Best of LinkedIn: Smart Manufacturing & Industrial AI CW 39/ 40

00:00:00: Provided by Thomas Allgeier and Freeness, based on the most relevant LinkedIn posts about smart manufacturing and industrial AI in calendar weeks thirty-nine and forty.

00:00:09: Freenes 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.

00:00:20: You can find more info In the description.

00:00:22: So um Imagine snapping a grainy photo of luxury watch With your smartphone Uploading it To an AI And like two minutes later receiving a fully structured, editable three-dimensional engineering file.

00:00:34: Right no calipers?

00:00:35: Yeah!

00:00:35: No Calipers...no days of manual reverse engineering just perfect digital model ready for the factory.

00:00:40: and I mean that is no longer science fiction.

00:00:42: it literally happened last week

00:00:44: It did And its completely redefining how we think about production.

00:00:49: So today are extracting top smart manufacturing and industrial AI trends dominating professional conversations across LinkedIn Exactly bypassing all the hype and looking exclusively at what is actually happening on the factory floor right now.

00:01:04: If you are working in operations or engineering or strategy this deep dive Is really about the tools?

00:01:10: And a massive capital investments that are reshaping your daily reality.

00:01:13: Yeah, and we're starting with a massive operational shift because for years You know manufacturing has treated AI as sort of fun science fair project

00:01:21: like a pilot program here there Right

00:01:23: predictive maintenance dashboard.

00:01:25: Here little test their.

00:01:27: But the patience for that experimentation is just gone.

00:01:31: Plant managers are demanding measurable return on investment and extreme reliability.

00:01:37: Yeah, they won results yesterday

00:01:39: Exactly.

00:01:39: And to get that They're radically changing their timelines.

00:01:43: Arboree Hakaj shared some really fascinating insights about how Covestro was operating right now.

00:01:48: Oh

00:01:48: The chemical company.

00:01:49: Right!

00:01:49: They are completely tossing out Their traditional ten-year technology roadmaps.

00:01:53: Wow

00:01:54: Ten years is a long time in tech anyway.

00:01:56: It's

00:01:56: an eternity.

00:01:58: So instead they are executing highly agile six-month game plans

00:02:02: Which I mean, it was really the only logical response to the current pace of innovation.

00:02:07: if you lock your facility into a ten year blueprint for artificial intelligence You are Basically accumulating technical debt before the ink even dries.

00:02:15: Yeah that makes sense

00:02:16: Because by the time you finish a two-year procurement cycle For like a server cluster The foundational AI models have advanced three generations.

00:02:24: Right,

00:02:24: it moves so fast!

00:02:25: Exactly.

00:02:27: but what Covestro's doing internally is actually even more important than their timeline.

00:02:31: instead of just going out and hiring external data scientists they built an internal AI Academy.

00:02:36: Oh to train their own people?

00:02:38: Yeah upskill their own domain experts.

00:02:40: uh their chemical engineers and plant operators.

00:02:42: Well because you can't just parachute a Silicon Valley software developer into a chemical plant and expect them to optimize it right.

00:02:49: No, exactly because those are fundamentally different environments.

00:02:53: A software developer works in this you know deterministic digital world.

00:02:57: they code one plus one.

00:02:58: it always equals two

00:03:00: right

00:03:00: but a chemical plant is a non-deterministic physical world.

00:03:04: One Plus One might equal to But It Might Also Equal In Explosion.

00:03:08: Oh Wow Depending on ambient humidity or pipe pressure Or Trace impurities and a catalyst?

00:03:14: A Plant Specialist Already Understands That Physical Chaos.

00:03:17: Right, they live it every day.

00:03:19: Exactly!

00:03:20: So, Covestro realized that is much faster to teach a chemical engineer.

00:03:24: how do you use machine learning than its data scientist?

00:03:27: how fluid dynamics work in a pressurized reactor?

00:03:30: That makes total sense.

00:03:32: and you know the distinction between digital consequences and physical consequences perfectly tees up an analogy Del Costi shared.

00:03:41: I cannot stop thinking about

00:03:42: it.

00:03:42: Oh, the pizza one?

00:03:43: Yes!

00:03:44: He pointed out an instance where a user asked generic consumer AI what to do if cheese wasn't sticking into their pizza and the AI scraped an old joke from forum just confidently presented as factual solution.

00:03:57: It literally told that users add non-toxic glue to the pizza sauce

00:04:01: Which is...I mean its funny glitch on your smart phone.

00:04:04: Yeah, it's hilarious

00:04:05: But it was terrifying.

00:04:07: prospect of factory floor Coast's underlying point there is that generic AI lacks physical context.

00:04:14: In an industrial environment, hallucinations are simply not an option.

00:04:18: Not at all.

00:04:19: If an AI hallucinates a robot's toolpath by like three inches it doesn't just give you bad recipe It shears a hundred thousand dollars spindle off of CNC machine

00:04:29: Or injures the worker.

00:04:30: Exactly!

00:04:31: The stakes are physical.

00:04:32: So how do we prevent the industrial equivalent of glue on pizza?

00:04:36: Thomas Ruer argued in his post Probably right is just useless in manufacturing.

00:04:41: Yeah, probably doesn't cut

00:04:42: it.

00:04:42: he pointed out that to actually scale this technology You need rigid orchestration and strict safeguards That mathematically define where the limits of the AI?

00:04:50: Actually are

00:04:51: right.

00:04:52: but wait I am getting stuck on something here.

00:04:54: So we're giving AI strict guardrails like putting up bumpers at a bowling alley so the ball doesn't go in The gutter.

00:04:59: okay yeah But at what point do those safeguarts restrict the speed and agility We were trying to achieve with AI in the first place?

00:05:06: i know It feels Like A contradiction but it is actually the key to scaling.

00:05:11: It's less like putting brakes on a car and more like giving the AI, uh strictly defined sandbox.

00:05:16: A Sandbox?

00:05:17: Okay!

00:05:18: Yeah inside that sand box.

00:05:19: The AI has infinite freedom to optimize adapt and move at lightning speed But it is mathematically locked out of touching anything outside of it.

00:05:29: I see

00:05:29: If a plant manager knows the AI physically cannot bypass a safety interlock or ultra critical pressure valve They will finally trust it enough to let it run autonomously.

00:05:39: The sandbox is what actually allows this feed.

00:05:43: Okay, that makes sense.

00:05:44: And when you build that trust I mean the financial returns are staggering.

00:05:48: Reiner Brem shared some hard numbers from Siemens.

00:05:51: Oh yeah?

00:05:51: The PNG case

00:05:52: Yeah.

00:05:52: they're running AI visual inspection across Procter and Gamble plants globally.

00:05:57: We aren't talking about checking if like a simple metal bracket Is right shape Right.

00:06:01: They are inspecting high speed, highly variable textured consumer goods in real time.

00:06:07: And because they're running deep learning models directly on the industrial edge... ...they are cutting scrap by ten to twenty percent!

00:06:13: Which is huge and they are deploying it five-to-ten times faster than traditional vision systems.

00:06:18: Wait really?

00:06:20: Five-to ten

00:06:20: times?!

00:06:21: Yeah cause traditional system require a programmer define every single lighting condition and pixel threshold

00:06:27: Right manually

00:06:28: Exactly.

00:06:29: Edge AI simply learns what Texture looks like dynamically.

00:06:33: It adjusts to tiny variations in the material without triggering a false alarm.

00:06:38: Petra Dietrich highlighted his similar leap with TRUMPF's TeachLine AI on their new welding machines.

00:06:43: Okay, so this is for companies doing small highly customized batches where programming The Welding Robot actually takes longer than the actual manufacturing.

00:06:51: right?

00:06:51: Precisely!

00:06:52: When you weld metal... ...the intense heat causes thermal distortion.

00:06:56: Right

00:06:57: The metal literally warps and shifts while you are working on it.

00:07:00: Normally, a human has to stop measure the deviation and reprogram the robot path.

00:07:05: Which takes forever?

00:07:07: Truef's AI uses sensors to recognize those deviations in real time... ...and automatically adjusts the robots' paths against original CAD data while sparks are still flying!

00:07:17: Wow

00:07:18: It is cutting programming & setup times by upto fifty percent.

00:07:21: Cutting set-up time in half when skilled welders are nearly impossible to hire right now Is massive competitive advantage

00:07:29: Absolutely.

00:07:30: And speaking of advantages, before we look at the massive software ecosystem shifts making all this possible just a quick reminder if you are finding the steep dive helpful for navigating these rapid industry changes make sure to subscribe so that you don't miss future additions.

00:07:43: Yes definitely hit subscribe!

00:07:45: Okay so manufacturers get these massive ROI numbers but they aren't building those tools from scratch?

00:07:51: No definitely not.

00:07:53: The technological infrastructure required to run, you know high-speed edge AI or real time thermal compensation is incredibly complex.

00:08:02: What we are seeing right now Is a massive consolidation of the industrial tech stack?

00:08:07: The ecosystem is basically converging to support physical AI

00:08:12: Which brings us back to that mind.

00:08:13: bending engineering example from Eric Fong

00:08:16: the CAD GPT one.

00:08:18: Yeah he nicknamed it CAD GP T. When he fed photos of an AP Royal Oak watch into an AI and got a three D model back, it wasn't just a flat hollow shape.

00:08:27: Right?

00:08:28: It was a fully structured CAD tree.

00:08:31: right for those of us who aren't design engineers What does that actually mean?

00:08:34: That is the crucial distinction here.

00:08:37: A generic three d scan gives you a mesh which Is basically a digital lump Of clay.

00:08:42: okay

00:08:43: It looks like a watch but The computer doesn't know what.

00:08:47: A CAD tree is entirely different.

00:08:49: It's a parametric recipe.

00:08:51: Oh interesting!

00:08:52: The AI recognized the individual components, it knew that Bezel was separate piece of metal and mapped exact geometry of hexagonal screws And understood the curvature of crystal.

00:09:04: It figured all out from photo?

00:09:06: Yes.

00:09:07: It built distinct editable file for each part All linked together in an assembly.

00:09:12: You could click on screw lengthen by two millimeters if you wanted to.

00:09:17: So the photograph is literally, The new first sketch.

00:09:20: Exactly!

00:09:22: That fundamentally collapses the product design life cycle.

00:09:25: but to run something like that or To run those high-speed vision systems at Proctor and Gamble You need serious computing power.

00:09:32: exactly where the action Is

00:09:33: you do?

00:09:34: And Daniel Brand shared an update on the New industrial AI suite release that focuses specifically On this.

00:09:40: they made it possible to Do on premise AI model retraining entirely on the shop

00:09:44: floor.

00:09:45: This is a major breakthrough for edge computing.

00:09:47: Really?

00:09:48: How come?

00:09:49: Well,

00:09:49: normally if the machine's environment changes say A factory window opens and changes to lighting on a visual inspection line The AI model gets confused.

00:09:59: Oh because the shadows change Right.

00:10:01: Historically you had to package all that new image data Send it into massive server farm in cloud Wait for the model be retrained And then push back down onto the factory

00:10:10: which takes time, takes bandwidth and obviously exposes your data to the internet.

00:10:15: Exactly!

00:10:16: Brand's update means you have direct GPU access right there on the machine.

00:10:21: The AI retrains itself locally adapting to new lighting or material in minutes without ever leaving building

00:10:27: And providing that kind of seamless end-to-end infrastructure is driving billions of dollars in corporate acquisitions right now.

00:10:34: Huge

00:10:34: acquisitions?

00:10:35: Yeah,

00:10:35: Neil Barua announced that Schneider Electric is acquiring PTC and a twenty two point six billion dollar all cash transaction.

00:10:43: That is massive.

00:10:44: It really is.

00:10:45: they are taking Schneiders dominance in physical energy management and factory automation and fusing it with PTC's PLM product life cycle management software.

00:10:55: Right

00:10:55: MPLM is essentially the digital filing cabinet that tracks a product from its initial design all the way through it's manufacturing and service life.

00:11:03: It is a clear signal that owning just the physical hardware or just the digital software isn't enough to survive the next decade.

00:11:10: You need both.

00:11:10: you have to own the entire loop.

00:11:12: Dion Smith highlighted another major move reflecting this A global partnership between Siemens And TD Cinex.

00:11:20: This Is The Perfect Illustration of IT & OT Convergence.

00:11:23: So IT being information technology, the cloud.

00:11:26: The server racks that enterprise software and OT being operational technology-the actual physical robots PLCs in factory machines.

00:11:34: right

00:11:34: historically those two departments didn't even speak to each other.

00:11:37: oh they had completely different priorities.

00:11:39: right it prioritized data security and scale.

00:11:43: ot prioritized physical safety and machine uptime.

00:11:45: Right keep the line running

00:11:47: but physical AI forces them to merge.

00:11:50: A cloud-based AI model from the IT side is useless if it can't talk to a robotic arm on the OT side, to adjust a

00:11:55: weld.

00:11:56: Yeah

00:11:56: This partnership bridges those ecosystems bringing IT's cybersecurity and scale down to the physical factory floor.

00:12:03: But wait I have to challenge this.

00:12:04: It sounds like industrial automation is acting more like a smartphone ecosystem.

00:12:08: now The hardware is just a vessel in real powers In these interconnected software partnerships.

00:12:15: So, are we losing the bespoke custom nature of manufacturing or were finally just standardizing it so that actually scales?

00:12:22: I mean isn't this whole point a highly tuned factory unique to its specific product.

00:12:27: It is valid concern but standardizing foundation actually unlocks more customization at the product level.

00:12:33: Howso

00:12:34: Think about your smartphone?

00:12:36: The operating system is identical with billion other phones.

00:12:38: right

00:12:39: Yeah

00:12:40: but that standard framework allows millions of entirely unique, highly specialized apps to function perfectly on your device.

00:12:47: By standardizing the IT OT stack a factory can rapidly switch from making one custom product to another without ripping out the underlying networking cables and control architecture.

00:12:58: Oh okay That makes sense.

00:12:59: But if we are Standardizing global networks in plugging operational Factory machines directly into cloud it infrastructure aren't we just Opening the front door to massive cyber threats and data theft.

00:13:12: Well, yeah That's the big fear

00:13:14: which brings us to the most intense debate happening right now digital sovereignty And the geopolitics of data.

00:13:21: it is The unavoidable reality of hyper-connected supply chains.

00:13:25: dr.

00:13:25: Ferri Abba Hassan and Elka Andla at Deutsche Telekom are actually leading the charge on this in Europe.

00:13:30: Oh really?

00:13:30: Yeah their stance Is that as the continent scales AI Delivering data sovereignty is a non-negotiable prerequisite.

00:13:38: Okay, they're just talking about basic cybersecurity.

00:13:41: Sovereignty means absolute legal control over your critical AI operations and the freedom to switch technology providers without vendor lock in.

00:13:49: And they are building that infrastructure.

00:13:51: to prove it.

00:13:52: I read that in less than six months They launched the Munich industrial AI cloud.

00:13:56: That's incredibly fast.

00:13:57: yeah It is a localized AI factory designed specifically to give companies massive computing power while keeping their proprietary manufacturing data physically and legally within their control.

00:14:08: Because in the modern industrial landscape, if you do not control the data... You simply don't control the factory!

00:14:13: Exactly.

00:14:14: Roland Rosen provided brilliant technical look at how this was being standardized through the Factory X Initiative.

00:14:20: They just released three open source MX port implementations

00:14:24: Which are named Hercules, Orion & Leo right?

00:14:27: Yes These utilize technologies like EDC, OPC UA and asset administration shells to guarantee sovereign data exchange.

00:14:35: Okay let's slow down and translate that for anyone who isn't a systems architect.

00:14:39: good idea what exactly is an asset administration shell?

00:14:42: And why does it matter?

00:14:43: think of an asset Administration Shell as a secure digital backpack That follows the physical component around its entire life.

00:14:51: A digital back

00:14:52: pack.

00:14:53: Let say you manufacture a highly specialized pump.

00:14:56: The digital backpack holds its CAD drawings, it's maintenance history and real-time performance data.

00:15:02: Okay

00:15:03: Now when you sell that pump to a chemical plant You need their machines to communicate with it.

00:15:08: OPC UA access the universal translator so the pump in the plant can speak the same language.

00:15:13: Got It!

00:15:14: The MXport implementations act as secure checkpoints.

00:15:18: They ensure that the chemical plant can read the maintenance data but absolutely cannot access proprietary engineering designs.

00:15:25: So it cryptographically enforces who gets to see what, meaning you can plug your machine into a global supply chain without giving away your intellectual property.

00:15:33: Exactly

00:15:35: And when you zoom out the national level The race to secure these capabilities is staggering.

00:15:40: Myer Agarai highlighted India's ISM-II.O program.

00:15:44: Oh yes!

00:15:45: India is pouring thirteen point five billion dollars Into building an end-to-end semiconductor manufacturing ecosystem on their own soil

00:15:53: Because they recognize that silicon is the new steel.

00:15:56: They are training a hundred thousand specialized technicians and driving automated wafer fabrication to ensure they aren't reliant on foreign supply chains for the chips that run these AI factories.

00:16:06: Meanwhile, Ude Senapati spent week in China meeting with their automotive ecosystem.

00:16:11: made it really profound observation.

00:16:13: What did he say?

00:16:14: He noted that China's auto industry isn't just responding to global manufacturing trends anymore through incredible speed actively writing the next chapter of the global industry.

00:16:25: They are turning technology into physical products at a velocity that is forcing the rest of world to entirely rethink what's

00:16:32: possible.".

00:16:33: So we have Europe building sovereign data fortresses, India spending billions to localize chip production and China accelerating hardware to unprecedented speeds.

00:16:44: But I see a glaring contradiction here.

00:16:46: We are treating industrial data the way nations treat their power grids or water supplies, locking it down behind borders.

00:16:52: Right!

00:16:53: But if we hyper-localize and silo all this data for security don't kill global efficiency in speed we just praised?

00:17:01: That is that exact height rope industry is walking right now.

00:17:05: True sovereignty about control not isolation.

00:17:10: The goal of those digital backpacks and secure checkpoints we discussed with FactoryX is to allow global collaboration without compromising security.

00:17:19: You can share the recipe with a partner, but there's no denying that geo-political walls around critical technology are getting taller!

00:17:30: So to see how all these high-level geopolitical strategies and converging software ecosystems actually look, when the metal hits the floor we have to look at the trade shows.

00:17:38: Yes!

00:17:38: The true test

00:17:40: And IMTS twenty twenty six just wrapped up... ...and it served as a massive reality check.

00:17:45: Theory is great But does the software actually run the machines?

00:17:48: And the resounding answer this year was yes.

00:17:50: Jasper Wildebore assured a fantastic breakdown of Siemens demonstrating live end-to-end digital threat.

00:17:57: Oh I saw that post.

00:17:58: Yeah

00:17:58: They didn't just show a bunch of disconnected software demos, they took real aerospace jet engine component and pushed it through single continuous workflow.

00:18:07: Right on the floor?

00:18:08: Exactly!

00:18:09: The AI optimized initial CAD design seamlessly handed off to generate cam programming guided robotic handling arms and controlled actual CNC machining.

00:18:19: It was one intelligent unbroken chain from pixels to physical titanium.

00:18:23: It's incredible But making that physical chain work requires a massive, hidden architecture.

00:18:30: Bryce Myers made really astute observation while walking the floor at IMTS.

00:18:34: What did he notice?

00:18:35: Well...he was surrounded by physical AI humanoid robots Collaborative arms Autonomous drones Navigating the aisles.

00:18:41: Right The flashy stuff!

00:18:43: but pointed out that intelligence driving these machines doesn't live inside hardware.

00:18:48: Wait A robotic dog isn't doing advanced spatial calculus in its own head?

00:18:52: No

00:18:52: exactly.

00:18:53: Meyers noted that all of this physical AI relies completely on massive cloud infrastructure.

00:18:58: The heavy lifting, training the neural networks on millions of simulations happens in hyperscale cloud data setters.

00:19:05: Oh okay!

00:19:06: Then the resulting intelligence is compressed and pushed down to the edge... ...to the robot's onboard processor to actually execute movements.

00:19:15: The physical robots are incredible feats for engineering but they're entirely tethered with Cloud architecture.

00:19:20: That makes a lot sense.

00:19:22: Amidst all the talk of cloud infrastructure, data sovereignty and titanium.

00:19:28: Demetrius Beliopoulos brought the focus back to something we consistently overlook... The human element!

00:19:34: The people running it all?

00:19:35: Yeah He shared a story about finally meeting a longtime LinkedIn connection Mike Nager in person at IMTS.

00:19:42: It is always validating when those digital professional networks translate into real-world collaboration.

00:19:48: Absolutely.

00:19:49: And what Mike's doing is vital.

00:19:51: He writes children's books, like one titled All About Smart Manufacturing explicitly designed to rebrand factories for the next generation.

00:19:59: That's so cool!

00:20:00: Right?

00:20:01: he is actively trying to teach kids that modern factories are not The Dark Dank in dangerous places from history books.

00:20:07: They're ultra clean high-tech hubs driven by AI, three D printing and robotics.

00:20:13: And you know that perception shift isn't just nice PR it Is an existential necessity.

00:20:19: We are currently facing a massive demographic cliff as the whole generation of experienced operators retires.

00:20:25: Which begs the question, we focus so much on the silicon and software but Mike's children books point out a massive looming gap?

00:20:33: Yeah Who is going to run these AI factories?

00:20:36: Are we innovating machines faster than we're inspiring next generations human operators?

00:20:41: Because what good an edge-retrained sovereign AI cloud if you can't hire anyone who wants walk into building.

00:20:47: That is the silent race happening right now.

00:20:50: The technology's moving fast because it actually has to fill that demographic void.

00:20:55: I hadn't thought of it like that!

00:20:57: The strategy, by using AI to make industrial software interface more like a consumer smartphone we dramatically lower the barrier-to-entry.

00:21:05: We need the technology be advanced enough so that a twenty two year old operator can step onto floor and be effective on day one actively mentored by machines.

00:21:14: AI.

00:21:15: Wow So the AI isn't just turning the wrench, it's teaching the worker.

00:21:20: Exactly!

00:21:21: That is a profound shift.

00:21:22: I mean if we look back at the ground we've covered today The evolution is just stunning.

00:21:28: We started with companies like Covestro throwing out ten-year plans for agile AI integration

00:21:33: Recognizing that generic AI Is great writing emails But will literally suggest gluing your pizza together in the physical world.

00:21:42: Right, which requires those strict sandboxes to protect production.

00:21:46: Then we saw the ecosystem consolidate to support that speed.

00:21:49: editable CAD models from a single photograph localized model retraining on the edge and The twenty two point six billion dollar merger bridging the historical divide between IT and OT

00:22:01: all of Which slammed right into the geopolitical reality?

00:22:03: That industrial data is the new critical infrastructure driving Europe's sovereign clouds and India's multi-billion dollar semiconductor push.

00:22:12: And it all culminated at IMTS, proving the digital threat is real provided we can inspire the next generation to actually run in...

00:22:20: Which brings up a fascinating —and frankly very challenging— paradox to leave you with.

00:22:25: Oh!

00:22:25: What is that?

00:22:26: Dr Eva Reasonhooper highlighted critical finding.

00:22:30: Right now, nearly ninety percent of all industrial AI activity is focused on optimizing operations within a single organization.

00:22:39: A company making its own proprietary line run faster

00:22:42: Which makes total sense.

00:22:43: that's where the immediate financial return for them.

00:22:46: True Yet The biggest existential sustainability challenges we face optimizing national energy grids, managing circular material flows and decarbonizing global supply chains require massive transparent collaboration across entire value chains.

00:23:04: Yeah you can't do that alone!

00:23:06: So the question you have to ask yourself is this if every single manufacturer is building highly sovereign cryptographically lockdown AI ecosystems to fiercely protect their proprietary data how will we ever achieve cross-industry collaboration required to actually save the planet.

00:23:25: Yeah, we started by saying AI needs strict sandboxes so it doesn't cause catastrophes on the factory floor but if we build the walls of those sandboxes too high We might just lock ourselves out at the global solutions that desperately need.

00:23:38: Exactly Well that is definitely something to think about.

00:23:40: Thank you for joining us on this deep dive.

00:23:42: See ya next time.