The preliminary study of acoustic feature-based detection method of inflow and infiltration in urban sewer systems

In urban sewer systems, inflow and infiltration (I/I) during rainfall can increase flow rates, potentially damaging pumping stations and affecting sewage treatment and public health. As infrastructure ages and extreme weather becomes more frequent, timely detection of I/I is critical. This study proposes an intelligent acoustic monitoring method to detect abnormal water intrusions, aiming to reduce the cost and deployment complexity of traditional monitoring systems while addressing the low responsiveness of manual inspections. A lab-scale pipe system was built to simulate various rainfall and flow conditions. The key variables included intruding water, flow rate, and pollution levels represented by suspended particles. In this experiment, a designed water tank is used as a controllable inflow that simulates different conditions. A low-cost water-proof recorder equipped with a MEMS microphone was used to capture sound in the sewer manhole at a 48-kHz sampling rate in the experiment. The recorded sounds were analyzed using empirical mode decomposition to extract intrinsic features, which were then classified using a naive Bayes algorithm. Preliminary results demonstrated anomaly detection accuracy exceeding 70% across diverse flow and pollution conditions, highlighting the method’s potential for cost-effective and responsive sewer system management.