honest about why this specific technology so critical right now.

00:18:33: Developing physical AI isn't just chasing the next cool technology trend.

00:18:38: it's an absolute survival mechanism for the industry.

00:18:42: Yes, I was reading a staggering statistic shared by Dean Bartels regarding the defense industrial base workforce in The United States.

00:18:49: he noted that over the past few decades That specific manufacturing work force dropped from roughly thirty million people down to approximately twelve million.

00:18:57: Wow

00:18:57: an eighteen million person deficit there represents a catastrophic loss of institutional knowledge and just raw physical manpower.

00:19:04: it really does.

00:19:05: Yet, despite losing more than half of that workforce overall industrial productivity continued to increase over the same period.

00:19:11: The only mechanism makes this mathematical reality possible is automation.

00:19:16: Physical AI and adaptable robotics are no longer just efficiency plays to improve margins.

00:19:22: We're facing a massive demographic cliff in severe systemic work force shortages across global manufacturing.

00:19:29: If we do not develop robotic systems that can dynamically learn and adapt to unstructured physical environments, We literally will NOT have the human hands required To build the infrastructure society depends on.

00:19:42: It truly is The ultimate enabler for future production.

00:19:45: You know the industry spent the entirety of last decade Connecting machines just a sort-of sense.

00:19:49: Just understand what was happening in the physical world.

00:19:52: Next decade Will be entirely defined by Machines learning to act autonomously within it which leaves us with a critical strategic question for anyone listening today.

00:20:00: Go ahead!

00:20:01: As the industry builds AI that learns from fleets of physical robots, what foundational data and engineering capabilities does your organization need to start building today so you aren't left hopelessly behind when learning directly from the physical world becomes the ultimate industrial advantage?

00:20:17: If you enjoyed this episode new episodes drop every two weeks.

00:20:20: Also check out our other editions on Digital Construction and Connected Tools & Equipment.

00:20:25: Thank you so much for joining us in this deep dive into the realities of smart manufacturing!

00:20:29: Remember to subscribe, keep connecting that data – we will catch ya next time!