Leak detection of pipes in underground pipe gallery is one of the important guarantees for the safe operation of pipe gallery. In order to realize the automatic operation of pipe gallery, and for the shortcomings of insufficient flexibility and poor information interaction of traditional detection methods, this paper designs a new method based on STM32F103. Rail-mounted pipeline leak detection robot. The robot is equipped with a MIC sound sensor as a leak detection device, and the collected signal uses a neural network to identify the audio features, and the monitoring host makes an alarm or other measures on the judgment result. After the test of the pipeline leakage detection system experimental platform, the robot system runs stably, and the detection rate of pipeline leakage reaches 93%, which meets the practical application requirements.