Taiwan Water Corp adopts AI for leak detection

台灣自來水公司採用人工智慧進行洩漏檢測

Average experience of technicians has dropped from 20.9 years in 2018 to 7.2 years


TAIPEI (Taiwan News) — Taiwan Water Corp. is using AI-assisted acoustic leak detection equipment to enhance water leakage management.

The technology transmits the sound of each underground leak to the cloud, where AI analyzes and identifies distinct leak signatures, significantly improving detection accuracy and efficiency. Over the past 12 years, Taiwater has reduced the nation’s leakage rate from 19.55% to 11.99%, conserving the equivalent of five Hushan Reservoirs, according to CNA.

Taiwater President Lee Tin-lai (李丁來) emphasized that proactive leak control has been key to this achievement. In the past, over 90 senior technicians used headphones to listen for underground leaks, a method that could lead to hearing damage from prolonged exposure.

In 2018, the average experience of leak detection technicians was 20.9 years. However, with many senior technicians nearing retirement and training for new recruits taking over five years, the average experience has now dropped to 7.2 years.

To address this, Taiwater has partnered with the Industrial Technology Research Institute to incorporate AI into their operations. The result is an AI-powered system that uses the institute's leak sound recognition module to analyze and categorize leak signals based on different environments and pipe materials.

Lee explained that pipeline leak detection follows a two-year cycle, during which more than 60,000 km of pipeline is inspected. This process involves three key phases: “awareness,” “location,” and “repair,” with locating the exact leak being the most difficult part.

The AI system assists with initial leak screenings, allowing less experienced personnel to perform inspections. Professionals are then called in for confirmation when high-probability leaks are detected.

Taiwater invested NT$20 million (US$660,055) to develop this system, overcoming challenges posed by the company’s expansive service area, which includes outlying islands with varied pipeline materials, geological conditions, and background noise. To improve the system's accuracy, Taiwater has trained the AI model using data from over 300 real-world leak cases from various regions, enhancing its ability to operate in diverse environments.

Looking ahead, Taiwater plans to continue refining the AI model and is considering the introduction of drones to further support its leak detection efforts.