This research addresses the issues of low efficiency, limited interaction, and high dependence on specialized knowledge in NDT operations. It designs and implements an intelligent operation management system integrating dual-modal (speech-text) interaction and RAG technology. The system employs a layered architecture, integrating a natural language processing engine and a dual knowledge base framework. Through intent recognition, structured data extraction, and multi-strategy knowledge retrieval, it enables real-time voice input of inspection data, querying of standard clauses, and intelligent report generation. Efficient human-machine interaction is achieved via a portable voice terminal. Experimental results show the system improves on-site data entry efficiency by approximately 6.1 times compared to traditional methods, achieves an 88% recognition success rate in high-noise environments, and enhances report generation efficiency by 63.8% in office environments. User satisfaction surveys indicate consistent approval of the system's usability, functionality, and interactive experience. This system effectively lowers the technical barriers for NDT field operations and report compilation, enhances operational efficiency and data reliability, and meets the application demands for intelligent and on-site industrial inspection.