Controlled Infrared Thermography Dataset for Quantitative Nondestructive Evaluation of Electrical Cable Thermal States

Imagine you are a detective trying to solve a mystery, but instead of looking for fingerprints or footprints, you are looking for heat. In the world of electrical engineering, wires and cables are the silent workers that keep our lights on and our devices running. Sometimes, these cables get too hot, which can be a sign of trouble like a loose connection or too much electricity flowing through them. If we don't catch this early, it could lead to a fire or a blackout. To spot these heat problems without cutting the power or touching the wires, engineers use a special kind of camera called an infrared camera. Think of it like a pair of glasses that lets you see the invisible heat radiating off objects, turning temperature differences into colorful pictures.

For a long time, scientists have been trying to teach computers to look at these heat pictures and automatically say, "Hey, that wire is getting too hot!" or "That one is perfectly fine." But to teach a computer, you need a really good textbook full of examples. The problem is, most of the pictures we have so far are a bit messy. They might be taken in different places, with the camera at different angles, or they only show wires that are either freezing cold or burning up. Real life, however, is rarely that black and white. Wires usually get warm slowly, stay warm for a while, and then cool down gradually. To build a truly smart computer detective, we need a dataset that captures these in-between moments, taken under strict, controlled rules so every picture is fair and comparable.

This is exactly what Hidir Selcuk Nogay from Bursa Uludag University has created. The researcher built a special, highly organized collection of 628 infrared images of electrical cables, designed to be the ultimate "training manual" for computer programs. The goal was to move away from just looking for extreme emergencies and instead create a dataset that captures the full story of a cable's temperature life cycle.

To make this collection, the researcher set up a controlled laboratory experiment. They used a fixed camera position, hovering about 50 cm above the cables, to ensure every photo was taken from the exact same angle. They also controlled the environment to keep the background temperature steady and removed any confusing reflections. Then, they tested three different ways of arranging the cables: straight like a ruler, bent into a U-shape, and coiled up like a spring. This was done because the shape of a wire changes how heat spreads across it, just like how a crumpled blanket keeps you warmer than a flat one.

The real magic of this dataset, however, lies in the four specific "thermal states" or chapters of the cable's story that were captured:

1. Cold: The cable is just sitting there, doing nothing, at its normal room temperature.
2. Warm: The cable has been heated for 20 seconds. It's starting to get hot, but it's still in the early stages of warming up.
3. Hot: The cable has been heated for 60 seconds. It has reached a high, steady temperature, representing a fully stressed state.
4. Cooling: The heating was turned off, and the cable was allowed to sit for 40 seconds as it slowly lost its heat.

By including the "Warm" and "Cooling" stages, the dataset captures the tricky, in-between moments where the temperature is changing gradually. This is crucial because in the real world, a cable doesn't jump instantly from cold to hot; it transitions through these intermediate phases.

To test if this new dataset actually works, the researcher ran a simple experiment using a standard computer vision tool (a type of AI called GoogLeNet) to see if it could tell the four different states apart. The results were impressive: the computer correctly identified the thermal state of the cable 98.95% of the time. This high score suggests that the controlled way the photos were taken created very clear, distinct patterns for the computer to learn.

Interestingly, the computer did make a few small mistakes, mostly confusing the "Warm" state with the "Hot" state. The paper explains that this isn't a failure of the dataset, but actually a sign that the dataset is realistic. Since "Warm" and "Hot" are just steps in a continuous heating process, it makes physical sense that they look very similar to each other. The dataset successfully preserved this natural overlap, which is exactly what a good training tool should do. On the other hand, the "Cold" and "Cooling" states were perfectly distinct, showing that the computer could tell the difference between a cable that was never heated and one that was cooling down after being hot.

In short, this paper doesn't just offer a new way to fix wires; it offers a new, high-quality, and fair set of examples for scientists to use. It fills a gap in the research world by providing a reproducible, balanced collection of images that includes the subtle, gradual changes of temperature, not just the extreme emergencies. This gives researchers a solid foundation to build better, more reliable systems for monitoring electrical safety in the future.