In irrigation systems, flow monitoring is the practice of continuously measuring the volume of water passing through a line per unit time and recording any deviation from the expected value. Visible symptoms (yellowing, wet patches) are typically lagging indicators; flow data makes it possible to detect a fault before the loss escalates.

Flow sensor technologies

Three measurement technologies are widely used in irrigation:

  • Turbine / paddle (mechanical) — low cost, ±2–3 % accuracy; sensitivity falls with particulates and at low flows.
  • Electromagnetic (mag-meter) — based on Faraday's law, with no moving parts, ±0.5 % accuracy; requires a conductive fluid (≥5 µS/cm).
  • Ultrasonic (transit-time / Doppler) — non-intrusive (clamp-on) or in-line, ±1–2 % accuracy; broad diameter range.

Mechanical meters typically provide a pulse output (litres per pulse); electromagnetic and ultrasonic meters offer 4–20 mA, Modbus RTU/TCP or HART interfaces. Relevant standards include ISO 4064 (water meters), OIML R49 and the AWWA C700 series.

Baseline learning

For anomaly detection, an expected flow profile is established for each line or zone. The profile may be calculated from nominal pipe diameter, number of outlets (sprinklers, drippers), nozzle flow coefficients and line pressure, or it may be learned by the device from the statistical distribution (mean, standard deviation, interquartile range) of values observed during the first few weeks. The baseline is updated dynamically with seasonal, temperature and pressure variables.

Deviation thresholds and classification

Flow deviations are commonly classified into three main categories:

  • Under-flow — 20–40 % below expected: filter fouling, scale deposits, root intrusion, partial blockage, low line pressure, sprinkler failure.
  • Over-flow — 30 %+ above expected: pipe burst, gasket leak, broken sprinkler body, tampering.
  • Passing valve — flow >0 when the valve should be closed: solenoid sticking, worn diaphragm, burred seat.

Thresholds are usually combined as fixed percentages, absolute values (L/min) and statistical z-scores. A typical rule: an alarm is raised when a 3 σ deviation persists for more than 30 seconds.

Leak-detection algorithms

The principal industry approaches are:

  • Minimum night flow (MNF) — the standard method in water utilities; persistent flow during the lowest-consumption hours indicates a leak.
  • Mass balance — the difference between the master meter and the sum of sub-line meters.
  • CUSUM and EWMA — cumulative control charts that catch small but persistent deviations early.
  • Flow signature analysis — the shape of the opening transient (ramp, plateau, close) helps distinguish the type of fault: a sudden drop suggests a sprinkler/valve issue, while a gradual decline points to filter clogging.

Reducing false positives

Under field conditions, pressure fluctuations, brief air pockets, wind-induced sprinkler deflection and simultaneous valve openings can trigger false alarms. Practical mitigations:

  • A 10–30 second debouncing window before any alarm.
  • Cross-validation with a pressure sensor (if flow drops, pressure should rise).
  • Variable thresholds by season and time of day.
  • Combining multiple statistical tests using AND logic.
Practical impact: in large landscape areas, a single unnoticed pipe break can lose dozens of cubic metres of water overnight. Continuous flow monitoring and threshold-based automatic shut-off open a response window before the loss grows and make it easier to prioritise field-team interventions.

Data model and standards

Flow data is typically stored with timestamp, line identifier, instantaneous flow (L/s or m³/h), cumulative volume, pressure and valve state. Common industrial communication layers are Modbus, OPC UA (IEC 62541) and MQTT. For reporting water losses, the IWA (International Water Association) Water Balance framework is the reference; real losses and apparent losses are tracked in separate categories.