How Our Cobalt Leak Classification System Uses Data
Our Cobalt leak classification system, powered by machine learning and large language models (LLMs), takes raw water flow readings from meters and turns them into practical insights about where leaks may be happening. The approach differs depending on whether the meter is large (2 inches or bigger) or small (less than 2 inches).
Large Meters (2″ and Larger)
For larger water meters, Cobalt focuses on identifying when a leak starts, how long it lasts, and how much water is involved.
- Cobalt defines a leak as continuous flow for at least six hours.
- It records the smallest flow during the leak, the exact start and end times, and how many leaks have occurred over the past six months.
- This gives facilities a clear history of leaks without trying to label the exact source.
Small Meters (Less than 2″)
For smaller meters—typically at small businesses or homes—Cobalt goes a step further by using machine learning to identify the likely cause of the leak.
- How Cobalt learned: It was trained using leak data from customers who resolved issues reported in the EyeOnWater app. For example, if someone reported that their toilet was leaking, that labeled example was used to teach Cobalt what a “toilet leak” looks like in the data.
- What Cobalt looks at: It considers patterns such as how long the leak lasted, how steady the water flow was, when it started (time of year, day of the week), where the meter is located (region), and whether there were recent leaks nearby.
- How Cobalt classifies leaks: Using these patterns, Cobalt groups leaks into categories like toilet leaks, sprinkler leaks, hose leaks, pipe leaks, or simply water that was accidentally left on.
- Simplified view: For easy understanding, Cobalt also groups leaks into two broad categories: indoor leaks (like toilets or sinks) and outdoor leaks (like sprinklers or hoses).
Why This Matters
By combining user-provided “labeled” examples with smart algorithms, our Cobalt leak classification system can automatically suggest the most likely source of a leak. This helps utilities and customers:
- Save water by resolving leaks faster.
- Save money by reducing wasted usage.
- Reduce guesswork about where to start looking for the problem.
