Best of LinkedIn: Smart Manufacturing & Industrial AI CW 37/ 38

00:00:00: provided by Thomas Allgeier and Frenness, based on the most relevant LinkedIn posts about smart manufacturing in industrial AI.

00:00:06: In calendar weeks thirty seven and thirty eight friend is a B to be market research company that supports Smart Manufacturing providers with building feature-by-feature competitive intelligence That shows exactly how their product stacks up against competition.

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

00:00:22: you know it's funny.

00:00:23: We're looking at all these top trends of smart manufacturing right now And there was just this undeniable overarching pattern.

00:00:31: Yeah, the reality gap

00:00:32: exactly The Reality Gap because you the manufacturers listening to this You've already bought the technology.

00:00:37: I mean that checks for the AI the digital twins all of it they cleared right.

00:00:42: but the gap between just having That shiny new tech and actually turning into measurable results across your whole enterprise Mm-hmm?

00:00:48: That is what's keeping plant managers up at night?

00:00:50: Oh totally Because having a A really cool tool in one pilot facility is radically different from weaving it into the gritty daily reality of twenty global sites.

00:01:00: So, In this deep dive we're going to unpack that and let's actually start right at The foundational layer which Is your MES Your manufacturing execution system

00:01:08: Right because It feels like everybody has an MES At This point.

00:01:11: And the twenty-twenty six research data From Rockwell Automation proves That they found that ninety three percent Of manufacturers currently have An MES in place

00:01:19: Which is huge.

00:01:21: But the drop off after that initial purchase is just, it's staggering.

00:01:25: Out of that ninety-three percent only twenty eight percent have actually deployed at enterprise wide.

00:01:30: Wow Yeah.

00:01:31: And It gets worse A mere twenty three percent report having their MES fully integrated across their ERP Their product life cycle management quality and OT systems.

00:01:41: So nearly four out of five manufacturers are just operating these massive disconnected islands of execution?

00:01:46: Exactly!

00:01:47: You really have to wonder how we get stuck there.

00:01:49: you know, We've been doing software integrations for decades.

00:01:52: Well part of the bottleneck seems to be a fundamental misunderstanding Of what an MES is actually built-to do.

00:01:57: Lepiret Prince Indingea Angasaki pointed this out brilliantly.

00:02:01: Yeah I saw that

00:02:02: People treat MES like a silver bullet But he argues you have to define it by what is not.

00:02:08: Right, like.

00:02:09: It's not an enterprise resource planning system handling your high-level financials right?

00:02:14: its Not in advanced Planning and scheduling tool or a warehouse management System tracking pallets.

00:02:20: definitely not.

00:02:21: And certainly isn't your scatter system controlling the actual machine voltages Or your CMSS handling maintenance tickets,

00:02:29: right?

00:02:29: It's just the execution layer.

00:02:30: its sole job is to execute production and bridge The physical floor.

00:02:35: two those other corporate systems

00:02:37: Yeah.

00:02:37: And when you try to make it replace them or we just don't build those bridges at all.

00:02:41: The operational bleeding starts immediately.

00:02:43: Oh for sure.

00:02:44: DJ Kamari shared this painfully common scenario of that exact integration failure.

00:02:50: So say your MES successfully orchestrates a production run of like, a thousand units.

00:02:55: The metal is cut the parts are assembled...the physical work is genuinely finished But

00:02:59: digital handshake fails?

00:03:00: Yes!

00:03:01: The confirmation transaction from the MES back to the ERP system just errs out

00:03:06: And suddenly the ERPs completely blind

00:03:08: Exactly.

00:03:09: The MES knows the goods are sitting on the dock, but the ERP still thinks that raw materials were in their warehouse and finished goods don't even exist yet.

00:03:18: It's like a kitchen right?

00:03:19: Yeah!

00:03:19: How

00:03:19: so?!

00:03:20: Think of the MES as your brilliant sous-chef aggressively running the kitchen floor... ...and the ERP is executive chef sitting at back office handling high level menu & ingredient orders.

00:03:32: If they refuse to talk The sous chef cooks a thousand meals, throws them on the counter and the executive-chef clipboard still says the fridge is full.

00:03:41: And nothing has been

00:03:42: served.".

00:03:43: So your inventory?

00:03:44: Your downstream supply chain?

00:03:45: It all begins working off stale hallucinated numbers...

00:03:49: Exactly!

00:03:50: You end up overnighting raw materials you literally already own….

00:03:53: That's painful.

00:03:54: Why are those integrations so notoriously difficult to achieve?

00:03:59: Well, Jembergian articulated this perfectly.

00:04:01: We tend to introduce the software long before the actual physical processes and data governance are ready for it.

00:04:08: The software architecture assumes a perfectly standardized harmonious factory floor

00:04:12: which if you run a plant... You know that the floor is just dealing with constant exceptions!

00:04:17: The ambient temperature changed so an operator tweaked recipe

00:04:21: Right or material batch was slightly off.

00:04:23: So a seasoned machinist made judgment call on feed rate.

00:04:26: So by rushing the software implementation without locking down process governance first, factories aren't digitizing best practices.

00:04:34: Nope they are just digitizing their inconsistencies.

00:04:38: They're taking existing chaos and scaling it at speed of light.

00:04:42: And vendors themselves kind add to this complexity don't think?

00:04:45: Oh absolutely you can't buy a box labeled MES and expect your unique chaos.

00:04:51: Cure Thieves of S&R broke down how divergent vendor architectures can actually be.

00:04:56: Right, like if you look at Plex MES their approach leans really heavily toward a cloud native platform.

00:05:01: Yeah it's highly standardized and gives you incredibly fast ERP in machine connectivity right out of the box.

00:05:07: But then compare that to AV Eva.

00:05:09: And they lean towards much broader highly customizable industrial software ecosystem.

00:05:15: It spans the MES SCADA The historian Heavy analytics?

00:05:19: Exactly!

00:05:20: If have a highly bespoke legacy-heavy plant with decades of custom machinery, trying to force a rigid cloud native platform onto it might trigger an organ rejection.

00:05:30: Right you have to match the architecture to your physical reality.

00:05:33: You have too.

00:05:34: And look if that baseline data isn't flowing flawlessly between the MES and ERP It becomes incredibly obvious why everyone's new industrial AI investments are currently face planting.

00:05:46: Oh yeah If your AI agent is feeding on that stale ERP data we just talked about, it doesn't matter how sophisticated the neural network is.

00:05:53: It's going to make catastrophic decisions.

00:05:55: Yeah The scale gap in industrial AI Is arguably defining crisis and manufacturing.

00:05:59: right now Everyone wants autonomous self-automizing factory But almost no one is actually running one.

00:06:05: Joydeep D shared some sobering numbers on this.

00:06:07: Eighty five percent of manufacturers are currently stuck in AI pilots.

00:06:10: Wow!

00:06:11: Eighty Five percent?

00:06:12: And, eighty-six percent are using AI strictly in isolated walled off pockets.

00:06:17: A mere fifteen percent would describe their AI adoption as extensive.

00:06:20: So that means eighty five percent of the industry is essentially funding incredibly expensive science experiments That do not move the needle on the balance sheet.

00:06:29: Pretty much.

00:06:30: I mean a pilot works because you have team data scientist babysitting single machine But when try to scale across one hundred machines.

00:06:37: The real world introduces variables model has never seen and it breaks.

00:06:41: Right, pilot purgatory.

00:06:43: Windy Woo diagnosed the root cause of this stagnation and it really comes down to a misplaced expectation.

00:06:50: Executives are waiting for chat.

00:06:52: GPT moment further factories.

00:06:54: Oh interesting

00:06:54: Yeah they see consumer AI.

00:06:56: write up perfect essay in three seconds And just assume we can optimize supply chain as easily.

00:07:01: But general LLMs are trained on open internet right?

00:07:03: They read Wikipedia Reddit news articles.

00:07:06: Knowing the dictionary definition of a centrifugal pump is completely useless on.

00:07:19: It doesn't know its current operating constraints, it's vibration history from yesterday or how it physically interacts with the valve right next to it.

00:07:27: And by the way if you want to keep tracking how these shifts actually transition from those failed pilots into real scaled production make sure you subscribe so you catch future editions of our deep

00:07:37: dive for sure because it really all comes down to context.

00:07:40: Andreas Heine argued that we spent the entire last decade obsessively connecting our machines.

00:07:48: So getting the data flowing is no longer the bottleneck.

00:07:51: The bottleneck is translating raw machine tags into human or machine meaning.

00:07:56: Right, because an AI model couldn't care less about your PLC addresses... ...or some random register number blinking on a server If the data just says tag four-four-oh two as at eighty….

00:08:05: …the AI's completely blind!

00:08:07: Yeah Is AD a temperature?

00:08:09: is it speed, will the machine explode at eighty-one?

00:08:12: or is AD an optimal target.

00:08:14: It desperately needs semantic context.

00:08:16: Walker Reynolds illustrated how you actually build that context and he contrasted unified namespaces with knowledge graphs.

00:08:24: The industry is obsessed right now which acts like giant centralized phone book for your factory

00:08:31: Right.

00:08:31: And a UNS is fantastic for two-D plumbing, it tells you that asset A and asset B both exist within the same infrastructure.

00:08:38: But

00:08:39: if phone book doesn't tell how those people feel about each other or who reports to whom

00:08:43: Exactly!

00:08:44: A knowledge graph however Is like family tree combined with physics engine.

00:08:48: It defines relationships.

00:08:50: It tells system.

00:08:52: motor twelve physically drives pump seven.

00:08:55: It tells the AI that order forty-four seven.

00:08:58: one was produced on line three using a specific batch of raw materials.

00:09:02: And when an agent has that three dimensional scaffolded meaning, it can traverse data and actually reason through complex problem much like your best senior engineer would.

00:09:12: When you finally architect context correctly results are just staggering.

00:09:16: Suthir Shankarappa shared this amazing case study from Apollo Tires that proves what happens to escape pilot predatory.

00:09:23: Oh I loved.

00:09:24: Right, they built an AI-driven advanced process control system on AWS for their high mix extrusion lines.

00:09:31: And extrusion is a really messy physical process.

00:09:35: Rubber is sticky temperature sensitive and incredibly difficult to standardize.

00:09:39: Yeah traditionally every time they had recipe change in that line They lost valuable time and literal tons of scrap material just waiting for the physical process To re stabilize under new parameters.

00:09:49: So didn't build dashboard predicts failure.

00:09:52: dashboards just give humans more homework, you know?

00:09:54: Right.

00:09:55: They built a model at the edge that calculates the optimal startup line speed before the extrusion even begins.

00:10:01: as The rubber flows an AI agent watches the profiles in real time and physically writes corrections straight to the PLC.

00:10:08: It's actually closing the control loop.

00:10:10: it takes this steering wheel

00:10:11: Yeah And the business impact is undeniable roughly twenty six percent less start-up rework forty percent faster stabilization times, and they are recovering about twenty production hours every single month.

00:10:22: Twenty hours of heavy machinery running perfectly instead just churning out scrap?

00:10:26: It's amazing!

00:10:27: But wait...handing the physical controls on a heavy extrusion line over to an AI agent…that sounds like insurance nowhere.

00:10:33: Oh totally.

00:10:34: If you're plant manager The idea that algorithm writing directly to your PLC without human-hitting approval is terrifying.

00:10:41: How do engineer enough trust actually let the AI drive?

00:10:45: Well it requires an incredibly rigid governance framework.

00:10:49: You never just jump from a pilot straight to full autonomy, Sandeep Kulkarni laid out the six pillars required to build a production-ready AI agent.

00:10:58: Okay

00:10:58: what are they?

00:10:59: Build test run secure observe and govern.

00:11:03: missing even one of these drops you right back into demoware.

00:11:06: Let's actually look at Observe and Govern because that is where the trust is forged.

00:11:10: Observe means your engineers have complete transparent dashboard visibility into the model's logic.

00:11:16: Right, you can see exactly why AI wants to increase this speed by five percent before it even happens.

00:11:20: And govern is about establishing hard physical safety guardrails.

00:11:24: You program the PLC to physically reject any command from the AI That pushes temperature or speed past a dangerous threshold.

00:11:31: You transition slowly.

00:11:32: You go from human in the loop, where AI just actions to human on-the-loop.

00:11:37: Where the AI acts but a human is constantly supervising the boundaries.

00:11:42: And Apollo

00:11:42: Tires utilized a massive ML OAPS pipeline to manage this.

00:11:46: The AI becomes force multiplier for operators not unsupervised rogue agent

00:11:51: Exactly.

00:11:53: and that brings us into most critical yet somehow overlooked piece of this entire puzzle the human workforce.

00:12:00: Oh yeah, we spend so much time talking about autonomous AI but we are facing an existential demographic cliff regarding the actual humans required to run these advanced facilities.

00:12:09: Morgan Davis highlighted data from Deloitte and Manufacturing Institute that should honestly be setting off alarm bells in every boardroom right now.

00:12:16: Yeah

00:12:17: The numbers are wild

00:12:18: Over two point six million baby boomers retiring for manufacturing Right Now.

00:12:21: That mass exodus leads upto two-point one million jobs at risk of sitting completely unfilled through the year twenty thirty.

00:12:28: Think about the physical reality of that.

00:12:30: You have a sixty-two year old control engineer who doesn't even need to look at the SCADA system, right?

00:12:36: They can just listen to the pitch of a bearing whining on the floor and know it's gonna fail in three hours!

00:12:42: Right... That visceral tribal knowledge is literally walking out the door taking his pension with him.

00:12:48: So

00:12:49: this where AI transitions from an optimization tool through a survival mechanism.

00:12:53: Companies are rapidly deploying AI co-pilots to capture that tribal knowledge.

00:12:58: Yeah,

00:12:58: look at the partnerships forming like Google Cloud working with Uniwell.

00:13:02: They're building AI agents That help junior operators prioritize a cascading wall of alarms based on historical context.

00:13:09: And Maggie Slowick pointed out tools like IFS Resolve Which is designed specifically To tackle this exact brain drain.

00:13:16: Oh how does it work?

00:13:17: Well, when a piece of machinery goes down.

00:13:19: A junior technician doesn't have to spend four hours digging through PDF manuals.

00:13:24: the system puts twenty years of senior technician troubleshooting expertise directly in front them which drastically speeds up first time fault fixes.

00:13:33: that's incredible.

00:13:34: and we also seeing digital twins evolve to accelerate how we train this incoming less experienced workforce.

00:13:41: David Morley and Mark Hines about showcased a fascinating initiative by Siemens called meet at The Machine And they're partnering with TA machine tools on this.

00:13:50: Because the traditional commissioning process is painfully slow, right?

00:13:54: You order a half-million dollar CNC machine you wait months for it to sit on a cargo ship bolted to the floor and then you begin the weeks long process of programming simulating and training your operators on It while literally just sits there idle.

00:14:09: but under this new initiative They integrate the machine's exact digital twin, the CAM software and processing logic into a virtual environment.

00:14:20: Your technicians can begin programming crashing virtual tools validating code and getting fully trained before the physical machine ever leaves The Loading Dock.

00:14:29: So they estimate this concurrent engineering could cut machines ramp up time by about fifty percent.

00:14:34: Yeah having your time to value on heavy equipment is an absolute game changer.

00:14:38: But we do have add reality check here.

00:14:40: There's this hilarious but terrifying cautionary tale from Brian Carroll.

00:14:45: Oh, the AR headset story?

00:14:47: Yes!

00:14:47: So imagine a technician on your shop floor wearing an augmented reality headset.

00:14:52: The system analyzes the machine in front of them highlights component and confidently instructs the technician to remove it.

00:14:59: Okay

00:15:00: But the technician is staring at empty space.

00:15:02: That component doesn't actually exist on this machine.

00:15:05: See, AR doesn't fix bad configuration management.

00:15:08: it just puts Bad Configuration Management into three D. Exactly!

00:15:11: Digital Twins and augmented training environments are incredibly dangerous if your foundational data hygiene is

00:15:17: poor.

00:15:18: A digital twin as only good at its underlying configuration data In that specific AR failure.

00:15:23: the Three-D model powering a headset was built on Revision A of the Machine.

00:15:28: Right The service manual engineer studied was revision B But the physical asset humming on the shop floor had been heavily modified over years and was actually on revision C. So

00:15:38: putting wildly inaccurate information into a highly immersive headset doesn't help.

00:15:43: If your PLM, your ERP and your MES aren't relentlessly synced to reflect the real-time physical state of that asset.

00:15:50: Your digital twin isn't a twin at all.

00:15:52: It's just a hallucination.

00:15:54: it really is.

00:15:54: in all these capabilities mastering data layer deploying AI utilizing accurate digital twins They're completely reshaping global manufacturing footprints.

00:16:05: Yeah, we traditionally assume that deciding where to build a new factory is purely erased at the bottom for the cheapest labor.

00:16:10: but Daniel Kepper shared a survey of one thousand executives regarding their footprint decisions and The math has fundamentally changed what

00:16:18: they find.

00:16:18: well yes Fifty-four percent of the decision is still driven by baseline costs, but forty six percent are now based entirely on qualitative factors.

00:16:27: The availability high in technical talent, local tech infrastructure and ecosystem readiness.

00:16:33: Which

00:16:33: means traditional high cost regions can actually win these massive facility investments?

00:16:38: Absolutely!

00:16:39: If a region has robust digital infrastructure and the skilled engineers required to run highly automated AI-driven advanced processes, productivity games can completely dwarf the savings of cheap manual labor in a less-developed region.

00:16:57: It's a race for throughput and resilience, not just hourly wages...

00:17:00: it's a completely different battlefield yeah And I think that leads to a final thought you should evaluate against your own operations.

00:17:06: Yeah You likely spend millions of dollars hardening your factory networks against cyber threats?

00:17:11: You build massive expensive redundancies To protect again supply chain disruptions.

00:17:16: but Based on the data and trends we've unpacked today, maybe the biggest single point of failure in your entire enterprise probably isn't a server or shipping lane.

00:17:25: It is that forty-year veteran operator who's retiring next month?

00:17:30: If you haven't captured their context and digitized their brain yet all these shiny robotics won't save your throughput when they walk

00:17:41: out.

00:17:45: Thank you so much for joining us on this deep dive and remember to subscribe.