Best of LinkedIn: Smart Manufacturing CW 19/ 20

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 nineteen-and twenty.

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.

00:00:20: You can find more info In The Description.

00:00:22: Welcome to the deep dive, everyone.

00:00:24: Yeah welcome we're super excited to get into this one Absolutely

00:00:27: because I mean if you imagine a robot today You're probably picturing something that just executes a pre-programmed script Right

00:00:34: like a giant arm bolted to the floor in his safety cage.

00:00:37: Exactly.

00:00:38: but what if it actually looks at a messy chaotic factory floor?

00:00:42: Reasons through that chaos and decides its next move in real time?

00:00:47: For decades the factory literally had to be built around rigid limitations of machine.

00:00:52: Which is incredibly expensive!

00:00:54: Very expensive.

00:00:55: But today, based on the top insights we're unpacking from LinkedIn over calendar weeks in nineteen and twenty We are looking at a fundamental reversal of that dynamic where to explore how The industry is finally moving past the hype To scale systems that actually perceive an adapt

00:01:10: right?

00:01:10: And then we've clustered these insights into a few core themes so we can get straight to the good stuff.

00:01:14: were talking physical AI and robotics software defined automation digital twins and smart manufacturing execution.

00:01:22: It's going to be a packed discussion.

00:01:23: No fluff, just the actual ROI and how this tech is scaling on the floor right now

00:01:29: Let's jump into that.

00:01:30: first theme then physical AI and robotics.

00:01:33: The overarching challenge here has always been how do you scale artificial intelligence when the physical factory floor is so inherently rigid?

00:01:41: For years we've had robots doing one highly repetitive task in absolutely perfect conditions, but the sources from these past two weeks show this massive shift toward what people are calling physical

00:01:51: AI.".

00:01:52: And a perfect illustration of this comes.

00:01:56: He was analyzing Path Robotics' new system, which is called ROVE.

00:02:00: Oh the quadruped thing right?

00:02:01: Yeah The four-legged robot.

00:02:02: and to understand why this is such a big deal you really have to look at the historical constraints of welding automation.

00:02:09: Traditionally if your automating welding You need perfectly structured environment

00:02:13: Right.

00:02:14: So If are manufacturing massive assembly like I don't know A piece heavy earth moving equipment or structural steel cannot perfectly fixture it into a fixed robotic cell, you just simply can not automate that process.

00:02:30: Because the traditional robot is completely blind.

00:02:34: I mean if that massive steel plate shifts even a fraction of a millimeter because of thermal expansion or for just an imperfect fit.

00:02:42: The pre-programmed robot just welds thin air.

00:02:44: Exactly, it ruins the whole part...

00:02:46: Which is a nightmare!

00:02:47: It is and Path Robotics is flipping that entire architecture.

00:02:51: They are pairing this quadruped robot with what they call an obsidian physical AI model.

00:02:56: Now legged robots have historically been considered far too unstable for high precision tasks like welding.

00:03:01: Yeah I would think the vibration alone will totally compromise the weld pool.

00:03:04: That's the conventional wisdom.

00:03:06: yeah But the physical AI model processes real-time sensor and visual data millisecond by millisecond, so it actively compensates for its own micro movements.

00:03:18: Wow!

00:03:18: And not just that—it compensates on uneven surface of factory floor with imperfect fit up of metal allowing to bring autonomous welding directly into workpiece.

00:03:28: It's basically like taking a train off tracks but instead giving you GPS and real-time reflexes.

00:03:34: It's a great way to put it!

00:03:35: So, you can actually feel the uneven terrain...and adjust its suspension instantly—you know?

00:03:40: To keep a glass of water from spilling…instead spending millions of dollars pouring concrete into this rigid factory environment around the limitations of the robot….the robot just navigates the messy reality in the facility.

00:03:53: Right —the factory stays messy – the robot gets smart …and this adaptability isn't limited only to heavy metal fabrication.

00:03:59: Oh definitely not.

00:04:00: Robert Little posted about a massive transformation happening in logistics.

00:04:05: Specifically, UPS is investing nine billion dollars in modernization and they are deploying technology from Pickle Robot Company to tackle the loading dock.

00:04:15: The Loading Dock Is Wild I mean, specifically the inside of a floor-loaded trailer.

00:04:20: It is arguably the most chaotic unpredictable environment in global supply chains

00:04:26: because

00:04:26: floor loaded trailers are essentially giant randomized three D puzzles.

00:04:30: The boxes shift they get crushed They tumble during transit.

00:04:33: you cannot program A robotic arm with a fixed set of XYZ coordinates to unload that mess.

00:04:39: now

00:04:39: it would just smash into things

00:04:41: right.

00:04:42: But these pickle robots use advanced machine vision to evaluate the chaos.

00:04:45: The camera parses the three-D geometry of shifting boxes, it calculates physics in center gravity and physically reasons out best trajectory for next package.

00:04:55: And does that without causing a localized avalanche?

00:04:58: Exactly!

00:05:00: Crazy part is they are doing this without requiring UPS modify physical building or standardize incoming freight

00:05:06: And this tech is moving out of the lab incredibly quickly.

00:05:09: Timo Kisner and Christian Doyle both pointed out that this integration of physical AI, Is already scaling to complex humanoid forms.

00:05:17: The humanoids?

00:05:18: Yeah!

00:05:18: Siemens, Humanoid & NVIDIA have actually partnered To test humanoid robots in live operations at the Siemens Electronics Factory In Erlingen Germany.

00:05:28: Wait really?

00:05:29: Live Operations?

00:05:30: Yes

00:05:31: They are processing immense amounts of visual and tactile data to perform tasks that require like genuine human-like dexterity, and spatial awareness.

00:05:41: Okay I have to jump in here.

00:05:42: whenever i hear humanoid robots my hype radar starts flashing red.

00:05:46: Fair enough Based on an insight from Tobias Claus I really have to ask, aren't we just throwing wildly expensive sci-fi technology at every single industrial problem now?

00:05:54: Just because it looks cool to investors.

00:05:56: Well Tobias Claus actually provides a very necessary economic reality check on that exact point he outlines.

00:06:02: the robotics boom is definitely not going to scale equally across all sectors

00:06:05: and goodness

00:06:06: right humanoid general purpose robots are fascinating.

00:06:10: but The real rule of thumb for Robotics adoption relies On four strict factors you need A clear environment a repetitive task, a measurable return on investment and a bulletproof safety case.

00:06:21: So the form of the robot strictly follows

00:06:25: Precisely.

00:06:26: That is exactly why automated warehouses are scaling much faster than, say autonomous tractors in open unpredictable agricultural fields.

00:06:35: It scales incrementally where the math makes sense.

00:06:37: today

00:06:38: it's total since.

00:06:39: so we're looking at deployments and logistics or specialized inspection tasks not some sweeping rollout Where humanoids suddenly replace the entire manufacturing workforce overnight.

00:06:50: but you know deploying a highly adaptable self-thinking robot like Rove exposes a massive bottleneck further down the stack.

00:06:56: Oh, so?

00:06:57: Well you can have a machine that perceives and adapts but if The Factory it operates in is running on rigid hardcoded software infrastructure from the nineteen nineties You hit a wall.

00:07:06: oh

00:07:06: absolutely!

00:07:07: The physical machines might be ready But the underlying digital architecture needs a complete rewrite

00:07:12: which brings us perfectly to our second theme the necessity of Software Defined Automation or SDA.

00:07:19: Etienne Lacroix summarized the core architectural problem beautifully.

00:07:23: He described it as fragmentation by design.

00:07:26: That's

00:07:26: a great phrase, Fragmentation By Design.

00:07:29: Right because right now The logic required to run A factory is locked inside proprietary PLCs Those programmable logic controllers that act As physical computerized brains of machines.

00:07:42: And you've also got separate robot controllers Separate safety plcs and specific hardware from specific vendors.

00:07:48: And none of them natively talk to each other?

00:07:50: It's literally like having a highly skilled workforce, where every single person speaks a different language and the company policy strictly forbids the use of a translator.

00:07:58: Exactly!

00:07:59: Because each hardware component has its own siloed environment.

00:08:02: changing anything is agonizing slow.

00:08:05: You are essentially locked into the Hardware Vendor's specific product roadmap.

00:08:10: So what's the fix?

00:08:11: Well Etienne Lacroix argues that To unlock real agility, the industry must fundamentally decouple hardware from software.

00:08:19: The control logic that runs.

00:08:20: robots, conveyors and actuators at all must be virtualized through open-software interfaces.

00:08:28: Fabrice Monnier shared a highly practical example of this decoupling in work but within municipal water utilities.

00:08:37: He spoke to operators who manage infrastructure literally keeping communities alive!

00:08:43: Municipal water supplies require near-perfect uptime.

00:08:46: Right, you can't just pause the city's water for a software update.

00:08:49: Exactly and when a legacy PLC controller fails in traditional setup The system goes down.

00:08:55: You need maintenance window to physically dispatch a technician Pull broken box out of cabinet Wire into new one And manually upload programming

00:09:03: Which takes hours.

00:09:05: But with open software defined automation The control logic lives on a software abstraction layer.

00:09:10: Yeah, it is completely hardware agnostic.

00:09:12: So

00:09:12: what happens when a node fails?

00:09:14: When a physical node fails on the network... ...the system automatically redistributes the logic over-the IP Network to backup server onsite in real time.

00:09:23: itself heals.

00:09:24: Wow!

00:09:25: It works exactly how cloud computing balances web traffic without shutting down water supply for millions of people.

00:09:31: That's incredible and Annamarie Grossfree posted about massive collaboration proving this work beyond just utilities too.

00:09:39: Audi and Siemens have successfully proven that these virtual PLCs can scale in real high-volume automotive production.

00:09:47: So they're doing this on the factory floor right now?

00:09:48: Yes,

00:09:49: This isn't theoretical!

00:09:50: By separating their application layer from hardware an engineer can develop control logic in one centralized location And instantly deploy it across multiple global plants

00:10:00: Regardless of a local hardware mix sitting at the floor.

00:10:04: Mexico versus Germany.

00:10:05: Okay, but I want to raise a structural concern here based on some insights from Pankaj Kulkarni and Dinakar Ramamurthy.

00:10:13: If we make the factory infinitely flexible if software logic is just floating between hardware components And robots are dynamically generating their own pads doesn't that?

00:10:23: Just scale the chaos.

00:10:24: That

00:10:24: Is The Big Fear.

00:10:25: Yeah

00:10:26: kulkarny Makes A Brilliant Point.

00:10:27: He Says Digitizing fragmented operating models doesn't create agility, it just scales the complexity of your broken processes.

00:10:35: And that is The Ultimate Architectural Trap of Industry-FourpointO.

00:10:38: Dinakar Ramamurthy highlighted his experience leading a global manufacturing network design across over seventy plants for a major automaker.

00:10:46: Seventy

00:10:47: Plants!

00:10:47: Yeah.

00:10:48: And he noted that the primary challenge wasn't deploying advanced AI.

00:10:51: It was standardizing operational behavior across all those disparate facilities.

00:10:56: You cannot drop an adaptive physical AI into a factory if underlying data foundation is fractured.

00:11:01: mess

00:11:02: He argues.

00:11:02: before you can talk about autonomous factories, you have to build robust digital spine.

00:11:07: A digital spine right?

00:11:08: This means industrial backbone securely connects edge computing on machines To cloud architectures but specifically designed to have a latency-sensitive workloads.

00:11:19: Because you can't have a robot waiting two seconds for a cloud server... ...to tell it stop before hitting human?

00:11:25: Exactly, the latency would be fatal!

00:11:27: But, standardizing that digital spine and unifying the data layer is what allows you to map the entire physical factory into a virtual space.

00:11:36: Once your hardware or software are decoupled in communicating over unified architecture You no longer have to rely on physical trial-and-error To design production line.

00:11:45: You can just simulate whole thing?

00:11:46: Right!

00:11:47: You can simulate thermodynamics, kinematics And material flow virtually Which perfectly transitions us Into our third theme Digital Twins & Simulation.

00:11:56: And this shifts the conversation entirely.

00:11:58: We aren't just talking about rotating pretty three-D CAD models on a screen anymore, are we?

00:12:02: Not at all!

00:12:03: Engineers are actively running predictive experiments on their production lines in the virtual world.

00:12:08: Holder Libtruth and Matthias Heinecke provided a standout example regarding PepsiCo.

00:12:13: What're they doing?

00:12:14: PepsiCo is utilizing The Siemens Digital Twin Composer integrated with NVIDIA Omniverse.

00:12:20: They took their static three D factory layouts integrated them with live data streaming from IoT sensors scattered across their physical lines and merged them into a single living digital twin.

00:12:32: And what's fascinating is the mechanism of how this actually works.

00:12:36: By running realistic physics simulations inside NVIDIA Omniverse, they can simulate the exact friction and weight of thousands of bottles moving down a conveyor belt.

00:12:44: It's

00:12:44: mind-blowing!

00:12:45: They identified up to ninety percent potential collision issues in bottlenecks before ever poured concrete or welded into physical conveyer.

00:12:52: Ninety percent?

00:12:53: That is huge!

00:12:55: And that predictive capability yielded a twenty percent increase in physical throughput, completely validated and software-first.

00:13:01: That is massive!

00:13:02: And Felix Hagen highlighted how this virtual first approach is fundamentally changing the pharmaceutical industry too... ...where

00:13:07: the stakes are just immense.

00:13:10: Right because a failed chemical batch doesn't just result in lost revenue it can disrupt the supply of life saving medications.

00:13:17: Regulators like FDA and EMA in Europe are now actively encouraging model-based evidence.

00:13:24: So pharma companies are using digital twins to simulate the complex thermodynamic and chemical reactions of a drug batch?

00:13:31: Exactly!

00:13:32: They

00:13:32: can predict a temperature anomaly that would cause a batch failure mathematically long before they mix physical active ingredients.

00:13:39: Yes, predictive quality control on whole new level.

00:13:41: Lars

00:13:42: Ochles actually posted about similar productivity leap at Willis Custom Yachts...a luxury boat builder.

00:13:47: They run a full digital twin of their CNC environment, the automated milling machines that cut their physical parts.

00:13:54: So they simulate cuts before it happens?

00:13:57: Yeah!

00:13:57: By simulating exact G-Code and toolpaths in the digital twin... ...they prove cutting tools wouldn't crash into material.

00:14:04: That gave them confidence to run massive CNC machine lights out over weekend with no humans in building.

00:14:10: Wow!

00:14:11: They achieved a five hundred percent productivity increase.

00:14:13: Five hundred percent.

00:14:15: But, you know this reliance on predictive simulation introduces a fascinating philosophical and operational tension.

00:14:23: It was raised by Sean Sehe.

00:14:25: he compared these industrial digital twins to the Truman Show.

00:14:29: Oh, I love that movie.

00:14:30: Wait how so?

00:14:31: While in the movie The Control Room above-the-town had a command layer That simulated weather and redirected traffic Before the main character ever reached the edge of his world.

00:14:40: Right they controlled everything from Above.

00:14:42: Exactly.

00:14:43: Industrial operators are building that exact same Command Layer for their factories today.

00:14:48: But say he poses the ultimate governance question to you the listener.

00:14:51: Okay what is it?

00:14:52: When the simulation armed with millions Of historical data points knows more than factory floor and the two systems disagree, who owns the authoritative decision?

00:15:01: Oh that's a tough one.

00:15:02: If The Digital Twin predicts the machine spindle is failing based on its vibration model but the physical sensors in the actual machines say temperature and vibration are perfectly fine Who do you trust?

00:15:13: Answering this question of who decides forces us to look at how decisions actually executed on the floor which brings up our final theme agentic manufacturing and smart execution.

00:15:25: Right because you can have the best predictive simulation in the world, but if that data doesn't safely govern the real-time physical actions of machines it's entirely useless.

00:15:35: Gwennel Huet pointed out a staggering inefficiency.

00:15:38: currently seventy three percent of manufacturing data goes entirely unused.

00:15:43: Seventy Three percent?

00:15:44: Yeah!

00:15:44: It is locked in silos stripped off its operational context or surface to operators far too late be actionable.

00:15:51: The entire goal of the industry right now is closing the latency loop between generating a predictive insight and executing physical action.

00:15:58: And doing that requires moving beyond traditional, if-then programming.

00:16:02: Kudzai Mandatera offered brilliant breakdowns on how expert operator intuition actually works!

00:16:07: This was fascinating….

00:16:09: When you watch as seasoned human operators run complex industrial machines it often takes them five to ten years master high value tasks because they learn by physically practicing through weird nonlinear edge cases

00:16:21: like the humidity and the plant changing, or a raw material being slightly out of spec.

00:16:25: Exactly!

00:16:26: You cannot write a simple line-of code to replicate that kind of deep contextual intuition.

00:16:31: Traditional control theory handles.

00:16:37: But that missing twenty percent, the nuanced behavior that dictates ultimate yield and quality has always been a black box reliant on human expertise.

00:16:46: And

00:16:46: Manda Teresa points out that multi-agent AI is now architected to replicate this.

00:16:51: instead of one massive confusing neural network they break task specific expertise down into discrete skills.

00:16:58: How does it look in practice?

00:17:00: Well, you have different AI agents handling different parameters.

00:17:03: One agent calculates the optimal control theory.

00:17:05: another agent strictly monitors safety thresholds and other uses learned policies to handle the weird environmental edge cases.

00:17:12: Ah

00:17:12: so they work together

00:17:13: Right And overall of them sits a supervisor agent.

00:17:17: So if throughput agent wants run machine ten percent faster but the safety agent calculate that resulting vibration exceeds the threshold The supervisor agent arbitrates it throttles speed to safe five-percent increase.

00:17:28: instead.

00:17:28: It is literally a team of specialized digital experts debating the best way to run the machine in real time.

00:17:35: Exactly!

00:17:36: And, The speed at which this is moving from theory to reality is just remarkable.

00:17:41: Zdenekmachala noted that In twenty-twenty his master's thesis proposed an autonomous shop floor agent operating inside a digital twin.

00:17:49: That

00:17:49: was five years ago Yes

00:17:50: and Five Years Ago it was dismissed as too academic because the infrastructure simply wasn't there.

00:17:55: Today, it is mainstream reality.

00:17:58: Alexandra P highlighted the new Siemens Eigen Engineering Agent which automates complex engineering workflows from planning

00:18:05: to execution.

00:18:06: and what kind of results are they seeing?

00:18:08: It's currently delivering upto an eighty percent improvement in solution quality And a fifty-percent boost in engineering efficiency.

00:18:14: Okay I have played devil's advocate here channeling some insights from Emory Brew and Chris Stevens.

00:18:19: An AI model is inherently probabilistic.

00:18:21: right yes evaluates data and guesses The most likely correct answer.

00:18:26: But a shop floor control system is strictly deterministic.

00:18:29: It has to execute exactly what it's told, precisely when it was told to do or heavy machinery breaks and humans get severely injured.

00:18:37: That's

00:18:37: very true!

00:18:38: So directly connecting a probabilistic guessing AI into a deterministic high-stakes control systems sounds like an absolute recipe for industrial disaster?

00:18:49: It is completely valid fear And it is exactly why an industrial orchestration layer isn't absolute requirement.

00:18:55: As Bru and Stevens point out, you cannot hardwire in AI directly to the machine's actuators.

00:19:01: You must have a secure deterministic framework sitting between the AI models.

00:19:06: So this orchestration layer acts as a translator and a strict firewall.

00:19:10: Exactly!

00:19:10: It takes the AI's probabilistic suggestion like, you know I am ninety-five percent sure we should increase the pressure And passes it through rigid deterministic safety protocol.

00:19:19: If requested pressure exceeds physical burst limit of pipe The orchestration layers just blocks command.

00:19:27: It keeps human operators safe while still allowing the system to leverage the agility of AI where appropriate.

00:19:33: Yes, and on a more foundational level implementing this execution layer that MES or Manufacturing Execution System is often where companies fail organizationally.

00:19:42: Well for sure

00:19:43: Kim Wilson & Joe Gerstle point out many manufacturers stumble because they write their RFPs completely backwards.

00:19:50: They start with checklists of technological buzzwords rather than the problem they actually need to solve.

00:19:56: Exactly, They focus on generic future lists instead of starting with actual business problems and desired operational outcomes and the measurable success criteria.

00:20:06: You can purchase the most sophisticated orchestration layer in the world, but if your MES implementation doesn't strictly align production execution with actual business goals it just becomes an incredibly expensive source of digital noise.

00:20:20: It constantly circles back to execution

00:20:24: It does.

00:20:25: And as we wrap up, I want to leave you with a final thought tomorrow over based on a fascinating insight from Daniel Keper.

00:20:31: What's

00:20:31: the takeaway?

00:20:32: Well he analyzed Tesla's move upstream into semiconductor fabrication To secure that compute power they desperately need for their AI and robotics programs.

00:20:41: Oh!

00:20:42: That makes sense.

00:20:42: In the age of physical AI, computing power itself is becoming a strategic manufacturing input just like raw steel or electricity.

00:20:51: Because sending real-time robotics data to remote cloud introduces unacceptable latency.

00:20:57: factories will require massive localized datacenters sitting right on the facility floor.

00:21:02: Wow so whole factory footprint changes.

00:21:05: For decades, the primary question for global manufacturing was where is human labor?

00:21:09: The cheapest.

00:21:10: But the new question is Where can the full stack automation localized compute massive energy grids and the surrounding ecosystems scale the fastest?

00:21:19: that shift will fundamentally rewrite the logic of global production networks over the next

00:21:23: decade.

00:21:24: That has a wild thought to end on.

00:21:26: if you enjoyed this episode New episodes drop every two weeks.

00:21:30: also check out our other editions on digital construction And digital power tools.

00:21:34: Thanks for joining us on this deep dive.

00:21:36: Keep questioning the old frameworks and keep looking at mechanisms driving real-world applications.

00:21:42: Have a great week, don't forget to subscribe so you never miss an insight!

00:21:46: We'll see ya next time.