Water pipeline leakage is a common and significant global problem. In-pipe inspection based on hydrophone is one of the most direct, accurate, and reliable solutions for leak detection and recognition. In this study, a scheme of in-pipe detector was designed to pick up and identify acoustic signal due to leak. To investigate the characteristic of acoustic signal, an experimental platform was built to simulate the leaks and obtain acoustic signals under different leak conditions in an industrial scale water pipeline. Because a decreased pressure as leak has an unstable fluctuation in time domain, the frequency composition of the signal was analyzed in frequency domain, and then the change of frequency amplitude can be referenced to recognize the leaks. Moreover, the effects of leak size, pipeline pressure, and water flow rate on the characteristic of acoustic signal were investigated. The results show that the signal’s intensity under leak conditions are significantly higher than that of no leak case, and it will increase as the increased leak size; the signal intensity under no leak case will increase with the growth of pipeline pressure; the flow velocity has little effect on the signal intensity. To increase the recognition accuracy, an artificial neural network model was developed for the leak prediction, and 18 cases through additional tests were selected to validate the accuracy of model. Comparing experimental and prediction results, maximum relative error is within 10.0%. It indicates that the prediction model has a reasonable accuracy for the leak recognition.

输水管道泄漏是一个普遍的、重大的全球性问题。基于水听器的管道内检测是泄漏检测和识别最直接、准确、可靠的解决方案之一。在这项研究中，设计了一种管道内检测器方案来拾取和识别由于泄漏引起的声学信号。为研究声信号的特性，搭建了一个实验平台，对工业规模输水管道中的泄漏进行模拟，获取不同泄漏条件下的声信号。由于泄漏引起的压力下降具有时域不稳定的波动，因此在频域分析信号的频率成分，然后可以参考频率幅度的变化来识别泄漏。此外，泄漏尺寸、管道压力、研究了水流量对声信号特性的影响。结果表明，泄漏条件下的信号强度明显高于无泄漏情况，并且随着泄漏尺寸的增加而增加；无泄漏情况下的信号强度会随着管道压力的增加而增加；流速对信号强度影响不大。为提高识别精度，开发了人工神经网络模型进行泄漏预测，并通过附加测试选取了18个案例来验证模型的准确性。比较实验和预测结果，最大相对误差在10.0%以内。这表明预测模型对于泄漏识别具有合理的准确性。结果表明，泄漏条件下的信号强度明显高于无泄漏情况，并且随着泄漏尺寸的增加而增加；无泄漏情况下的信号强度会随着管道压力的增加而增加；流速对信号强度影响不大。为提高识别精度，开发了人工神经网络模型进行泄漏预测，并通过附加测试选取了18个案例来验证模型的准确性。比较实验和预测结果，最大相对误差在10.0%以内。这表明预测模型对于泄漏识别具有合理的准确性。结果表明，泄漏条件下的信号强度明显高于无泄漏情况，并且随着泄漏尺寸的增加而增加；无泄漏情况下的信号强度会随着管道压力的增加而增加；流速对信号强度影响不大。为提高识别精度，开发了人工神经网络模型进行泄漏预测，并通过附加测试选取了18个案例来验证模型的准确性。比较实验和预测结果，最大相对误差在10.0%以内。这表明预测模型对于泄漏识别具有合理的准确性。无泄漏情况下的信号强度会随着管道压力的增加而增加；流速对信号强度影响不大。为提高识别精度，开发了人工神经网络模型进行泄漏预测，并通过附加测试选取了18个案例来验证模型的准确性。比较实验和预测结果，最大相对误差在10.0%以内。这表明预测模型对于泄漏识别具有合理的准确性。无泄漏情况下的信号强度会随着管道压力的增加而增加；流速对信号强度影响不大。为提高识别精度，开发了人工神经网络模型进行泄漏预测，并通过附加测试选取了18个案例来验证模型的准确性。比较实验和预测结果，最大相对误差在10.0%以内。这表明预测模型对于泄漏识别具有合理的准确性。比较实验和预测结果，最大相对误差在10.0%以内。这表明预测模型对于泄漏识别具有合理的准确性。比较实验和预测结果，最大相对误差在10.0%以内。这表明预测模型对于泄漏识别具有合理的准确性。