The application of General large language models (LLMS) in the field of non-destructive testing (NDT) generally has many problems, such as inaccurate understanding of professional terms and difficulty in accurately adapting to dynamically updated regulations and standards. In response to these problems, the adaptation scheme for the lightweight field based on low-rank adaptive (LoRA) technology has been studied and evaluated. A dedicated dataset for the NDT domain containing over ten thousand high-quality data points was constructed. LoRA technology was applied to efficiently fine-tune base LLMS with parameter scales ranging from 7 to 9 billion (7-9B). The research results show that the optimized model has achieved significant improvements in evaluation metrics such as BLEU-4 and ROUGE-L. The above-mentioned models all demonstrated outstanding performance in the NDT professional knowledge Q&A session, successfully confirming that LoRA technology has good adaptability in the NDT field and proposing a technical path that is efficient in resource utilization and has practical application potential for the development of intelligent NDT systems. Future research will integrate cutting-edge technological achievements to further enhance the learning ability and practical application performance of the model.