Latest / The Manufacturing Podcast with Fexingo: Factories, Supply Chains, and Industrial Business / How a Japanese Factory Uses Paper-Thin Sensors to Cut Waste
Transcript
- Lucas: There is a factory on the island of Shikoku in Japan that has been making paper for over a century. Mitsubishi Paper Mills, their specialty papers division. And in the last two years they've essentially turned their own product into a sensor network. Luna: Wait — the paper itself becomes a sensor? Or are they printing sensors onto paper? Lucas: Both, actually. They developed a conductive paper — cellulose fiber embedded with carbon nanotubes and silver nanowires. You can print circuit patterns onto it using standard inkjet printers. So the paper acts as both the substrate and the sensing element. They've deployed these paper-thin sensors across their own production line. Luna: So they're monitoring their own paper-making process using sensors made of paper. That is incredibly meta. Lucas: It is. And it's not just a proof of concept. They've got about 400 of these sensors deployed along a single line — monitoring moisture content, temperature, vibration, even chemical composition in the pulp slurry. The sensors cost roughly twenty cents each to produce, compared to ten to fifty dollars for a conventional industrial sensor. Luna: At that price, you can basically carpet-bomb the factory floor with data points. Lucas: Exactly. And that density matters. In a paper mill, moisture variation of even half a percent across the web can cause breaks, waste, energy inefficiency. With sparse sensors you might catch the macro trend but miss the local hotspot. With these cheap paper sensors they placed them every thirty centimeters across the width of the paper web. Luna: What kind of results did they see? Lucas: Over an 18-month period, raw material waste dropped 27 percent. Energy consumption per ton of paper fell 12 percent. And unplanned downtime from web breaks — which is the biggest productivity killer in paper — dropped by nearly 40 percent. They published these numbers in a technical paper for the Japan TAPPI conference earlier this year. Luna: A 40 percent reduction in web breaks alone probably pays for the whole sensor project in a few months. Lucas: That's exactly what their operations team said. The ROI was under six months. And now they're commercializing the sensor technology externally. They sell it as 'Mitsubishi Sensing Paper' — rolls of sensor-embedded paper that factories can cut to size and attach to equipment. Luna: I love that it's a product they can sell, not just an internal efficiency gain. That's the kind of pivot that turns a cost center into a revenue stream. Lucas: Right. And the applications go way beyond paper mills. They've tested these sensors in automotive paint shops — measuring humidity and temperature in curing ovens. In food processing plants — monitoring cold chain integrity. In data centers — detecting micro-vibrations that precede hard drive failure. Luna: Are there limitations? I imagine paper isn't great in high-moisture or high-temperature environments for long periods. Lucas: That's the tradeoff. The sensors are biodegradable and have a limited lifespan — roughly six months to two years depending on conditions. But that's actually fine for many use cases. If the sensor degrades after a year, you just peel it off and stick a new one. At twenty cents, it's cheaper than wiring a traditional sensor or sending a technician to replace a battery. Luna: So it's almost a consumable — like the sticky notes of industrial sensing. Lucas: Exactly. And that changes the economics of data collection for older factories. Retrofitting a 1970s-era production line with conventional sensors often costs more than the line is worth. You need conduit, wiring, power, integration. With paper sensors, you just stick them on and they communicate wirelessly via RFID or Bluetooth Low Energy to a gateway. Luna: That makes me think about the labor shortage problem. If you can cheaply monitor equipment condition, you can shift from time-based maintenance to condition-based maintenance. Fewer skilled technicians needed, and those you have focus on the machines that actually need help. Lucas: That's exactly the strategy at Mitsubishi's own plant. Their maintenance team went from doing 60 percent scheduled checks and 40 percent reactive repairs to the reverse — 70 percent predictive, 30 percent reactive. And they did it without adding headcount. Luna: There's a broader trend here too. The cost of sensing is dropping so fast that the bottleneck is no longer hardware — it's interpretation. You can generate gigabytes of data from paper sensors, but you need the analytics layer to turn it into decisions. Lucas: That's where Mitsubishi partnered with a Tokyo-based AI startup called Abeja. They built a machine learning model that ingests all 400 sensor streams and predicts web breaks 20 minutes before they happen. Operators get an alert — 'moisture gradient exceeding threshold at position 142' — and they can adjust the steam pressure or the draw speed before the break occurs. Luna: Twenty minutes is an eternity in a paper mill running at 90 meters per minute. Lucas: It is. And the model itself improves over time. It started with about 70 percent accuracy on break prediction. After six months of data, it hit 93 percent. That's the flywheel — cheap sensors generate more data, which trains better models, which make the sensors more valuable. Luna: So what's the catch? Why isn't every factory doing this? Lucas: Three barriers. First, the sensors need a stable wireless environment. Factories with lots of metal interference or legacy radio systems can struggle with data dropouts. Second, the analytics layer still requires data science talent, which is scarce. Mitsubishi could afford to hire a team because they're a large company. A 50-person factory might not have that luxury. Luna: And the third barrier? Lucas: Cultural. Many plant managers are used to walking the line, feeling the paper, listening for problems. Handing that judgment over to an algorithm is a hard sell. The Mitsubishi plant manager told me it took about a year of parallel runs — algorithm recommendations alongside operator intuition — before the operators trusted the system enough to act on alerts alone. Luna: I can see that. In a high-stakes process, you don't want to be the person who ignored the machine and caused a $50,000 break. Lucas: Exactly. But once the trust is there, the results speak for themselves. And I think this is one of those quiet revolutions that doesn't make huge headlines — it's not a flashy robot or a new 3D printing material. It's a twenty-cent piece of paper that tells you when something is about to go wrong. Luna: Sometimes the most impactful innovations are the ones we barely notice. Speaking of things that deliver value quietly — if today's episode gave you a useful idea or a new way to think about your own factory floor, that's exactly what we're trying to do here. Lucas: And the way we keep these conversations ad-free and focused on real case studies is listener support. If you've gotten value from the podcast over the months, you can help keep it going at buy me a coffee dot com slash fexingo. No pressure, no perks — just a way to say this matters to you. Luna: And it really does help. Every contribution goes directly into research and production time, so we can keep digging into stories like this one. Lucas: Alright — back to the factory floor. One more thing that stood out to me about the Mitsubishi project: they're now licensing the sensor design to other paper manufacturers. So a company that makes cardboard boxes in Ohio could, in theory, produce its own sensors using the same conductive paper recipe. Luna: That's almost like open-sourcing the sensor. If every paper mill becomes a sensor fab, the cost could drop to pennies. Lucas: Exactly. And that's the direction we're heading — commoditized sensing that's so cheap you don't think twice about deploying it. The question is whether the analytics layer can keep pace. Because the bottleneck shifts from 'how do I get data?' to 'how do I use it?' Luna: And that's a good problem to have. Thanks for listening to The Manufacturing Podcast.