试飞遥测噪声数据升维与神经网络分析技术研究

Research on Uplifting Flight Test Telemetry Noise Data and Neural Network Analysis Techniques

摘要:

试飞噪声是飞行试验过程中舱内外的噪声集合。在新型号试验试飞转场频繁、构性差异大且重大改装密集的情况下,试验过程中的噪声需要在飞行后对数据解析才能对其成因和机理进行分析,分析周期较长,对试飞噪声的定位较为困难,也增加了排故试飞架次数。本研究提出基于DNT(Data dimensionality enhancement—neural network technology)数据升维神经网络的试飞噪声识别系统,对已做分析的噪声数据与分析结果构建映射关系,设计数据升维模块并设计神经网络进行拟合训练,形成对噪声数据敏感的网络模型,实现对噪声信号的快速识别及分类。对5类噪声进行了多次识别,结果表明DNT网络模型识别平均精度为83.2%,查全率为95%,具有较好的识别效果。同时构建网络模型优化评估策略,根据不同外场试飞任务输入不同网络模型参数并获得网络性能评估结果,获取更好的适配网络和更准确的识别结果。

Abstract:

Flight test noise encompasses the collection of noise both inside and outside the cabin during flight testing. With frequent transfers during flight testing of new models, significant structural differences, and intensive major modifications, the noise generated during the testing process requires post-flight data analysis to determine its causes and mechanisms. This leads to a long analysis cycle, making it difficult to pinpoint the source of flight test noise and increasing the number of troubleshooting flight sorties. This research proposes a flight test noise identification system based on DNT (Data dimensionality enhancement—neural network technology) data upscaling neural network. It constructs a mapping relationship between analyzed noise data and analysis results, designs a data upscaling module, and designs a neural network for fitting and training, forming a network model sensitive to noise data, thereby achieving rapid identification and classification of noise signals. Multiple identifications were performed on five types of noise, and the results showed that the average accuracy of the DNT network model was 83.2%, and the recall rate was 95%, demonstrating good identification performance. Simultaneously, a network model optimization and evaluation strategy is constructed, inputting different network model parameters according to different field flight test tasks and obtaining network performance evaluation results to acquire better-adapted networks and more accurate identification results.