基于大模型的FADEC软件需求符合性检查方法  

Requirement Conformity Check Method for FADEC Software Based on Large Language Model

作　　者：李栋[1] 张星星 宫丽娜 熊波 魏明强 LI Dong;ZHANG Xing-xing;GONG Li-na;XIONG Bo;WEI Ming-qiang(Nanjing University of Aeronautics and Astronautics,Nanjing 211000,China;Aero Engine Control System Institute,AECC,Wuxi 214000,China)

机构地区：[1]南京航空航天大学,江苏南京211000 [2]中国航发控制系统研究所,江苏无锡214000

出　　处：《航空计算技术》2025年第1期88-92,共5页Aeronautical Computing Technique

基　　金：国家自然科学基金项目资助(62202223)。

摘　　要：全权限数字式发动机控制技术(FADEC)在航空发动机中得到了广泛应用,软件是FADEC的核心。软件的需求符合性检查可以在开发阶段发现错误,提高开发效率和软件质量。大语言模型的训练语料丰富,具有理解软件需求和代码的能力。以航空发动机FADEC软件的容错需求检查为例,引入大语言模型,利用其上下文学习能力,从软件需求提取信号处理规则;采用思维链提示思想,提取和翻译MBD模型的数据流;通过语义相似度来对处理规则和数据流配对,并采用动态提示词来增强大语言模型的逻辑判断能力,最终自动化实现对MBD模型的软件需求符合性检查。实验结果表明,基于大语言模型的方法在MBD模型的需求符合性检查上有效。Full Authority Digital Engine Control(FADEC)technology has been widely applied in aero engines,and software is the core of FADEC.The requirement conformity check of software can detect errors during the development stage,improve the development efficiency and software quality.Large language models have rich training corpora and the ability to understand software requirements and codes.Taking the fault-tolerant requirement check of aero engine FADEC software as an example,the large language model is introduced.Firstly,its in-context learning ability is utilized to extract signal processing rules from software requirements.Secondly,the idea of the chain-of-thought prompt is adopted to extract and translate the data flow of the MBD model.Finally,the processing rules and data flow are paired through semantic similarity,and dynamic prompt words are used to enhance the logical judgment ability of the large language model,ultimately achieving the software requirement conformity check of the MBD model automatically.Experimental results show that the method based on the large language model is effective in the requirement conformity check of the MBD model.

关 键 词：航空发动机 提示工程 控制系统软件 需求符合性 大语言模型 

分 类 号：V249[航空宇航科学与技术—飞行器设计]