Flow sensors are often presented as a simple way to detect water leaks. Install a meter, watch the water flow and generate an alert when something looks wrong. In real projects, it is not that simple.
The main problem is selectivity. A flow sensor knows that water is moving through a pipe, but it does not know why. The same flow can come from a shower, a washing machine, a toilet filling, somebody running a bath or a real leak. The more water consumers you put behind one meter, the harder it becomes to reliably say that a leak has happened.
This is the first thing we look at when we design a flow-based leak detection system.
Selectivity is the main problem
Flow-based leak detection works best when the meter monitors something with a simple and predictable water use pattern. A toilet is a good example.
We use flow monitoring on toilet water lines because normal behavior is easy to understand. Water flows while the tank is filling, and then it stops. If the flow does not stop, or if it starts again much more often than expected, there is probably a problem.
This can detect something that a normal floor leak sensor may never see. A leaking toilet can waste water for weeks while keeping the floor completely dry, because the water goes directly into the drain.
The same approach becomes much harder when one meter monitors a complete apartment. At that point the system sees every water consumer through the same measurement point, and the signal becomes much less clear.
Why flow detection works poorly for a whole apartment
A lot of residential flow detection systems use one basic rule: how long has the water been flowing without stopping?
This can work, but there is always a compromise. If you set the time too short, normal water use creates false alarms. A long shower, filling a bathtub or another normal event can look like a leak. If you set the time too long, the alarm comes late.
For a large leak, waiting 20, 30 or 60 minutes before generating an alarm can mean that a lot of water has already escaped.
This is why we do not see a whole-apartment flow meter as a direct replacement for point or rope leak sensors. A point sensor detects water where water should not be. A flow sensor detects water use that may be unusual. These are different things.
Small leaks are difficult to detect
Small leaks create another problem. Every meter has a minimum flow that it can measure reliably, and this parameter is very important for leak detection.
A meter can be accurate at normal household flow and still perform poorly when only a very small amount of water is moving. If the meter cannot see low flow, the software cannot detect it either.
There is also a problem with normal apartment usage. A small leak can disappear inside the normal flow from a faucet, shower or washing machine. From one measurement point, it may be impossible to separate a very small leak from normal water use.
This is one reason why claims that one flow meter can detect every type of leak in an apartment should be treated carefully.
Flow detection is usually slower than a direct leak sensor
Reaction time is another important difference.
If a hose under a dishwasher comes loose, a rope sensor under the dishwasher can generate an alarm as soon as water reaches it. A flow-based system first needs to decide that the water use is unusual.
If the rule is based on continuous flow for a certain amount of time, the system has to wait before it can make that decision. For some types of leaks this is fine. For leaks that can quickly damage floors, walls or apartments below, direct water detection is usually faster.
This is why we see flow monitoring and direct leak sensors as technologies that solve different problems.
Where flow monitoring works well
There are several cases where flow monitoring makes a lot of sense.
The first is a selective water line with predictable behavior, such as a toilet. The second is consumption monitoring in large buildings. The third is utility and city water networks.
At building or utility scale, the objective changes. We are no longer trying to find water under a dishwasher. We are looking for changes in water consumption across a building, district or network.
For a large property, even simple consumption monitoring can be useful. If a building normally uses a certain amount of water and consumption suddenly increases, that is something worth checking. The system can also look at nighttime consumption, long periods of continuous flow or changes in the normal daily profile.
Most building projects do not need machine learning
For normal building projects, we usually start with simple rules. They are easier to understand, easier to test and easier to change.
For example, we can generate an alert when flow continues longer than expected, when nighttime consumption is too high or when daily consumption increases significantly compared with the normal level.
There is also an operational advantage. When the maintenance team receives an alert, we can tell them exactly why it was generated.
For example, the message can say that water has been flowing continuously for 45 minutes, or that nighttime consumption is 70 percent above the normal level.
This is much more useful than showing an unexplained anomaly score.
For residential and commercial buildings, simple rules are often enough.
When we use forecasting and anomaly detection
The situation is different for water utilities and city networks.
There we have much more data, larger networks and more stable consumption patterns. We are also not trying to understand the behavior of one person in one apartment.
For these projects, forecasting and anomaly detection become much more useful. We have worked with models based on LSTM, Isolation Forest and Prophet for this type of application.
The basic idea is to build an expected consumption profile and compare it with actual water use. If a district normally has one consumption pattern and current usage is significantly higher than expected, the system can flag that area for investigation.
This does not automatically mean that a pipe is broken. It means that something unusual is happening and the utility team knows where to look.
At utility scale, that is already very useful.
Why the same approach does not work as well for one apartment
A model needs useful and relatively stable data.
For a city network, there can be thousands or millions of measurements and clear patterns across days, weeks and seasons. One apartment is very different.
One person may live there today and four people may stay there next week. Somebody can take a 10-minute shower one day and a 40-minute shower the next. The apartment can stay empty for a week and then suddenly have high water use.
Technically, it is possible to build models around this data. The real question is whether the added complexity gives a better leak detection system.
In many residential projects, it does not. We prefer a simple rule that is easy to explain over a complicated model that still produces false alarms.
Hardware selection still matters
The software cannot detect something that the meter cannot measure.
For flow-based leak detection, we pay particular attention to low-flow performance. The maximum flow rate is important for sizing the meter, but it tells you very little about how good the meter is at finding small leaks.
You need to check the minimum flow the meter can measure and how accurate it is close to that limit.
Communication is another practical point. For a centralized building system, the meter needs to send data to the main platform. Depending on the project, this can be Modbus, pulse output, M-Bus, LoRaWAN or another interface.
We do not choose a meter only because it supports a specific protocol. Measurement quality, installation cost and the ability to get useful data at the right frequency are more important.
Installation location defines what you can detect
A meter at the main building inlet tells you what is happening with the whole building. A meter on one apartment tells you what is happening in that apartment. A meter on one toilet tells you what is happening with that toilet.
This sounds obvious, but it is one of the most important parts of the design.
If you want better leak location, you need more selective measurement points. There is always a balance between hardware cost and how accurately you want to locate the problem.
One meter for the whole building is cheap, but a leak can be anywhere. One meter for every apartment gives much better information, but the project becomes more expensive. Meters on individual high-risk lines give even better selectivity, but installing them everywhere usually does not make economic sense.
We normally choose measurement points based on the actual problem we are trying to solve.
Flow monitoring and direct leak sensors solve different problems
We do not normally choose between flow monitoring and leak sensors by asking which technology is better.
We look at what we need to detect.
If the concern is water leaking under a dishwasher, refrigerator or air conditioner, we normally use a point or rope sensor. If the concern is a toilet continuously wasting water, a flow sensor can be much better.
If the objective is to understand whether the total water consumption of a large building is changing, flow monitoring is the right tool. If the objective is to detect anomalies across a city water network, we can add forecasting and anomaly detection on top of the flow data.
Different problems need different tools.
What should an alert show?
A centralized system should not just say that there is a possible leak. The operator needs enough information to understand the event.
For a selective line, the alert can say that Apartment 802, Toilet 2 has had continuous water flow for 18 minutes. For a building-level meter, it can say that nighttime consumption is 42 percent above the normal level. For a utility network, it can say that current consumption in Zone 14 is significantly above the expected profile.
The important part is that the alert explains why the system thinks something is wrong.
This makes troubleshooting much easier and also helps the building team trust the system.
Automatic shutoff needs more confidence
Flow monitoring is often combined with an automatic shutoff valve. Technically, this is easy. Operationally, it needs to be treated carefully.
A false alert is annoying. A false alert that automatically turns off somebody's water is a much bigger problem.
This is especially important for apartment buildings, where normal water use can vary a lot. Before using automatic shutoff, we want to know that the detection rule is stable and that false alarms are very rare.
For simple and selective lines, this can be easier. For a complete apartment, we would be much more careful.
Automatic shutoff is a separate topic and we will cover it in more detail in another article.
What we use in practice
Our approach depends on the scale of the project.
For a simple and predictable water line, we use direct rules. Toilet monitoring is a good example because normal water flow is easy to define.
For residential and commercial buildings, we normally use hardcoded alerts based on flow duration, consumption limits or changes in expected water use.
For water utilities and city networks, we can use forecasting and anomaly detection models such as LSTM, Isolation Forest or Prophet because there is enough data for them to be useful.
We do not add machine learning just because it sounds more advanced. The detection method needs to match the problem.
The main rule
The most important thing we have learned about flow-based leak detection is that selectivity matters more than the algorithm.
The more different water consumers there are behind one meter, the harder it becomes to say what is normal and what is a leak.
A flow sensor can be a very useful part of a water monitoring system, but it needs to be installed for a clear reason.
For a toilet, the question is simple: why is water still flowing?
For a building, the question is different: why has consumption changed?
For a city network, the question becomes: where is actual water use different from what we expected?
Once that question is clear, choosing the meter, the rules and the software becomes much easier.
