Best of LinkedIn: Smart Manufacturing & Industrial AI CW 35/ 36

00:00:00: provided by Thomas Olguyer and Frenis based on the most relevant Lincoln post about smart manufacturing in industrial AI, in calendar weeks thirty five and thirty six.

00:00:09: Frenes 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.

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

00:00:22: You know it feels like we're trying to inject AI into absolutely everything right now.

00:00:26: completely

00:00:26: It's like everywhere.

00:00:28: you look all right

00:00:29: but Here is the uncomfortable truth for anyone working in manufacturing.

00:00:34: You could have the most advanced AI model on Earth, but if there's say a five millisecond latency spike on your factory floor...

00:00:41: Just five milliseconds?

00:00:43: Exactly!

00:00:44: It could cause catastrophic safety shutdown.

00:00:47: So today we are stripping away fluff and buzzwords to examine raw mechanical reality of industrial AI.

00:00:54: We're looking at what is actively breaking down the shop floor And

00:00:59: if you are a professional in smart build and manufacturing, You already know the noise out there is just deafening.

00:01:05: I mean every vendor promises their new neural network Is going to magically double your output overnight.

00:01:11: Right it's just height.

00:01:12: Exactly Yeah.

00:01:13: So our mission for this deep dive is cutting through all that.

00:01:16: We're looking at the top Smart Manufacturing & Industrial AI trends That have been circulating on LinkedIn Extracting exactly what we need To Know about The Reality right now.

00:01:25: Because before we talk about what AI can do, We really need to look at why these digital transformations are currently failing.

00:01:32: I mean the failure rate is kind of The elephant in the room.

00:01:34: it absolutely is

00:01:36: and Jacobo lorak Kassal laid out the root cause for this beautifully.

00:01:40: he spent time observing over Fifty different factory floors.

00:01:44: Wow,

00:01:45: fifty that's a lot of ground to cover

00:01:46: Yeah and he saw very clear repeating pattern.

00:01:49: these massive manufacturing transformation failures They rarely have anything to do with technology gaps.

00:01:56: Okay so what is the actual issue then?

00:01:58: they're

00:01:58: failing because leadership Is misdiagnosing the problem entirely.

00:02:03: He saw CEOs chasing these shiny AI pilots while their overall equipment effectiveness Their OEEs just sitting at like Wait, fifty-eight percent.

00:02:13: Let's ground that for a second because an OEE of fifty eight percent means at nearly half the time your Multi million dollar equipment is either broken down producing scrap or just completely idle.

00:02:25: Exactly you are hammering money on the basics But you're trying to implement a neural network.

00:02:30: it makes no sense

00:02:31: right.

00:02:31: Junkabo illustrated this with a really common scenario.

00:02:34: So let's say your facility has energy costs suddenly double, okay?

00:02:38: Leadership immediate reflex might be to say well we have an energy crisis We need an AI agent to optimize our power consumption.

00:02:45: throw tech at it.

00:02:46: But Junkago points out that you likely don't have an Energy problem at all.

00:02:50: You have a visibility problem.

00:02:51: Oh because you literally Don't know which specific production line is drawing the power or why

00:02:56: exactly if your cost go up.

00:02:58: That is a visibility issue.

00:03:00: Buying an AI tool doesn't help if you don't even know what your looking at.

00:03:04: I mean, it's like buying a state of the art GPS for car that has flat tire.

00:03:08: That is perfect analogy.

00:03:10: You're just not going anywhere.

00:03:11: And this actually ties right into a post by Abhishek Chakrabarty.

00:03:15: He makes this critical point that autonomous plants require highly structured engineering knowledge before you can ever trust and AI to act.

00:03:24: Right.

00:03:24: because you cant plug gen AI in factory and say optimize this

00:03:29: Exactly It needs to understand the equipment hierarchy first.

00:03:32: Hmm, it has to know the failure modes and The basic cause-and-effect logic of machinery because

00:03:38: AI doesn't inherently Know how a physical plant operates

00:03:41: right?

00:03:42: And speaking of trusting the AI to act Chris Graham brings up another massive issue.

00:03:47: He argues that the next major bottleneck in industrial AI isn't gonna be model accuracy at all.

00:03:52: Oh really what is it then?

00:03:54: its decision authority Like what authority should the AI actually have?

00:03:58: Oh, that makes total sense.

00:03:59: Does it have the clearance to shut down a line?

00:04:01: exactly?

00:04:01: who pulls the trigger?

00:04:03: and Mike Guilfoyle adds to this too.

00:04:05: He says that workforce transition mechanics not The model capabilities themselves are What will actually govern the pace of industrial AI adoption.

00:04:15: right because if the workers don't trust It doesn't matter how good the AI is.

00:04:19: but let me push back on us a little bit though If the tech is ready and capable, but the governance and workforce just aren't there yet.

00:04:30: Are we basically just waiting on human psychology to catch up with algorithms?

00:04:34: Well I wouldn't call it just psychology.

00:04:36: you can train a work force or establish that governance if your systems don't even speak the same language to begin with.

00:04:42: Oh

00:04:42: okay fair.

00:04:43: And That brings us straight into backbone of all this.

00:04:46: which data architecture in connectivity?

00:04:48: You cannot build trust without transparency.

00:04:50: And you can't have transparency with out connectivity?

00:04:53: Exactly!

00:04:53: Senad Sawak used this brilliant analogy for this, he calls OPCUA the English Language of Industrial Automation The

00:05:00: English language.

00:05:01: I like that.

00:05:02: Yeah because it standardizes date exchange across all these wildly different industrial systems from your PLCs to SCADA to your MES.

00:05:10: Right.

00:05:10: so data doesn't just stay trapped in proprietary silos

00:05:13: exactly.

00:05:14: It acts as a universal translator.

00:05:16: but And this is a big but while standards help.

00:05:19: Andre Malishenko highlighted his totally new problem regarding edge platform proliferation.

00:05:24: Oh

00:05:25: I saw that one, it's so painfully accurate

00:05:27: Right.

00:05:28: He reminded everyone ten years ago if you wanted to update the software on factory machine You literally walked over with USB

00:05:35: stick Cold sneaker net.

00:05:37: Yeah.

00:05:38: But now Every single new vision system or packaging machine arrives with its own proprietary management

00:05:44: dashboard.

00:05:45: It's out of control.

00:05:46: We have basically turned highly skilled machine operators into like managers-of-management systems.

00:05:51: So absurd.

00:05:52: just picture the dashboard your car, right?

00:05:54: Right now imagine if checking you're speed Your gas and your check engine light meant You had to log in two three separate iPads bolted to your dashboard

00:06:03: while driving down The highway.

00:06:04: exactly that is what operators are dealing With on Monday mornings.

00:06:08: Hey, real quick if you're finding value in how we unpack all this take a second to subscribe To the deep dive so you don't miss future editions.

00:06:15: We really love breaking This down for you absolutely.

00:06:17: but getting back to this UI nightmare How do manufacturers actually escape?

00:06:22: This trap?

00:06:23: John Yoon's asks what I think is the ultimate architecture stress test question.

00:06:28: he asked What happens when you add the second plant?

00:06:31: Oh, that is a million dollar question.

00:06:34: Right

00:06:34: because if Plant B requires another massive web of custom pipelines and manual data mappings just to function it won't scale!

00:06:42: It's impossible...

00:06:42: ...it shatters whole system.

00:06:44: Exactly so John advocates for using process models.

00:06:47: they give this consistent way represent manufacturing entirely independent all messy hardware differences underneath.

00:06:53: And That Is So Critical Because once you achieve that standardized model, the possibilities totally change.

00:06:59: Stefan Obbs dug into this.

00:07:01: he found that combining engineering data with live runtime data is the key.

00:07:05: and He does.

00:07:06: using MCP servers

00:07:07: right?

00:07:07: Yeah multiple model context protocol Servers when use those to combine The original Engineering intent With the messy reality of the Live Runtime Data.

00:07:17: That Is what unlocks deeper AI Insights.

00:07:20: it breaks down Those silos.

00:07:21: Right It marries the blueprint with the Reality.

00:07:24: so Once you have that data architecture flowing and standardized, what do you actually do with it?

00:07:28: You

00:07:28: build predictive models.

00:07:30: Exactly!

00:07:31: Which links us to our next theme... Digital Twins & Predictive Engineering.

00:07:35: Because seeing the data is great but predicting what happens in X's where real value is

00:07:40: Yeah.

00:07:40: And Brent Roberts shared a phenomenal insight here about how predictive engineering analytics basically fills-in sensor gaps.

00:07:47: Sensor Gaps.

00:07:49: Physics is inconvenient.

00:07:51: You can't always just stick a physical sensor exactly where you need it.

00:07:54: Oh, like inside of high-pressure fluid flow or something

00:07:56: Right!

00:07:56: Or if you needed to measure the remaining fatigue life of deep internal gear A sensor would get crushed.

00:08:02: So validated simulation models estimate these behaviors instead.

00:08:06: Okay so turns physics based predictions into actual decision ready evidence

00:08:11: Exactly...you measure what you want on outside and the model calculates what must be happening on the inside.

00:08:17: But there's a historical roadblock to making that work, which Shankaraman pointed out.

00:08:22: he asked you know why do predictable equipment failures keep repeating across product generations?

00:08:28: Oh it's great question!

00:08:29: Why does new pump fail exactly same way old one did Right?

00:08:33: And answer is validation models are living in total silos.

00:08:39: Of

00:08:40: course they're.

00:08:40: You have simulation engineers validating designs But then the live field telemetry just sits in some IoT platform and The actual service failures.

00:08:49: they sit in warranty databases,

00:08:51: so they never actually talk to each other.

00:08:53: Never.

00:08:54: we have to close the loop To feed that field failure data back into the upfront simulation.

00:08:58: And when companies do close that loop it is wild.

00:09:02: what happens?

00:09:03: Jasper Wildbore showcased the CNC digital twin at the IMTS booth.

00:09:08: Well, I heard about this!

00:09:09: It's amazing.

00:09:10: it validates The entire machining process before a single physical chip of metal is ever cut.

00:09:15: That is incredible.

00:09:16: And Katharina with Westridge discussed A really comprehensive semiconductor Digital Twin.

00:09:21: It connects the design Manufacturing operations and the infrastructure Altogether just to prevent lost yield.

00:09:28: Since semiconductors a tiny defect Is incredibly expensive.

00:09:32: Tens of thousands of dollars just gone.

00:09:35: Wow, but let me ask you a focused question on this.

00:09:38: sure if we are creating these perfect digital replicas that can basically predict their own failures Are we essentially removing the concept of unplanned downtime entirely?

00:09:50: Or are we just shifting the burden to the IT department to keep the twin perfectly accurate?

00:09:55: Well, I mean the digital twin only matters if it can actually affect the physical world right?

00:09:59: Okay.

00:09:59: If the Twin predicts a failure but still relies on a human to walk over and turn a wrench you haven't really closed the loop.

00:10:05: And that bridges perfectly to our final theme Robotics & Physical AI.

00:10:09: Right The new frontier Max Wright had a really fascinating take on this.

00:10:13: He points out that advanced robotics is forcing us To completely rethink the traditional automation pyramid Because

00:10:19: the old pyramid just doesn't hold up anymore

00:10:21: Exactly.

00:10:22: The old layers PLC, SCADA, MEs ERP they just don't work when an autonomous robot needs live OT data to avoid crashing but also needs AI models with production context at the exact same time.

00:10:36: Right, it needs data from very bottom and top of the pyramid simultaneously Exactly And Yuri Trishkov detailed hardware reality trying to actually do this.

00:10:46: Deploying AI in a factory isn't about running biggest model on some cloud data center.

00:10:52: Because of latency issue we talked earlier.

00:10:55: It requires three-tier heterogeneous silicon architecture.

00:10:59: Okay break that down.

00:10:59: for me

00:11:00: Sure, so first you have the mobile or front line edge.

00:11:03: This is for a battery efficient vision directly on the robot.

00:11:06: then You Have The Deterministic Industrial Gateway.

00:11:09: this Is For Hard Real-Time.

00:11:11: I like millisecond Latency For Safety.

00:11:13: So

00:11:13: The Robot Doesn't Run Into A Wall

00:11:15: Or Person Yes.

00:11:16: And Finally You Have THE PRIVATE AI FACTORY.

00:11:19: THIS SITS LOCALLY BEHIND THE FIREWALL FOR FINE TUNING.

00:11:22: So it's a mix of different ships doing exactly what they're best at.

00:11:25: And to bring this to life, there was this amazing post about the SAP embodied AI Jam at The Swiss Smart Factory.

00:11:33: This was discussed by Dr.

00:11:35: Lele, Claudia Fiume and Alexander Finger.

00:11:37: Oh!

00:11:37: The hackathon right?

00:11:38: Yeah

00:11:39: They demonstrated an employee literally just using chat interface in an ERP system To assign to a humanoid robot.

00:11:47: A chat interface, like texting the robot?

00:11:50: Basically

00:11:50: yeah and the robot then confirmed it and physically executed the task of handling boxes in The Warehouse

00:11:57: Just completely bypassing the whole automation pyramid...

00:12:00: ...the business intent translated instantly into physical kinetic action.

00:12:03: It is wild!

00:12:04: And you know that really leaves with one less thing to chew on today.

00:12:08: What's

00:12:08: that?!

00:12:09: If we are moving toward a reality where an AI agent can diagnose the supply chain shortage, restructure and engineering digital twin.

00:12:16: And then dispatch a humanoid robot to The Warehouse floor to fix it-

00:12:19: Which is happening?

00:12:20: Right!

00:12:21: Then the ultimate competitive advantage for manufacturers won't be their technology stack... ...the tech will just be commoditized.. ..the real advantage would be their operational courage.

00:12:31: Operational

00:12:31: courage?!

00:12:32: Yes!!

00:12:32: The willingness actually let the system run itself.

00:12:35: Wow!

00:12:36: Operational Courage The willingness to let go of the steering wheel.

00:12:40: That is an incredible thought to leave on!

00:12:44: If you enjoyed this episode, new episodes drop every two weeks.

00:12:47: Also check out our other editions on digital construction and connected tools & equipment.

00:12:50: Thank You so much for joining us And don't forget to subscribe.

00:12:53: See ya next time.