Early underground pipeline collapse detection and optimization based on water hammer vibration signal

Ground collapse of buried pipelines is a common accident. Early-stage collapse of buried pipelines often manifests as bottom collapse, surface exposure, or complete suspension. Through theoretical analysis and COMSOL finite element simulation model, it was found that the early-stage collapse type and length have varying degrees of influence on the attenuation of water hammer wave amplitude and the characteristic frequency of the pipeline. Based on the above analysis, this paper proposes a method to use the water hammer vibration wave generated by valve closure in flowing pipelines as a detection excitation source to identify early-stage collapse types. In the experiment, 3000 sets of vibration data were collected and 5 average time-domain features were extracted. At the same time, wavelet packet technology was used to decompose and reconstruct the frequency domain sensitive band of the pipeline, and 5 frequency-domain features such as centroid frequency were extracted. These features were combined to create a new training set. The BP neural network machine learning model was trained and the parameters were adjusted by genetic algorithm to find the optimal parameters and training effect. Genetic algorithm was used to optimize the BP neural network, and the test accuracy reached 96.97%. By using the water hammer in the pipeline as an excitation source to detect early-stage collapse, the collected signals were extracted and processed, the dimensionality of the training dataset was reduced, and the neural network input layer was set to 10, thereby reducing computational complexity and significantly improving the recognition accuracy. This model can be used for early-stage collapse and type identification and can be combined with climbing robots for real-time monitoring, providing a theoretical basis for pipeline safety.