面向复杂工况的航空发动机气路失稳过程智能故障预测

For the domestic aero-engine in the complex working condition, the gas path abnormal fault often appears due to stalling and surge, which can serious threat to flight safety. Therefore, in this work, the intelligent prognosis and fault diagnosis of the gas path aerodynamic instability in aero-engine becomes essential to ensure safe operation of aero-engines and reduce maintenance costs. Firstly, the holographic dynamic characteristics of key performance parameters in aero-engine is extracted through deep learning algorithm. Moreover, the real-time prediction model of adaptive rolling window is established to determine the dynamic trend of each parameter in aero-engine aerodynamic instability developing process. Secondly, based on the artificial intelligence, the gas path aerodynamic instability fault diagnosis is carried out with transfer learning to identify the abnormal knowledge and the intelligent fault model. And the incremental update approach is applied to build the gas path abnormal condition knowledge database in aero-engine. Finally, the aero-engine gas line abnormal conditions intelligent fault prognosis software platform will be implemented on the test site prototype application demonstration in the cooperation institution, and it will provide the theoretical guild and the algorithm support for the domestic aero-engine gas path instability intelligent fault prognosis.

国产航空发动机在复杂工况运行中，会遇到失速、喘振等严重威胁飞行安全的气路异常故障。本项目将开展航空发动机气路失稳过程的智能故障预测研究，以保障发动机的安全运行和降低其维护成本。首先，引入自适应滚动窗口技术保证不同时空尺度下深度学习算法的实时性，并根据提取的失稳过程气路关键性能参数全息特征，预测航空发动机气动失稳过程中各参数的动态变化趋势。接着，通过将航空发动机气路失稳异常工况的故障诊断同迁移学习相结合，构建增量更新的气路失稳故障知识库，完成航空发动机气路失稳异常工况的知识迁移与智能推理研究。最后，将开发航空发动机气路失稳异常工况智能故障预测软件平台，并在合作单位利用实际现场试车数据进行应用验证，为我国航空发动机气路失稳过程的智能故障预测提供理论指导和算法支撑。

项目从国家重大需求的实际工程出发，研究航空发动机气路失稳异常工况的智能预测和故障诊断的实际问题。通过建立自适应滚动窗口的深度学习实时预测模型，确定气动失稳过程中关键性能参数的动态变化趋势。并将航空发动机气路失稳异常工况的预测与故障诊断同人工智能相结合，能根据不同发动机进行气路失稳异常工况特征的归纳、迁移和融合，并运用迁移学习对失速喘振等航空发动机气路失稳故障进行准确判断，构建航空发动机气路失稳异常工况自动增量更新知识库管理平台。项目的完成将有助于建立完善的航空发动机气路失稳异常工况智能故障预测算法，深入研究不同算法的优缺点与适用性，为实际工程试验中提供提理论指导和算法支持。