Analog Gauge Reading

Industrial facilities run on thousands of analog gauges: pressure dials on steam systems, temperature indicators on heat exchangers, flow meters on pipelines. Operators walk rounds reading each gauge and logging values by hand. Replacing every analog instrument with a digital sensor is expensive and often impractical for legacy equipment.

Datature Vi takes a different approach. You photograph your gauges, label each photo with the correct reading, and train a model that learns to read those gauges from images. Operators snap a photo during their rounds, and the model returns the reading as a number, ready for your SCADA or historian system.

No hardware retrofit needed. If you can photograph the gauge, you can digitize the reading.

For an interactive overview of this application, visit the gauge reading use case on vi.datature.com.

Common applications

Choose your task type

Annotation examples

Tips:

• Photograph each gauge from directly in front to minimize parallax error
• Include gauges at different reading positions across their full range
• If your gauges have fogged glass or faded markings, include those conditions in training

Deploy and test

import json
from vi.inference import ViModel

model = ViModel(
    run_id="your-run-id",
    secret_key=".your-secret-key.",
    organization_id="your-organization-id",
)

result, error = model(
    source="gauge_photo.jpg",
    user_prompt="Read the gauges visible in this image.",
    generation_config={"temperature": 0.0, "do_sample": False}
)

if error is None:
    readings = json.loads(result.result)
    for r in readings["gauges"]:
        status = "OK" if r["in_range"] else "OUT OF RANGE"
        print(f"{r['gauge_id']}: {r['value']} {r['unit']} [{status}]")

Training tips

Photograph from the front: parallax (viewing a gauge from an angle) changes the apparent reading. Train on front-facing photos and instruct operators to photograph the same way.

Cover the full scale range: include images of gauges at low, mid, and high readings. A model trained only on mid-range readings will be less accurate at the extremes.

Include degraded conditions: real plant gauges have fogged glass, faded numbers, vibration blur, and condensation. Include these in training so the model handles them in production.

Label with numeric precision: "42.5 PSI" is more useful than "about 40 PSI." Match the precision to what a trained operator would record.

Next steps