应用神经网络诊断航空发动机气路故障的前景

Prospect for neural networks used aeroengine fault diagnosis technology

介绍了近几年来国内外应用神经网络对航空发动机气路故障进行诊断的基本方法和研究进展。对单一故障进行定性的诊断已经取得了试验验证 ,结果表明神经网络具有较高的诊断准确率。对反映发动机气路部件健康状况的气流量、效率等参数的多故障、定量的诊断则取得了一些仿真研究成果。相对于基于发动机气动热力学数学模型的方法而言 ,神经网络方法具有更大的工程应用潜力。

Introduced several methodologies for usage of neural networks in fault diagnosis for aero engines. First, qualitative diagnosis for a single fault was discussed. The methodology was verified by experiments on the test cell and demonstrated a relatively high success rate. Followed by a discussion of a measurement preprocessor that can perform sensor data validation and reduce the effect of data deviation. A quantitative diagnosis for multi fault was discussed. In the simulation investigation, all major component performance parameters that reflect deterioration of engine, such as airflow capacities and efficiencies, can be accurately calculated by neural networks. Comparing approaches based on a mathematical thermal dynamics engine model, neural networks as a potential tool can be applied to engine condition monitoring and modular fault isolation.