【作者】 刘越；

【导师】 王栋； 段超；

【作者基本信息】 南京理工大学 ， 航空工程（专业学位）， 2021， 硕士

【摘要】 固体火箭发动机装药的形状和几何参数可以确定发动机燃气生成率随时间的变化规律,进而决定发动机的工作压强和推力随时间的变化,所以装药设计是固体火箭发动机设计的核心内容之一。装药设计过程中隐含着大量经验知识,传统设计方法通过方案比较、论证来实现设计知识的重用,对设计人员的经验要求高且效率低。本文围绕构造专家系统的关键技术,结合固体火箭装药设计的基本理论,将人工神经网络技术引入到装药设计专家系统中,对其知识库和决策推理算法进行了研究和探索,并开发了相应的专家系统软件。首先,本文研究了适应神经网络模型的固体火箭装药设计知识表示方法,给出了单药型案例知识表示,并利用面向对象技术进行封装,实现了适应多药型的案例知识表示。将知识库以关系型数据库形式进行组织、存储与管理。其次,基于人工神经网络技术进行决策推理算法的研究。进行单药型神经网络结构和参数的设计,在燃面推移程序生成的模拟数据组成的样本中完成训练,依据训练结果调整网络结构与参数,优化模型的性能。根据专家系统的设计需求,进行适用多药型的组合模型神经网络结构设计,提出两种组合模型结构方案,并对其神经网络进行了设计和优化。依据装药设计实例的验证结果,各模型的预测误差基本满足装药方案初步设计的要求,表明人工神经网络在固体火箭装药设计专家系统中具有可行性和实用性。将专家系统用于辅助固体火箭装药设计,可减小装药设计对于专家经验的依赖性,降低对设计人员的经验要求。基于人工神经网络实现的决策推理算法,充分利用了已有装药设计知识,已优化的神经网络模型可快速计算推荐装药方案,得到较为准确的结果,提高装药设计效率和设计水平。最后,本文在Windows平台完成专家系统软件的开发,包含数据库管理、装药设计推理、模型再训练等。运用SQL Server完成数据库的创建,Python完成神经网络模型的训练与保存,C#完成人机交互界面的设计,实现了固体火箭装药设计专家系统软件的基本功能。更多还原

【Abstract】 The shape and geometric parameters of Solid Rocket Motor(SRM)grain can determine the change rule of gas generation rate in motor with time,and then determine the change of motor working pressure and thrust with time,so grain design is one of the core parts of SRM design.There is a lot of experience knowledge in grain design process.In traditional design,the reuse of design knowledge is realized through scheme comparison and theory,which requires high experience but low efficiency for designers.Based on the key technology of expert system construction and the basic theory of SRM grain design,this paper introduces the Artificial Neural Network(ANN)technology into the expert system of grain design,researches and explores the knowledge base and decision inference algorithm,and then develops the corresponding expert system software.Firstly,this paper studies the knowledge representation method of SRM grain design adapted to neural network model,gives the knowledge representation of single-grain type case,and realizes the knowledge representation of multi-grain type case by using object-oriented technology.The knowledge base is organized,stored and managed in the form of relational database.Secondly,the decision inference algorithm based on artificial neural network technology is studied.The structure and parameters of the ANN with the single-grain were designed,and the training was completed in the sample composed of the simulation data calculated by the combustion surface transition program.The network structure and parameters were adjusted according to the training results,and the performance of the model was optimized.According to the design requirements of expert system,two structural schemes of combination model are proposed,and their ANN model have been designed and optimized.According to the verification results of grain design examples,the predicted errors of these models basically meet the requirements of grain preliminary design,which indicates that ANN is feasible and practical in SRM grain design expert system.The expert system is used to assist the SRM grain design,which can reduce the dependence of grain design on expert experience and reduce the experience requirements for designers.The decision inference algorithm based on ANN makes full use of the existing grain design knowledge,the optimized ANN model can quickly calculate the recommended grain design scheme,obtain more accurate results,and improve the efficiency and design level of SRM grain design.Finally,this paper completes the development of expert system software on the Windows platform,including database management,charge design reasoning,model retraining,etc.SQL Server is used create the database,Python completes the design of human-computer interaction interface,which realizes the basic functions of the SRM grain design expert system software.更多还原