An Intelligent IoT and ML-Based Water Leakage Detection System

Abstract

Water is a precious resource, and much of it is wasted due to the leakage of pipelines. Timely identification of leakage could curb the wastage. Traditional leakage detection methods are time-consuming, inefficient and cause substantial water loss. Onsite, real-time leakage detection could reduce water loss and mitigate associated environmental and economic impacts. In this paper, we have proposed and developed an edge ML-based low-power IoT device to detect water leakage and notify the user. We developed the device in three stages. At first, an experiment was set up to capture real-life audio data of leak and non-leak signals. A piezoelectric contact microphone was used to capture the audio signals and to keep the unwanted environmental noise minimal. In the second step, an ML model was developed. The ML model was then quantized and pruned, resulting in a lightweight model of 11 KB in size with 98.96% accuracy. Finally, a node was implemented with an ML model and radio communication capability. If it detected any leakage, the node started beeping noise and broadcasting low-energy RF messages. The primary node could alert the user of potential leakage. These sensor nodes could be set up in the home or industrial environment. The device’s maximum current draw was 216.8 mA while transmitting data, and the minimum current draw was only 5.46 mA while sleeping. The node could run for more than 25 days on a single 3500 mAh battery.

Keywords

• Water leakage detection
• Internet of Things (IoT)
• machine learning (ML)
• sensor networks
• edge IoT