The rapid development of the service robotics industry has led to diversified module and product selections, but has also made it increasingly difficult to accurately match user requirements with appropriate modules. Existing service robot module retrieval and matching methods commonly suffer from strong manual dependency, low matching accuracy, and long response times, particularly evident when dealing with modules that appear similar but have significantly different technical parameters. How to bridge the semantic gap between professional terminology and everyday expressions, accurately assess inter-module compatibility, and efficiently handle complex matching of multi-functional requirements has become a critical challenge that urgently needs to be addressed in this field. To tackle these issues, this paper proposes a Knowledge-enhanced Adaptive Multi-source Retrieval (KAMR) framework for service robot module matching. By integrating domain knowledge graphs with multi-source retrieval technologies, this framework establishes a precise mapping mechanism from natural language user requirements to professional modules, capable of assessing query complexity and adaptively handling queries of varying complexity levels to provide high-quality module recommendations. Additionally, this study constructs an open-source service robot module description dataset containing 2,051 entries. Experimental results on this dataset demonstrate that the KAMR framework outperforms existing baseline methods across various query types from simple to complex and multiple application scenarios (multi-functional general, industrial warehousing, educational technology demonstration, and commercial service), particularly showing significant advantages in complex query processing.