Best of LinkedIn: Smart Manufacturing CW 27/ 28

00:00:00: This episode is provided by Thomas Allgeier and Frennus, based on the most relevant LinkedIn posts about smart manufacturing in calendar weeks twenty-seven and twenty eight.

00:00:10: Frenness 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: Right now There's A Factory in Europe purposefully making catastrophic manufacturing errors every single day.

00:00:32: Wait, on purpose?

00:00:33: Completely on-purpose!

00:00:34: I mean they're overloading machines messing up material flows and just causing absolute chaos on the floor... And it's actually saving them millions of dollars.

00:00:41: Okay that sounds insane but i know there is a

00:00:43: catch.

00:00:44: There is..and we are going to explain exactly how they do that without immediately going out into business in few minutes.

00:00:50: But our real goal for you today if your operating anywhere in the manufacturing or construction space Is cut straight through the noise.

00:00:58: We're looking at the absolute top trends and actionable insights shared across the industry by leaders on the ground over

00:01:17: hard, profitable execution on the shop floor.

00:01:20: Absolutely!

00:01:20: But as we're going to see in this deep dive you can't just buy a shiny new algorithm and expect it to fix a broken factory... The technology is only good at what its foundation built-on

00:01:30: which is why kind of have start with serious reality check right before even talk about autonomous agents or humanoid robot Which

00:01:36: will get too We'll.

00:01:38: but first look at incredibly unglamorous realty.

00:01:44: I was reading a post from Jacobo Lurek Kassal recently, and he pointed out a really uncomfortable truth.

00:01:50: Oh about his factory visits?

00:01:52: Yeah He has personally walked over fifty different factory floors across Europe over the last few months And he noticed this terrifying trend.

00:02:01: You have manufacturers out there signing multi-million dollar contracts for AI solutions while their OEE Their overall equipment effectiveness is sitting at a miserable fifty-eight percent.

00:02:14: Wow,

00:02:15: which means they're essentially running at almost half capacity like their machines are either down running slowly or just producing scrap almost half the time

00:02:24: exactly.

00:02:24: and Chacobo makes this brilliant argument when a plant manager sees their energy costs suddenly spiked by forty percent Or they realize that can't scale production because they cant find qualified staff They immediately assume they have technology problem.

00:02:37: So they go buy software?

00:02:39: Yep But they don't have a technology problem.

00:02:40: They have visibility and process problems,

00:02:42: right?

00:02:43: Because if you don't actually know which specific line is consuming that energy

00:02:47: exactly

00:02:48: or your workflows constantly require Your senior experts to intervene because absolutely nothing was ever documented.

00:02:54: an AI pilot Is not going to save.

00:02:56: it's just

00:02:56: not

00:02:57: buying AI to fix A basic visibility problem.

00:03:02: I mean, it's like buying a faster car when you don't even have a map.

00:03:05: That is the perfect analogy!

00:03:06: You're not going to reach your destination any fast or are just gonna get lost at much higher speed... ...you're buying this solution before you've bothered to map out

00:03:13: that problem.

00:03:13: Yeah and Britton Roberts made very similar point recently about information flow.

00:03:18: He noted that most of the AI conversations happening in manufacturing boardrooms right now, they start by arguing over which language model to use.

00:03:27: Oh!

00:03:27: That's entirely the wrong place to start?

00:03:29: It

00:03:29: is...AI does not magically fix broken information flow if your data is trapped in manual clipboard reporting or

00:03:36: locked inside those disconnected proprietary legacy systems

00:03:40: Exactly The most advanced algorithm in world totally useless there.

00:03:45: AI only becomes valuable when the right information reaches the right people or machines fast enough to actually alter an outcome.

00:03:53: Okay, I have to push back a little here Or at least ask for some mechanical clarity.

00:03:57: everyone in this industry knows that data silos are bad?

00:04:01: We've been talking about breaking down data silos since the nineteen nineties But what does fixing that flow actually look like today?

00:04:07: That's the multi-million dollar question, right?

00:04:09: Right.

00:04:10: And it is where we bring in some insights shared by Connie Lumentit and Shankar Rahman.

00:04:15: They argue that if you want an AI ready factory today You need a data infrastructure built on two things.

00:04:21: Okay what are they?

00:04:22: A data ontology and knowledge graphs all anchored to a PLM backbone.

00:04:27: Wait ontology You're losing me.

00:04:29: That sounds like something I slept through at a college philosophy class, not something you find on a greasy factory floor!

00:04:34: It does sound super academic but physically it's very practical.

00:04:37: A data ontology in manufacturing is just a strictly defined structure of what your data actually means In the business context.

00:04:45: Right now without an ontology, a factory's data Is just a bunch of scattered spreadsheets.

00:04:50: You have financial data in your ERP system Shop floor data and your MES And defect data into separate quality systems.

00:04:58: Right, and they don't speak to each other.

00:04:59: So there are essentially speaking completely different languages.

00:05:02: A part number in one system might be called something completely different than another Precisely.

00:05:07: But if you use a knowledge graph And Connie actually uses the Siemens Rapid Minor Graph Studio as an example of this You physically connect those data points conceptually.

00:05:16: Oh,

00:05:17: I see!

00:05:17: It takes a raw string of numbers and defines it like supplier A provides raw material X which is used in chemical formula one-twenty three Which is actively being produced on millimachine five.

00:05:29: Ah

00:05:29: okay so it gives the AI the ability to actually Reason across relationships rather than just crunching a flat spreadsheet.

00:05:37: Yes, exactly.

00:05:38: So if Millie machine five suddenly starts producing out of tolerance parts the system understands instantly that The root cause might trace all the way back to I don't know humidity change from supplier A's warehouse.

00:05:50: you nailed it.

00:05:50: and Critical piece that Shankaroman points out is the PLM backbone product lifecycle management.

00:05:57: That's your digital spine.

00:05:58: It holds the absolute single source of truth for engineering design.

00:06:02: Got it.

00:06:02: So when you're knowledge graph sits on top of that governed PLM back bone.

00:06:07: The AI finally has a trustworthy consistent foundation to actually analyze.

00:06:11: And once you have that foundation, and once systems are speaking the same language You can do some incredible things without even installing physical hardware.

00:06:22: Oh,

00:06:22: absolutely!

00:06:23: I was looking at a post by Sean Dick Barman and he shared this brilliant practical example of this called soft sensors.

00:06:30: Ah yeah, soft sensors are mathematically fascinating...

00:06:34: Right because in heavy industry Physically measuring a process variable can sometimes be impossible.

00:06:39: It might too hot or to corrosive, or too remote To put physical thermometer and gauge in there.

00:06:45: Right.

00:06:45: so instead you use soft sensor to mathematically estimate what is happening inside that environment Using the signals your plant has already generating from outside.

00:06:54: Yeah And Shenan Dip used an example of cement kiln which was perfect illustration.

00:06:59: In cement manufacturing, one of the most critical quality parameters is something called clinker-free lime.

00:07:04: Right

00:07:04: and if that level is off your cement is basically useless?

00:07:08: Exactly!

00:07:09: But to measure it traditionally a human operator has to physically extract a sample from this massive rotating kiln send it to a laboratory run chemical tests And they get the results maybe four hours later...

00:07:23: ...and in manufacturing A four hour delay Is an eternity.

00:07:28: Every single hour between a quality deviation happening inside that kiln and you actually detecting it in the lab Costs, you massive amounts of fuel

00:07:37: yeah And it ruins.

00:07:38: The kilns thermal stability

00:07:40: and obliterates your profit margin.

00:07:41: because getting a lab result four hours late isn't actionable intelligence.

00:07:45: It's just a history lesson

00:07:46: exactly.

00:07:47: But a soft sensor fixes this.

00:07:49: To use an analogy, it's kind of like trying to guess the exact internal temperature of an oven but you don't have a meat thermometer right?

00:07:55: Instead You build a mathematical model based on watching how fast The cheese is melting through the oven window combined with How much gas the oven Is pulling.

00:08:02: that's A great way to put It.

00:08:04: so the soft Sensor takes historical process data and combines it With live operating signals Like the torque Of the kiln motor or the exhaust Gas Temperature to predict That clinker quality in near real time.

00:08:16: It

00:08:17: literally predicts the lab result before the lab even runs

00:08:40: data

00:08:42: and using it to predict an act is the exact bridge To our next major shift, right?

00:08:51: Because once your visibility's fixed AI stops being just a passive dashboard giving you a recommendation It graduates.

00:08:58: It starts taking autonomous action on the floor.

00:09:00: See, this is where things get slightly intimidating to be honest.

00:09:03: Andros Varro was at the Automate twenty-twenty six conference recently and he noted that.

00:09:15: Yeah, he highlighted the Siemens eigenengineering agent.

00:09:17: Right?

00:09:17: He

00:09:18: did.

00:09:18: and what is crucial here Is that this wasn't just a chatbot writing a polite email to a supplier.

00:09:24: The Eigen engineering agent was actively executing PLC coding And configuring HMI visualization.

00:09:30: Let's define those quickly.

00:09:31: for anyone not deep in automation.

00:09:33: yeah the plc the programmable logic controller is essentially the rugged physical brain attached to an industrial machine that tells it exactly when to fire piston or spin a motor.

00:09:43: Right, and the HMI is the human-machine interface like touch screen operator uses.

00:09:48: getting in AI right code incredibly complex

00:09:51: extremely because requires adhering to strict industrial safety standards.

00:09:55: In past writing PLC logic took days of manual programming by highly specialized engineer

00:10:00: And on dress saw this agent doing autonomously ensuring correctness reliability

00:10:05: And it isn't just a flashy trade show demo anymore.

00:10:08: Christiane Ribeiro posted some insights from the Realize Live Europe event, specifically looking at Rolls Royce.

00:10:15: Oh yeah!

00:10:15: The AI copilot.

00:10:17: Yeah.

00:10:17: they deployed an AI-powered production co-pilot and they managed to cut their manufacturing programming time by up to eighty percent.

00:10:25: That's

00:10:25: wild!

00:10:26: Cutting engineering time by eighty percent at a company like Rolls-Royce is a staggering financial win,

00:10:31: it really is.

00:10:32: which brings us the core business question how do we actually justify these massive AI investments to a board of directors?

00:10:40: because we aren't just buying new conveyor belt with unknown output We are buying digital intelligence.

00:10:46: that such good point and Arthi Sariman shared fantastic framework for exactly this problem.

00:10:51: She argues that manufacturing leaders have to stop treating AI like an IT science project.

00:10:56: Okay,

00:10:56: so how should they treat it?

00:10:57: You have to evaluate using the exact same ruthless financial discipline you use when buying a multi-million dollar CNC machine.

00:11:06: instead of asking in the Ikea department which trendy A.I tool we buy...you go on the floor and ask which operational decisions occur hundreds times per day cost us in labor and delay.

00:11:22: Oh, so we're optimizing the costs per operational decision?

00:11:25: I love that!

00:11:25: Right think about the daily life of a production planner.

00:11:29: they might make two hundred fifty scheduling decisions every single day rerouting apart because the machine went down adjusting a batch size because supplier was late.

00:11:37: right let's say each decision takes them to minutes.

00:11:40: digging through spreadsheets.

00:11:42: If an AI co-pilot reduces that time from two minutes to thirty seconds per decision while actually producing a more mathematically optimized schedule, you aren't just saving the planner's time.

00:11:52: You're increasing factory throughput.

00:11:54: Exactly!

00:11:54: You are reducing raw inventory holding costs and recovering massive amounts of human capacity.

00:12:00: But...and

00:12:01: this is a massive factory size.

00:12:03: but we have to be incredibly careful here.

00:12:05: The closer AI gets to physical floor The higher the stakes.

00:12:09: That's silly.

00:12:10: Antoine Bissom, founder of POCA put out a stark necessary warning about this.

00:12:14: he said and I quote when an industrial AI model hallucinates a torque spec or lockout procedure

00:12:30: It

00:12:30: really is.

00:12:31: If a marketing department uses AI to write a blog post and the AI hallucinates, you get a funny typo!

00:12:37: If an industrial AI hallucinate's the torque parameter on aerospace turbine assembly or it gets safety lockout procedure wrong at high voltage stamping press...

00:12:45: You got catastrophic machine jailers?

00:12:47: Yes millions in scrap parts.

00:12:48: OR YOU GET A HUMAN BEING SERIOUSLY INJURED.

00:12:50: So if we want massive efficiency gains of AI making decisions & writing code but absolutely cannot afford for learning by making trial-and-error mistakes physical machines, how do we bridge that gap?

00:13:03: We make the mistakes in a perfect virtual world first.

00:13:28: digital lighthouse and it is entirely built around the concept of The Digital Twin.

00:13:33: Right,

00:13:33: And Matthias pointed out that they simulate absolutely everything before a single wrench Is turned in reality?

00:13:39: Everything!

00:13:40: Let's say They want to reroute their autonomous material carts or they wanna change the sequence Of an assembly line Before they touch the physical floor...they run It into the digital twin.

00:13:49: they intentionally push the system to the breaking point in the simulation, and see where the bottlenecks happen.

00:13:54: Exactly!

00:13:55: They only alter the physical factory when computer mathematically proves it will work... ...and mechanical results are wild.

00:14:01: Yeah you'd think a digital twin is just nice three-D map but used them to simulate thermodynamics of their factory ventilation

00:14:08: which is brilliant.

00:14:09: Instead

00:14:09: of just blasting air conditioning everywhere based on a thermostat, they map the exact heat signatures of machines running at peak load versus idle.

00:14:18: By simulating their airflow digitally They optimize ventilation and drop entire factory's energy demand by seventy percent.

00:14:25: Seventy percent And by simulating routing of autonomous guided vehicles reduced unnecessary material movement by forty percent.

00:14:33: This isn't a pilot program, this is just how they operate.

00:14:36: every single day.

00:14:37: We're seeing the same virtual first approach trickling all the way down to the individual machine level too.

00:14:43: Dereja Sankarpal highlighted company called Willis Custom Yachts.

00:14:47: Oh I saw that.

00:14:47: Yeah They build these massive custom sport fishing yachts with incredibly complex flowing geometries!

00:14:55: They use the digital twin inside Siemens and Xcam To simulate full CNC machining environment.

00:15:01: And for those unfamiliar, CNC machining is subtractive manufacturing taking a giant block of metal and using spinning cutting tools to carve out a precision part.

00:15:09: It's incredibly easy to accidentally crash the tool into the metal if your code is wrong which destroys a fifty thousand dollars.

00:15:18: spindle in a fraction second Exactly!

00:15:20: So Willis Custom Yachts creates complete digital replica.

00:15:23: The exact cutting tools, the exact holding fixtures...the physical limits of machines travel.

00:15:29: So smart

00:15:30: By validating every single toolpath continuously in the virtual world before the file ever reaches The shop floor.

00:15:37: their programmer reported cutting setup times and programming In some areas by fifty percent.

00:15:41: Wow,

00:15:42: and Ross Kiesel echoed this exact workflow as well.

00:15:45: He talked about using run my Virtual Machine software with heller machine tools.

00:15:50: They are literally training operators and managing complex engineering changes on the digital twin Before the physical machine ever cuts a single piece of raw material.

00:15:58: And when you trust that virtual physics engine, You can push the machines much harder.

00:16:02: Erdem Osterk shared a jaw-dropping case study about an aerospace manufacturer using SenseNC's CAM optimization software.

00:16:09: Where they do?

00:16:10: Typically A human programmer guesses a safe conservative cutting speed so The tool doesn't break.

00:16:16: But SenseNC uses physics-based algorithms to calculate exactly how much force the cutting tool can take at every millisecond.

00:16:23: It speeds up the tool in straight lines and slows it down intelligently, tight corners... So

00:16:28: they aren't relying on trial or error?

00:16:30: They're letting cure physics dictate feed and speed rates inside the CAM environment!

00:16:35: Right…and by doing that saved over a thousand machining hours of just two aerospace parts over six months period.

00:16:43: that is insane.

00:16:44: That's

00:16:45: essentially unlocking forty two days of extra free machining capacity.

00:16:49: and they didn't have to buy a single new physical machine, They just used the digital twin to extract the maximum physical capability

00:17:01: fix the data flow with ontologies, we've allowed AI to start making decisions.

00:17:05: We used digital twins to validate those decision safely right?

00:17:09: The final step is pushing that validated optimized intelligence back out into the physical world not just as code but his physical motion.

00:17:16: Okay, so we are talking about the rise of physical AI.

00:17:20: Louis Feinstein posted a really fascinating observation... ...about how robotics space is rapidly converging.

00:17:27: For the last ten years Robotics has been running on two separate parallel tracks.

00:17:32: One side you have autonomous mobile robots or AMRs basically smart wheeled platforms that have mastered navigating busy warehouses and logistics.

00:17:42: Right On the other side, you have stationary humanoid arms that have mastered dexterous fine manipulation.

00:17:49: And now those two worlds are crashing into eachother.

00:17:53: We're finally building systems capable of both mobility and complex physical work.

00:17:58: But The math required to do this is staggering.

00:18:00: I can imagine

00:18:01: An AMR on wheels Can just hit the brakes if someone walks in front it.

00:18:04: A bipedal humanoid robot carrying a thirty-pound box has to constantly calculate its dynamic center of gravity, adjusts it's force distribution and figure out how step over an air hose on the chaotic factory floor without falling.

00:18:15: It requires millions of micro decisions per

00:18:18: second.".

00:18:18: Which is exactly why everything we just discussed about modeling in the digital twin first...is so vital!

00:18:25: Michael Walker highlighted how Siemens Process Simulate software is already allowing manufacturers to drop digital humanoids into their virtual factories.

00:18:33: Right, so you can test it all out?

00:18:35: Yeah!

00:18:35: You can test the cycle times evaluate the ergonomics and make sure that humanoid work safely alongside traditional human workers before a physical robot ever rolls off of its delivery truck.

00:18:45: And they are arriving much faster than people realize.

00:18:49: Tobias Claus shared an update on the UB TechWalker S-II Robot which has being deployed right now to take over highly repetitive, physically demanding tasks in manufacturing environments.

00:19:01: Wow!

00:19:02: It's no longer a question of if humanoids will join the workforce

00:19:05: but how quickly?

00:19:06: But doesn't this bring us back to Anton Besson warning about physical blast radius?

00:19:10: I mean humanoid robot.

00:19:12: hallucinating movement while holding heavy tool next to a worker is terrifying thought

00:19:17: it And Kalmos wrote an incredibly insightful piece about this exact challenge.

00:19:23: He noted that the next phase of physical AI isn't about giving robot better joint motors or a slicker chassis, it is entirely about assurance.

00:19:31: Assurance?

00:19:32: Meaning mathematically proving the robot won't hurt anyone and break anything!

00:19:36: Yes...

00:19:37: The industry has moved past demo phases.

00:19:39: The question no longer can this robot pick up apart in pristine controlled lab environment.

00:19:45: The question is, can we trust this system to dynamically adapt to variations and stay within strict safety boundaries on a live messy production line?

00:19:54: Right.

00:19:54: As Calvin brilliantly points out A robot that achieves a ninety-five percent success rate in the university lab sounds great on paper But if you put a ninety five percent accurate robot on high speed production line It's failing fifty times a day

00:20:07: And definitely cannot run profitable business.

00:20:09: stop in line fifty times.

00:20:12: Heavy industry requires ninety-nine point nine percent reliability.

00:20:15: Bridging that final gap, earning that absolute operational trust is the real competitive moat for physical AI right now.

00:20:22: So if we step back and look at this entire journey We've mapped out We have gone from the incredibly unglamorous foundational work of fixing messy data silos with PLM backbones and ontologies.

00:20:37: Yeah, we moved to AI agents actually writing the PLC logic for machines?

00:20:42: We looked at how digital twins validate that logic so we don't blow up the factory.

00:20:46: And finally we arrive at humanoid robots physically executing the work.

00:20:51: It is a complete, closed-loop autonomous

00:20:54: ecosystem."

00:20:55: It's

00:20:55: full realization of The Smart Factory.

00:20:58: Which leaves us with highly provocative thought for you to mull over this week.

00:21:02: If we are truly moving toward this Autonomous Reality where AI agents write the engineering code digital twins simulate and validate physical safety of it And humanoid robots physically execute labor on floor What exactly is new role in Human Engineer?

00:21:16: Are you actively preparing your career right now to shift away from being an operator of physical machines and stepping up to become an orchestrator of digital logic?

00:21:24: That might just be the defining question for the next decade of manufacturing.

00:21:28: Thank you so much for joining us as we explored these incredible insights from professionals building this future on-the-ground!

00:21:34: If you enjoyed this episode, new episodes drop every two weeks.

00:21:38: Also check out our other editions on Digital Construction & Digital Power Tools.