Supervised learning-based classification of acoustic emission and vibration signal for identifying condition change of district heating system

Abstract

District heating pipes facilitate the distribution of heat energy across urban areas by utilizing high-temperature and high-pressure water. However, rigorous management is crucial due to the potential formation of sinkholes and the risk of human casualties. This study considered various pipe conditions and pipeline configurations and conducted pilot tests both above and below ground. Signals from the pipes were acquired using accelerometers and acoustic emission sensors, filtered based on specific frequency ranges, and compressed using statistical characteristic equations. The features extracted from the original and filtered signals were classified using a support vector machine. Individual classification models were developed and assessed under various conditions to compare their classification performance. The results indicated that the signals could effectively differentiate between abnormal pipe conditions in different states. The classification accuracy exceeded 90%, even without accounting for factors such as temperature, pressure, and pipeline arrangement. The experiments demonstrated a significant difference in signals between normal and damaged pipes.