Contact Us

Drop us a message and we’ll be in touch!

     

    Leak Classification

    Listening to your infrastructure

    Water leaks can be hard to hear

    React AI were contracted by our client to advise on water leak classification based on acoustic signals.  Our client has world-leading hydrophones which operate in groups and can accurately locate leaks by correlating the signals received across two hydrophones.

     

    Detecting whether a signal is a large or small leak, however, is another matter: leak noise is often viewed in the industry as effectively random, since it varies day-to-day with water pressure; and small leaks rushing through a small hole can be just as loud as large leaks flowing more slowly through a larger hole.  Our client wanted advice on using machine learning to classify the acoustic signals received: ideally directly to flow rate, though both pipe size and material offer useful guidance on which leaks are likely to be major.

    Services Provided

     

    React AI worked as embedded team members in their data-science department, operating within the client’s data-science environment and tooling to provide:

     

    • Pre-project workshop to review options and firm up approach.
    • Literature review.
    • Acoustic data preparation.
    • Guidance on and assistance implementing a range of approaches, including direct classification; building models to infer flow rates from individual readings; detecting time-based anomalies; general and region specifc models; and other approaches.
    • Training and general consultancy.
    • Reports on work undertaken, documentation of all results, conclusions and recommendations for future stages.