Answers are grounded in this paper and cite their sources.
Acoustic leak detection in water distribution networks (WDNs) is often hindered by complex acoustic environments and a lack of automated, evidence-grounded diagnostic capabilities. This study proposes a retrieval-augmented generation large language model (RAG-LLM) framework as an intelligent decision-support and reporting tool for acoustic leak detection. First, a feature vector database is constructed from historical pipeline acoustic cases. Second, relevant historical cases are retrieved from the database. Finally, the retrieved cases are incorporated into the prompt of an LLM to generate comprehensive leak-detection reports. Evaluations using samples from Dalian and Hong Kong indicate that this evidence-based retrieval mechanism compensates for the limitations of a standalone LLM in acoustic leak detection, achieving high recall, low false-positive rates, and performance comparable to specialized supervised learning models. The results further indicate that deep representations capture acoustic patterns more effectively than handcrafted domain features in the evaluated datasets. Overall, the proposed framework supports the automated generation of pipeline leak diagnoses and reports and provides a practical pathway toward smart water management systems that integrate advanced acoustic analysis with LLMs.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: