乘坐飞机已经成为人们出行的重要选择之一，与汽车、火车及轮船等交通工具相比，飞机是快速且安全的，每百万次才可能发生一次事故。但由于其运行的特殊性，一旦发生事故都将十分严重。而我国的民航行业规模早已稳居世界第二，如此大数量大规模的民用飞机使得对其可靠性和安全性的提升提出了更高的要求。民机的结构振动情况作为影响民机可靠性和安全性的重要内容之一，一直都是民机设计部门重点关注的内容。通过对民机结构振动研究和与振动相关的预测研究的调研，发现当前研究缺乏民机结构振动响应的直接预测，且民机振动相关的预测研究方法均为单一预测方法。针对上述情况，本研究基于收集到的C919机型三个架次的完整试飞数据，考虑使用集成学习的混合预测方法来利用飞行参数预测民机结构振动响应，并开展相应原型系统的研制工作。本研究首先使用Python语言对所有试飞数据进行了预处理，包括所有数据文件的读取、变量名称的读取、监控时间的读取、监控数据的读取、数据的时间匹配和数据欠采样等内容。然后以皮尔逊相关系数作为衡量变量，分析了飞行参数与振动响应之间的相关性，为后续开展特征选择提供依据。在完成特征选择之后，考虑到飞行参数与振动响应采样频率的不同，为提高预测精度，使用快速傅里叶变换对数据进行了时频域变换。同时，应用滑动平均的方法对数据进行了降噪处理。本研究采用基于集成学习的方法来开展民机结构振动响应预测，包含线性回归、人工神经网络和支持向量回归三个个体学习器。将三个架次的试飞数据分别定义为训练数据集、验证数据集和测试数据集，其中的测试数据集用来验证预测结果的有效性，预测精度的衡量指标为均方根值误差。最后根据上述内容，结合HTML5语言和LayUI（前端）、PyCharm+Flask（后端）框架，研制了民机结构振动响应预测原型系统，介绍了原型系统的操作界面，并给出了民机机身、机翼、舵向和尾翼的部分测点预测结果，验证了原型系统的有效性。本研究旨在协助民机设计部门优化维修资源部署，研究成果可为民机的预防性维修提供应用借鉴。
Traveling by plane has been the choice of more and more people. Compared with cars, trains, and ships, planes are fast and safe, with an accident rate of one per million. However, because of the particularity of its operation, any accident will be quite serious. In the scale of civil aviation, China is the second-largest in the world. Such a large scale of civil aircraft puts forward higher requirements for its reliability and safety improvement. The structural vibration of a civil aircraft, which affects the reliability and safety of the civil aircraft, is the focus of the aircraft design department.This study investigated the research of structural vibration and the prediction related to the vibration of civil aircraft. The current research of civil aircraft structure lacked the direct prediction of the vibration response, and only individual methods of prediction were used. Given the above situation, this study considered using flight parameters to predict the vibration response of the aircraft structure based on the collected complete flight test data of three sorties of the C919. The hybrid prediction method of ensemble learning was adopted and the corresponding prototype system was developed.In this study, the Python language was used to preprocess all test flight data, including the reading of all data files, the reading of variable names, the reading of the monitoring time, the reading of the monitoring data, the time matching of data, and the under-sampling of data. The measured variable was the Pearson correlation coefficient. After the feature selection was completed, the time-frequency domain transformation was performed using Fast Fourier Transform to improve the prediction accuracy, taking into account the difference of the sampling frequency between the flight parameters and the vibration response. At the same time, the method of moving average was used to denoise the data.In this study, the ensemble learning method was used to predict the vibration response of aircraft structures. Individual learners are the linear model, the neural network, and the support vector machine. The test flight data of the three sorties were defined as training data set, verification data set and test data set respectively. Each data set represented the test flight data of one sortie respectively and did not influence each other. The test data set was used to verify the validity of the prediction results. The measurement index was the root mean square value error. Finally, the prototype system for the prediction of vibration response of civil aircraft structural was developed combined with HTML5 language, Layui (front-end), and PyCharm+Flask (back-end) framework. The operation interface of the prototype system was introduced. And the prediction results of some observation points of the aircraft fuselage, wing, rudder direction, and tail fin were given, which verified the effectiveness of the prototype system. The purpose of this study is to assist aircraft design departments to optimize the deployment of maintenance resources, and the research results can provide a practical reference for aircraft preventive maintenance.
