# A camera or a phone photo reads the analog gauges, so inspection rounds are no longer written by hand (Nachi-Fujikoshi, Kobe Steel group, a plastics maker)

> AI reads analog gauges from a camera or phone photo: Nachi-Fujikoshi checks remote gauges from the office; a plastics maker cut gauge rounds 70%.

- Canonical URL: https://aiagentsformanufacturing.com/use-cases/ai-analog-gauge-reading-camera-inspection-rounds-nachi-kobe-steel/
- Department: Maintenance (https://aiagentsformanufacturing.com/departments/maintenance/)
- Evidence level: Verified. Real deployment at a named or precisely described company, with a measured result from a primary source.
- Added: 2026-10-27
- Last updated: 2026-10-27
- Publisher: AI Agents for Manufacturing (https://aiagentsformanufacturing.com)
- Manufacturers named: Kobe Steel (https://aiagentsformanufacturing.com/companies/#kobe-steel), Nachi-Fujikoshi (https://aiagentsformanufacturing.com/companies/#nachi-fujikoshi)
- AI tools deployed: hakaru.ai (GMO GlobalSign) (https://aiagentsformanufacturing.com/tools/#hakaru-ai-gmo-globalsign), LiLz Gauge (https://aiagentsformanufacturing.com/tools/#lilz-gauge), Solomon META-aivi (https://aiagentsformanufacturing.com/tools/#solomon-meta-aivi)

## Role

Maintenance (daily equipment checks and inspection rounds on analog gauges and meters). Typical users: maintenance technician, operator doing daily checks, maintenance supervisor, reliability or utilities engineer, quality or lab technician who records readings

## The problem

Most plants still have hundreds of analog gauges and small meters (pressure, temperature, current, filter differential pressure) with no data output. Someone walks the plant, reads each needle by eye, writes the value on a paper form and types it into a spreadsheet or system later. It is slow, tiring and error-prone, checks get skipped, and nobody can verify afterwards that the value written was the value shown. At Nachi-Fujikoshi's robot plant in Japan, daily start-of-shift checks covered many kinds of equipment (cranes, compressors, torque wrenches); a check could take 10 minutes and was repeated several times a day, some gauges were a 10-minute round trip from the office and others sat high up where the check itself carried a fall risk. At a plastics manufacturer, operators read analog and digital gauges every hour, wrote them on paper and then keyed them into the IT system; tired operators made reading and copying mistakes and lost forms left gaps in the records for audits.

## How AI helps

An image-recognition model reads the needle or the digits from a photo of the gauge and turns it into a number with a time stamp and the photo attached. There are two ways to take the photo. Fixed: a small battery-powered camera is stuck in front of a gauge and takes a few pictures a day (LiLz Gauge says three shots a day give about three years of battery life), the reading is done in the cloud and shows up on a screen in the office. Mobile: the technician keeps doing the round but takes a photo with a phone or tablet instead of writing the value, and the app reads it and uploads it to the database. Thresholds can then trigger an alert. The existing gauges are not replaced or rewired.

## Step by step

1. List the gauges on your daily and weekly rounds: location, what it measures, how often it is read, the normal range, and how long it takes to reach it.
2. Pick 5 to 10 gauges for a pilot: the ones far away, at height, in hot or unsafe areas, or read most often.
3. Choose fixed cameras for gauges read several times a day or hard to reach, and phone photos for the rest of the round.
4. For a few weeks, keep the manual reading and compare it with the AI reading on every gauge; note the ones the AI misreads (glare, angle, small digits) and fix the camera position or lighting.
5. Set the normal range for each gauge and decide who gets an alert when a value goes out of range, and what they do.
6. Once the readings match, replace the paper form for those gauges, keep the photos with the readings as proof, and extend gauge by gauge.

## Tools

- LiLz Gauge (battery-powered IoT cameras with cloud machine-learning meter reading): used at Nachi-Fujikoshi
- hakaru.ai by GMO GlobalSign (meter-reading AI from photos, used through its API in a Claris FileMaker app): used in the Kobe Steel group
- Solomon META-aivi (photo-based gauge reading on phones and tablets, analog and digital): used at the plastics manufacturer
- Other tools to evaluate: meter-reading apps on industrial camera platforms (for example Kyocera's Meter Reader on Actcast), the gauge-reading options of your CMMS or inspection-rounds app, and a smart transmitter where a gauge is critical enough to justify one
- For a first test: any LLM that reads images (ChatGPT, Claude, Copilot, Gemini) on a few gauge photos, with the prompt below

## Result

Nachi-Fujikoshi (machinery maker, consolidated sales of 258 billion yen), robot division plant, LiLz Gauge: cameras were installed first on the gauges that were a 10-minute round trip from the office and on checks at height; they now watch duct temperature (to prevent overheating), refrigerant pressure gauges and current readings, which staff check remotely from the office, detecting equipment problems earlier; the plan is to extend it to all equipment in the robot division; no time or cost figure published (LiLz customer interview, November 2023). Kobe Steel group, hakaru.ai (GMO GlobalSign) connected by API to a Claris FileMaker app, first deployed at Kobelco Research Institute, the group's materials and parts testing company, to record test and inspection readings from analog, digital and panel meters: about 20 users; each value is now stored with its photo, so a third person can check a reading that only the person measuring could confirm before; no more handwritten status notes or misread handwriting; old instruments kept in service. The meter-reading AI was reading 81% of values correctly at one point and was then retrained on the most-used meters, which the user says improved accuracy and efficiency, without a new figure; angles, glare and small decimal points made some instruments hard to read (vendor interview, information as of February 2022). A plastics and synthetic materials manufacturer (not named), Solomon META-aivi on phones and tablets: routine gauge inspections 70% faster than the previous manual process, readings uploaded automatically to its IT system and fewer recording errors (Solomon case study).

As reported by the source; most figures are vendor or self-reported and not independently verified.

## Human review: cautions

- A camera that reads a gauge is a monitoring aid, not a safety system: alarms that protect people or equipment must stay on proper instruments and interlocks.
- Readings can be wrong (glare, steam, dirty glass, camera angle, small decimal points; Kobe Steel saw 81% correct readings before the model was retrained): compare with manual readings before dropping the paper round, and keep the photo with each value.
- If a gauge matters a lot (critical pressure, regulatory record), a transmitter wired to the control system may be the better fix; cameras fit the many gauges that do not justify one.
- The figures are published by the vendors (LiLz, hakaru.ai, Solomon); Nachi-Fujikoshi and Kobe Steel give no time or cost saving, and the 70% comes from an unnamed customer.

## Try it yourself (prompt)

```text
You are a maintenance planner who organizes inspection rounds in a plant.

I will paste: (1) the list of gauges and meters on our rounds [tag or location, what it measures, unit, normal range, how often it is read, walking time to reach it, safety notes such as height or heat]; (2) a few photos of these gauges, if I have them.

Do this:
1. Sort the gauges into three groups: fixed camera worth testing (read often, far away, at height or unsafe), phone photo during the round, keep as is. Give the reason for each.
2. For each photo I attach, read the value, give the unit and a confidence (high / medium / low), and describe what makes the gauge hard to read (glare, angle, scale, needle hidden). If you are not sure, write 'not readable' instead of guessing.
3. Estimate the walking and writing time per week that the first group represents, using only the times I gave you.
4. Propose a 4-week pilot plan: which gauges, how we compare AI readings with manual readings, what counts as a good match, and who gets an alert when a value is out of range.

The AI reading must be checked against the manual reading during the pilot; do not treat it as a safety device. Use only the data I paste in; if something is missing, write 'not found' or list it as a question instead of guessing. Do not send, order, post or change anything in any system: produce drafts for my review and approval.

Gauge list:
[paste]
```

## Related use cases

- [A walking robot runs thermal and acoustic inspection rounds and turns what it finds into work orders (AB InBev, BMW, Nestlé Purina)](https://aiagentsformanufacturing.com/use-cases/robot-thermal-acoustic-inspection-rounds-ab-inbev-bmw/)
- [Predictive maintenance from vibration sensors, planned in the CMMS](https://aiagentsformanufacturing.com/use-cases/predictive-maintenance-vibration-sensors-cmms/)
- [Condition-monitoring alerts ranked by an expert so teams act on the right ones (Boise Cascade, Arbec, Kerry Group)](https://aiagentsformanufacturing.com/use-cases/prioritized-condition-monitoring-alerts-assetwatch/)
- [AI reads technicians' free-text work orders and assigns the failure codes, so the CMMS shows what really fails (Janssen Pharmaceutica, oil and gas operator)](https://aiagentsformanufacturing.com/use-cases/ai-failure-codes-from-maintenance-work-order-text/)

## Sources

- https://lilz.jp/interviews/8vgn500dic0o
- https://www.hakaru.ai/case/case_kobelco.html
- https://www.solomon-3d.com/case-studies/meta-aivi/automated-gauge-reading-ai-vision-ocr/

---
Please cite the canonical URL: https://aiagentsformanufacturing.com/use-cases/ai-analog-gauge-reading-camera-inspection-rounds-nachi-kobe-steel/