Best of LinkedIn: Smart Manufacturing CW 15/ 16

00:00:00: This episode is provided by Thomas Allgaier and Frennus based on the most relevant LinkedIn posts about smart manufacturing in calendar weeks, fifteen and sixteen.

00:00:09: Frenness is a BDB 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.

00:00:19: You can find more info In the description.

00:00:25: So Welcome to the deep depth.

00:00:27: I mean, imagine this for a second.

00:00:28: you're looking at a twenty billion dollar semiconductor facility

00:00:31: right?

00:00:31: Oh yeah massive scale

00:00:33: exactly.

00:00:33: You've got like the most advanced robotics on earth clean rooms that are you know Orders of magnitude more sterile than an operating theater machines manipulating matter at the atomic level and then Suddenly production just grinds to a dead halt.

00:00:48: Right,

00:00:48: the nightmare scenario What?

00:00:49: The culprit It's not uh... it is not a sophisticated cyber attack or massive power grid failure.

00:00:55: Its literally a sloppy note scribbled by technician during your fifteen minute shift change.

00:00:59: I mean sounds almost comical when you frame like that but thats actual reality on ground right now.

00:01:04: Yeah lets be real its wild It

00:01:05: totally is.

00:01:06: We spent calendar weeks, fifteen and sixteen looking at this massive influx of insights across LinkedIn straight from the people running these facilities And the overarching theme was well it was a really loud wake-up call.

00:01:19: smart manufacturing as no longer just this you know concept or acute pilot program Exactly its in the execution phase and the friction points we are hitting.

00:01:28: our entirely unexpected Today were basically going to map out for We'll start with the human reality on the shop floor, then look into industrial AI and why.

00:01:38: context is everything.

00:01:39: Context is huge!

00:01:40: Right?

00:01:41: And from there we will explore digital twins for hardcore execution... ...and wrap it all up with software-defined automation.

00:01:48: Perfect.

00:01:48: So let's just stay that semiconductor example a minute because I think this perfectly captures tension between high tech and Human Reality.

00:01:57: Matthew D shared this incredibly grounded observation about US semiconductor manufacturing.

00:02:02: Oh, I saw that one and really eye-opening.

00:02:04: Yeah because if you read the news You'd think the absolute biggest bottleneck in chip manufacturing is like acquiring EUV tools.

00:02:12: And for anyone outside of space Extreme ultraviolet lithography machines are arguably the most complex pieces of equipment in human history.

00:02:21: Right, they use microscopic mirrors to bounce light and print features smaller than a virus!

00:02:27: They're multi-hundred million dollar engineering marvels...

00:02:30: Exactly.

00:02:31: but Matthew argues that lack those machines or even like government funding isn't actually what's choking yield?

00:02:38: Yeah he points out real yield.

00:02:40: bleed happens right there during shift handoff.

00:02:43: A senior tech knows, say chamber three on a specific tool has this weird thermal quirk but they just don't document it properly.

00:02:50: They leave a vague log note night shift comes in runs a batch and suddenly you've got hundreds of thousands dollars of scrap silicon.

00:02:57: It's crazy!

00:02:58: Because technology scales infinitely But human discipline pays the bills.

00:03:03: If the communication protocol is broken The billions spent on hardware Just do not matter.

00:03:07: Absolutely, and honestly this exposes a massive structural flaw in how we think about automation.

00:03:13: We just assume automation is always overseeing the process but bad process governance actually creates this dangerous illusion of normal operations.

00:03:22: Oh that's a good way to put it an illusion

00:03:24: Yeah!

00:03:25: And Michael Adesina shared a story that highlights his perfectly.

00:03:29: so he was auditing a multi-module tool... ...and he noticed a warning flag on a Process Module.

00:03:36: But get THIS.

00:03:37: It hadn't triggered a hard fall.

00:03:38: Wait really?

00:03:39: No alarms!

00:03:40: None, it didn't shut down the line.

00:03:41: no flashing red lights across the floor... ...it had just been sitting there in completely silent failure state for nearly ten days while production kept rolling.

00:03:49: And days quietly failing in the background.

00:03:52: Okay how does that actually happen?

00:03:53: like why did this system catch it?

00:03:55: Because the system only catches what humans actually configure it to catch.

00:03:59: When Michael dug into the root cause, he found out that failure threshold for this specific tool had been misconfigured to zero M-tor.

00:04:07: Zero?

00:04:07: Wait...M-tor is a pressure measurement right?

00:04:10: Right..militor It's used in the near vacuum environments these machines operate but the actual baseline pressure was significantly higher than zero.

00:04:20: So the software was checking the current pressure against a limit that had absolutely nothing to do with physical reality.

00:04:26: That is wild, so it's not that this sensor failed to see the pressure drop.

00:04:31: It said the software essentially saying you know hey as long as the pressure isn't literally absolute zero we're totally fine.

00:04:38: Precisely I mean automation is obedient to a fault.

00:04:41: A misconfigured threshold doesn't just miss a problem...it actively suppresses it!

00:04:48: Absolutely not.

00:04:50: Which brings me to this quote I saw from Jacobo Lurekasal.

00:04:53: that really struck a nerve, he pointed out the costliest mistake in digital transformation isn't buying the wrong software stack it's hiring an IT director to solve and operations problem.

00:05:04: Oh wow yeah!

00:05:05: That is bold statement.

00:05:06: It sounds bit harsh at first rate but think about it.

00:05:09: its like installing state-of-the art home security system But leaving front door wide open.

00:05:16: The IT world you know, operational technology the physical machines.

00:05:21: they operate on completely different paradigms.

00:05:24: They really do.

00:05:25: I mean an IT director thinks in terms of data packets network security server uptime if a server goes down You reboot it.

00:05:32: no big deal right?

00:05:33: But if an OP system goes down?

00:05:35: Yeah,

00:05:35: exactly.

00:05:36: If OT goes down a two ton robotic arm might swing out of control or and entire continuous casting line of molten steel just freezes up.

00:05:44: the stakes are totally different yeah.

00:05:46: And Jacobo's point is that implementing a manufacturing execution system isn't really in IT project it to change management Projects It Just Happens To Involve Software.

00:05:55: You Need A Leader Who Knows Exactly What Happens When Align Stops At Two IM.

00:05:58: And that credibility, it really comes from understanding the end user.

00:06:02: Actually Erdem Osterk shared this fantastic anecdote about this.

00:06:05: He brought his six-year old daughter Ella into his facility.

00:06:08: Yeah

00:06:08: I love this one

00:06:09: It's so good.

00:06:10: So within five minutes she was running a tap test Which for context is when you strike a tool to capture force and acceleration signals To check its structural dynamics.

00:06:19: Pretty complex stuff

00:06:20: Totally But he picked up instantly.

00:06:23: but then her very first question why do you need cables?

00:06:26: Why can't you just listen to the

00:06:27: sound?".

00:06:28: Man,

00:06:28: that is a brilliant question.

00:06:30: To leave it to a kid right... It strips away all those legacy assumptions we have about how machines should be monitored

00:06:36: Exactly!

00:06:37: And Erdem's realization was really profound.

00:06:41: If a six-year old can intuitively grasp The goal of this process, WHY IS THE INDUSTRY STILL BARRIING THAT PROCESS BEHIND THESE INCREDIBLY DENSE UNUSABLE INTERFACES?

00:06:50: Because ease of use Is treated like luxury in manufacturing software.

00:06:54: Exactly Erdem argues it is actually the single biggest barrier to adoption today.

00:06:59: If the interface requires a PhD to navigate, The front-line workers will just go right back using clipboards and vague log notes.

00:07:06: Yeah And then we're going straight into the shift handoff problem.

00:07:08: So if we accept that human foundation usability governance context is paramount How do you start applying advanced tools like industrial AI To help these operators instead of confusing them

00:07:22: more?

00:07:22: It's huge question.

00:07:23: It is, because the numbers are getting impossible to ignore.

00:07:26: Like Nithyanayathan Parambathu shared stats showing AI predictive maintenance cutting unplanned downtime by up to thirty percent and computer vision catching defects with ninety-nine percent accuracy.

00:07:39: Those figures are super impressive but you know it's actually driving this sudden explosion?

00:07:43: Its not just better math its...the economics Peter van Schalkwijk highlighted his fundamental shift The cost of AI reasoning is completely collapsing.

00:07:51: How much we talking?

00:07:52: Well, back in twenty-twenty two if you wanted to process a billion tokens of data and think about token as like piece of word or datapoint it cost about twenty thousand dollars.

00:08:00: Okay, twenty grand today?

00:08:01: Today that exact same billion tokens costs under one hundred dollars.

00:08:06: Wait wait!

00:08:07: Under a

00:08:07: hundred?!

00:08:08: But hold on... If inference costs are almost zero And everyone is basically downloading the same foundational AI models Doesn't mean AI completely commoditized?

00:08:17: It's

00:08:17: really valid point

00:08:18: Like..If a mid market facility and fortune five hundred giant are running the exact same brain, how does anyone actually build a competitive advantage?

00:08:27: Won't the big players just buy up all the compute and win?

00:08:30: see that is The billion dollar question, and Peter's answer.

00:08:34: Is that?

00:08:34: The competitive mode is absolutely not the model itself.

00:08:37: Everyone has a model.

00:08:38: the mode is what we call the operational context layer

00:08:41: a context layer.

00:08:42: Yeah John Nixon summed it up perfectly.

00:08:44: He said AI without context is a hallucinogenic risk

00:08:49: a hallucina genetic risk.

00:08:50: I love that phrasing meaning The AI will confidently give you the wrong answer because it just doesn't know the specific rules of your factory.

00:08:57: Exactly, and Mobakey expanded on this too.

00:09:00: he pointed out that throwing a powerful AI at a massive lake of unstructured data Doesn't magically solve your problems.

00:09:07: It usually just exposes how noisy and messy Your data actually is

00:09:10: right?

00:09:10: It's just garbage in very smart garbage out

00:09:13: pretty much yeah for AI predictive maintenance to work A can't Just operate on generic rules.

00:09:18: like You Know if the temperature exceeds a hundred degrees trigger an alarm.

00:09:22: It has to understand the subtle multivariable patterns.

00:09:25: specific It needs to know that when vibration increases by two percent on spindle four and the ambient humidity is high, And the supplier for the current batch of steel changed yesterday.

00:09:38: Oh wow!

00:09:38: Yeah That specific combination leads to a failure in exactly three hours.

00:09:43: Ah

00:09:43: okay I see it's like The difference between hiring A brilliant engineer fresh out at MIT versus a Brilliant Engineer who has spent twenty years listening To this specific hum Of your machines.

00:09:54: The raw intelligence is the same, but the context changes everything.

00:09:58: That's a perfect analogy and David Rogers actually shared a case study of what building that context looks like in practice with HP Indigo.

00:10:04: so they manufactured commercial printing presses And They were just drowning In over ten terabytes Of daily operational data.

00:10:10: Ten

00:10:10: terabytes A day

00:10:11: Every single Day.

00:10:13: But The data was totally siloed.

00:10:15: Trapped in different ERP systems, different quality control formats and different machine logs.

00:10:20: It used to take them upto three days just to trace the lineage of a single manufactured part.

00:10:25: Three-days digging through databases.

00:10:27: Just find out where defective parts came from If that part is failing on field A three day delay Is an absolute Eternity.

00:10:35: Exactly, so they brought in data bricks to create a unified data backbone.

00:10:39: But they didn't just dump the data into a pile.

00:10:42: They mapped the relationships between the iKeyData and the otdata .They built that operational context layer.

00:10:49: they cut their part traceability time from three days down to just sixty minutes.

00:10:54: Wow, From seventy-two hours to one hour.

00:10:56: that is a staggering operational shift.

00:10:59: and you know we are seeing this move out of the back office indirectly onto the shop floor too.

00:11:04: yeah like Manisha Priyadarshini noted that at Siemens Amberg which has widely considered one of the most advanced digital factories in the world They aren't just playing around

00:11:12: right it's not lab experiment

00:11:14: Exactly.

00:11:15: Neural networks and automated visual inspections are already fully embedded at the field level, running in real time as part of a daily routine.

00:11:22: But here's the critical transition point.

00:11:24: If building an operational context layer is that important to making AI work you really can't wait until the factory has built.

00:11:32: start mapping it.

00:11:33: You have to map the context before even pour concrete

00:11:37: Which perfectly brings us two digital twins.

00:11:40: If you're walking your shop floor today, You really have to look at your three D models and ask yourself a hard question.

00:11:46: Yeah?

00:11:46: Do?

00:11:46: is this just an expensive video game we use to impress the board of directors?

00:11:50: or Is it in actual hardcore execution tool?

00:11:53: because The conversation on LinkedIn has completely shifted away from Just pretty Three-D visualizations?

00:11:59: Oh totally Philo, Presh and Jennifer Petrosky shared a use case from PepsiCo that illustrates what a true execution grade digital twin actually looks

00:12:07: like.

00:12:08: Okay break down first.

00:12:09: So PepsiCo used the Siemens Digital Twin Composer coupled with NVIDIA Omniverse.

00:12:14: They didn't just build a static three-D CAD model.

00:12:16: they built a kinematic physics based simulation

00:12:20: Physics Based meaning it acts like The real physical

00:12:23: world.

00:12:23: Exactly, they modeled the exact thermal properties... ...the conveyor belt speeds.. ..the mechanical constraints.

00:12:29: and get this!

00:12:30: Even the pads the operators would walk!

00:12:33: Wait- They mapped human paths too so that effectively spun up a parallel universe inside of server where could run factory before building it?

00:12:42: Let's dig into mechanism for that.

00:12:44: What does actually allow them to do?

00:12:45: It allows them to run thousands of what-if scenarios, like What happens if we increase line speed by a ten percent?

00:12:51: Where does the bottleneck form will be?

00:12:53: operators bump into each other.

00:12:55: By simulating The physical reality first PepsiCo identified ninety percent Of potential integration issues before they ever bought A piece of steel or modified a physical

00:13:04: layout.

00:13:05: That's incredible.

00:13:05: so what did that mean for the bottom line?

00:13:07: They boosted throughput by twenty percent and reduced their capital expenditure by ten to fifteen percent.

00:13:13: See, when you can test a theory in software rather than testing it by tearing up a concrete floor the ROI is massive.

00:13:21: Zucca Weisman pointed out that we are seeing this heavily in aerospace and defense right now too.

00:13:25: Oh for sure!

00:13:26: The tolerances there are unforgiving

00:13:28: Right microscopic tolerances...and incredibly high stakes.

00:13:31: They're using these digital twins to ramp-up complex production lines fifty to one hundred percent faster just to meet urgent global demand.

00:13:41: And Sirin Flanagan took this a step further by looking at the next frontier, which are AI factories.

00:13:47: We're entering an era where data centers become so massive and power hungry that they act as part-power utility in parts super computer.

00:13:55: Oh

00:13:55: man I imagine cooling requirements alone require an insane amount of predictive modeling.

00:14:00: Absolutely!

00:14:01: The complexity managing thermal load is so high that traditional engineering methods simply fail.

00:14:09: Ciaran noted that they are optimizing for entirely new metrics that didn't even exist a few years ago, things like time to token and tokens per watt.

00:14:17: Wow!

00:14:17: Yeah you literally cannot design a facility that complex on a two-D blueprint.

00:14:22: You have build it in digital twin simulate the immense thermal power interactions And only when math holds up virtual world do actually break ground.

00:14:30: Okay, so let's connect all these pieces.

00:14:31: We have the human foundation.

00:14:33: we had the localized AI context...we've simulated The perfect factory using a digital twin.

00:14:38: right but at the end of the day A Digital Twin has to translate into physical action.

00:14:44: How do we make the actual?

00:14:45: Physical hardware on the floor flexible enough To keep up with all this software?

00:14:51: That leads us to the massive structural shift happening in software-defined automation.

00:14:55: Yeah, Shadi Nassafat and Domenico Napoli from Schneider Electric have been highly vocal about this shift toward open software defined automation.

00:15:04: To understand why this is revolutionary you have to realize just how rigid industrial automation has historically been

00:15:10: Right!

00:15:11: Historically if you buy a robotic arm or control system From let's say vendor A You are permanently locked into Vendor A's proprietary software ecosystem.

00:15:19: It's a walled garden Like a phone that only runs one app forever.

00:15:22: Exactly, and the core of this system is usually PLC, the Programmable Logic Controller.

00:15:27: For those outside the engineering bubble A PLC essentially has highly ruggedized industrial computer.

00:15:32: It's brain tells valve to open or conveyor stop.

00:15:37: Historically hardware and software in a PLC were inextricably linked.

00:15:41: You couldn't separate them But Rene McBride highlighted massive breakthrough The rise of virtual PLCs.

00:15:50: We are now moving control logic entirely into software decoupling it from the physical hardware.

00:15:55: Wait, let me stop you there because I know old-school engineers here.

00:15:59: we're moving real time machine control Into software and they immediately break out in a cold sweat

00:16:04: all day panic.

00:16:05: Yeah They imagine like a windows update crashing while a robotic arm is swinging at car chassis across a room.

00:16:12: How do you move?

00:16:13: Physical control to virtual environment safely

00:16:16: It's a completely valid fear.

00:16:18: But we aren't talking about running a factory on standard consumer operating systems here.

00:16:22: They use real-time operating systems, running on highly reliable edge servers.

00:16:26: The latency is measured in microseconds

00:16:28: Okay so it's bulletproof

00:16:29: Basically.

00:16:29: And by moving the logic to an Edge server You turn the Factory floor into something resembling and open App Store.

00:16:35: you can choose best hardware from vendor A Best sensors from vendor B Run them all using centralized software.

00:16:41: So build application once and deploy across ten different factories globally with just keystroke.

00:16:47: Exactly

00:16:47: That flexibility is incredible.

00:16:49: Yeah, but and there's always a... But here is the blind spot.

00:16:52: Tolla Chauanca made a brilliant observation about why all this physical AI and software defined capability Is currently failing to scale in a lot of organizations.

00:17:02: Oh

00:17:02: I saw that post.

00:17:03: it's really harsh reality check

00:17:05: It is.

00:17:06: Tolla points out that companies are buying All these capabilities piecemeal.

00:17:10: they buy advanced sensors so The system has perception.

00:17:13: They implement eGendic AI, which you know is AI that doesn't just analyze data but can autonomously execute actions based on reasoning.

00:17:21: Right

00:17:22: and they buy digital twins for simulation.

00:17:25: So they have all the ingredients But they are completely missing the orchestration layer.

00:17:28: Yes The orchestration layers.

00:17:30: what forces IT systems OT hardware And physical operations to actually communicate an act as a unified entity?

00:17:37: right

00:17:38: I've heard this described as the conductor of the orchestra, but honestly i think a better analogy is an air traffic controller.

00:17:44: Think about a really busy airport.

00:17:46: you can have the most advanced autonomous fuel efficient planes in the world.

00:17:50: they all have perfect perception perfect reasoning But if they're only optimizing for themselves You know For their own fuel efficiency or flight path and there was no Air Traffic Controller coordinating The entire airspace

00:18:03: those planes are going to crash into each other on the runway

00:18:05: exactly.

00:18:06: And that is exactly what happens on a factory floor without orchestration.

00:18:10: The robots work, but not together.

00:18:13: Right the robotic arm optimizes its speed But the conveyor belt isn't ready!

00:18:17: The AI optimizes energy usage by shutting down a chiller... ...but it ignores thermal requirements of next

00:18:23: batch.

00:18:24: It optimizes but for the wrong objective.

00:18:26: Exactly!

00:18:27: Tullest point is that physical AI is a full stack, not single capability.

00:18:31: If you don't have orchestration acting as an air traffic controller You just have a bunch of highly intelligent machines actively working against each other.

00:18:39: It all comes back to alignment And honestly I think this ties everything we've talked about today together beautifully.

00:18:47: We discussed agentic AI physics-based digital twins, virtual PLCs.

00:18:52: You would think the primary hurdles in smart manufacturing were purely technical... Yeah you'd think

00:18:56: so yeah

00:18:57: But Breanne Carroll shared an insight that completely flips that narrative.

00:19:01: He warned That In The Rush To Digitize Everything Our teams are becoming socially anti social

00:19:07: Socially Anti Social Meaning?

00:19:10: The technology is actually creating new silos between people.

00:19:13: Precisely With all this talk of AI and twins.

00:19:17: by relying entirely on dashboards in automated alerts, we were losing the hallway conversations.

00:19:22: We are losing that quick intuitive gut check between day shift to night shift Which

00:19:26: loops us right back to Matthew D's point at very beginning.

00:19:29: The fifteen minute shift handoff...the human element.

00:19:32: Right Carol argues That we don't actually have a technology problem than manufacturing now.

00:19:39: If you deploy incredibly fast technology without establishing human trust and alignment first, it doesn't lead to transformation.

00:19:47: It just leads to fragmentation at a much faster speed.

00:19:49: That's

00:19:50: powerful take away.

00:19:51: You really is so as you head back to your facility today or you know As you sit down for the next operation strategy meeting I want you keep this central thought in mind to mull over.

00:20:01: Technology enables but people multiply.

00:20:04: You can build the most advanced digital twin in the world, but if you leave the operators behind at the analog age The entire system collapses.

00:20:12: well said.

00:20:13: If you enjoy this episode new episodes drop every two weeks.

00:20:17: Also check out our other editions on Digital construction and digital power tools.

00:20:21: keep it short and confident And with a thank-you and reminder to subscribe.