ROI Case File No.502 'The Twenty-Year-Old Optical Sensor That Missed the Breakage'
The Twenty-Year-Old Optical Sensor That Missed the Breakage
Chapter One: Every Time It Stopped, the Machine Was Idling
"The optical sensor flags things that aren't drill breakages as drill breakages."
Kohei Katase, Manufacturing Team Lead at TechInnovations, stood in front of the machining station. Drills about one millimeter in diameter were rotating in parallel. "Metal powder, coolant spray, slight vibration—anything that blocks the light triggers a false alarm. Every time, the machine stops and an operator runs over to check."
"How often are misdetections occurring?" Claude asked.
"Twelve times a day on average," Katase replied. "Each check takes an operator five to ten minutes. If a drill actually broke, response is necessary—but 90% are false alarms. And the remaining 10% of real breakages sometimes get missed by the optical sensor. The most dangerous case is when machining continues with a broken drill."
"The sensor itself has been in use for twenty years, as I understand it," I confirmed.
"It was state-of-the-art at the time," Katase replied. "But the workpieces have shrunk in diameter, and the detection environment has changed. We looked at overseas-made cutting-force sensors, but the size didn't fit and we gave up. At one millimeter drill diameter, even compact off-the-shelf sensors are too big. On top of that, multiple drills are driven by a single motor, so detecting individual drills by current draw is also difficult."
"This isn't just a sensor problem—it's also an operational design problem," I replied. "Let's redesign the whole thing with JOURNEY."
Chapter Two: JOURNEY Asks—The Path From Detection to Recording
"This case calls for JOURNEY."
Claude wrote seven letters on the whiteboard. J, O, U, R, N, E, Y.
"JOURNEY is a framework for decomposing the flow a user or object travels through into stages, and visualizing the experience and problems at each stage," I explained. "It was originally a customer journey technique, but it also applies to detection processes on the factory floor. A drill is installed, machining begins, an anomaly occurs, it's detected, it's responded to, and it's recorded—decompose this journey and you'll see it isn't a problem with the sensor alone but with the design of the entire flow from detection to recording."
"First, let's measure the current cost," Gemini said, opening ROI Polygraph. He input the machining data Katase had provided.
"The monthly detection cost has come out," Gemini read. "Operator labor for handling misdetections averages 70 hours per month at 2,800 yen per hour, or 196,000 yen monthly. Production losses from machine stoppages due to misdetections average 1.8 million yen monthly. Defective-product costs from missed real breakages average 600,000 yen monthly—machining continuing with a broken drill produces defects that must be scrapped. Drill-life prediction isn't possible, causing under-replacement (or over-replacement) costs averaging 400,000 yen monthly. Drill breakage data sits on paper and is never analyzed, an opportunity cost averaging 200,000 yen monthly. Total: 3.196 million yen monthly. Annualized: approximately 38.4 million yen."
Katase stared at the figures. "I had been thinking misdetections were just a hassle on the floor."
"Now let's design with JOURNEY," I continued.
[J·O—Journey Start: Installation and Operation Begin]
"We'll examine the first two stages," Claude said. "The drill is installed, and machining begins. At present, only the installation timestamp is recorded. Which operator installed which drill in which position, what the machining conditions were at start—if these were recorded, later analysis would be possible. We'll have operators scan a QR code at installation that automatically captures machining conditions."
[U·R—Use and Response: Anomalies and Response During Use]
"We'll change the detection method during use," Gemini continued. "From optical sensors to force sensors. When a drill breaks, the cutting reaction force drops instantly to zero. Detecting that change directly is structurally far less prone to false positives. Small force sensors compatible with one-millimeter diameter have become commercially available over the past few years."
"What about the issue of multiple drills running off one motor?" Katase confirmed.
"We install sensors at each individual drill position," I answered. "Instead of overall detection by current draw, each position's reaction force is measured independently. The installation footprint will fit within the existing optical sensor location."
[N—Notify: Notification Design]
"At the notification stage, we cut false detections with a two-stage check," Claude continued. "The moment the force sensor reacts, we automatically capture the most recent three seconds of machining data and re-evaluate it on three axes—vibration, reaction force, and rotation speed. Force sensor for primary judgment, pattern analysis for secondary. Two layers raise the confidence and further reduce false alarms."
[E—Evaluate: Evaluation and Feedback]
"At the evaluation stage, we build in drill life prediction," I continued. "We accumulate data on machining count, machining conditions, and reaction force trends from installation to breakage. Once enough data is collected, we build a life-prediction model. Optimal preventive replacement timing becomes visible."
[Y—Yield: Recording and Continuous Improvement]
"Finally, we design the recording stage," Gemini continued. "Move paper records to digital, with detection data automatically accumulating in a database. Monitor misdetection rate, missed-breakage rate, and prediction accuracy monthly. The more data accumulates, the higher detection accuracy climbs."
[Estimating the Payback]
"Let's run it through ROI Proposal Generator," Gemini proposed.
- Initial cost: 4.8 million yen (force sensor installation, pattern analysis platform, QR identification system, database build, on-site training)
- Monthly cost: 120,000 yen (sensor maintenance and analysis platform fees combined)
- Monthly savings: misdetection handling reduction = 160,000 yen; misdetection-driven production loss reduction = 1.5 million yen; defect cost reduction from fewer missed breakages = 480,000 yen; replacement cost optimization via life prediction = 280,000 yen. Total: 2.42 million yen monthly
- Net monthly savings: 2.42 million − 120,000 = 2.3 million yen
- Payback period: 4.8 million yen ÷ 2.3 million yen ≈ 2.1 months
"Just over two months for payback," Gemini summarized. "What's particularly large is the production loss reduction. Misdetection-driven machine stoppages may be short individually, but they accumulate. We were losing about 1.5 million yen monthly."
Katase looked at the numbers. "I thought this was about replacing a sensor. Viewing it as a journey from detection to recording, the investment target looks completely different."
"Discussing the sensor in isolation is discussing one stage of the journey," I replied.
Chapter Three: Redesigning Detection by the Journey
"Here's the rollout plan," I said, standing at the whiteboard.
"Week 1: select force sensors, obtain samples. Weeks 2–3: pilot installation on existing machines, verify reactions under machining conditions. Weeks 4–5: develop the pattern analysis algorithm, implement two-stage judgment logic. Week 6: build the QR identification system and automatic machining-condition capture. Week 7: design the database, migrate paper records to digital. Weeks 8–9: run in parallel, measure misdetection rate and missed-breakage rate. Week 10: switch over from optical sensors, run the new method standalone."
"Do we keep the existing optical sensors?" Katase confirmed.
"For the first three months, yes," Claude replied. "Run them in parallel to confirm force sensor accuracy, then remove the optical sensors once stable. Sudden switching carries too much risk."
Katase made a note. "I understand now why I couldn't bring myself to remove a sensor we'd used for twenty years. I wasn't looking at a comparison."
Chapter Four: The Day Data Told Us Before It Broke
Seven months later, a report arrived from Katase.
Three months after the force sensor went live, misdetections dropped 91% versus baseline. From twelve a day on average, it fell to about one. "Zero misdetections is impossible, but a drop from double digits to roughly one is a level the floor experiences as a complete change," Katase wrote.
Real-breakage detection improved too. The minor breakages the optical sensor couldn't catch were reliably picked up by the reaction-force drop. "The case of machining continuing on a broken drill became structurally impossible," the report said. Defect rates fell roughly 60%.
The most surprising change appeared in drill life prediction accuracy. After three months of accumulated installation-to-breakage data, the life prediction model came online. Preventive replacement timing was optimized, cutting over-replacement by about 30%. "Annual drill cost dropped by an amount well beyond what we had projected," Katase wrote.
As a side effect, the machining data itself became an internal asset. With reaction force, vibration, and rotation speed accumulated on three axes, machining-condition optimization talks began. "Data collected for breakage detection turned out to be usable for machining-quality improvement discussions," the report said.
The twenty-year-old optical sensors completed their role at week 10 and were removed. "At the final inspection, the optical sensors themselves weren't broken. It wasn't that the technology was old—it was that the way of using it was old," Katase wrote.
The final line of the report read: "Replacing the sensor was never the goal. Redesigning the journey from detection to recording was the goal. The moment we decomposed it with JOURNEY, it stopped being a replacement conversation."
The day when data told us a drill's life before it broke became ordinary, the report said.
"Misdetections look like a sensor problem, but they're also an operational design problem. Installation, machining, detection, notification, evaluation, recording—unless you decompose what is happening at each stage of the journey, the conversation ends at replacement. JOURNEY asks for a line, not a point. View detection through recording as a line, and you'll see the structure that turns data into an asset. What the twenty-year-old optical sensor was missing wasn't drill breakages but the design to accumulate data. The day data started warning before drills broke, what the floor was watching was the machining itself."
Related Files
Tools Used
- ROI Polygraph — Visualizing misdetection labor, production loss, and life-prediction opportunity cost
- ROI Proposal Generator — Payback simulation for migration to a force-sensor foundation
