自适应核模式分析方法及其在航空发动机部件性能衰退识别中的应用

Self-Adaptive Kernel Pattern Analysis Method and Its Application in Aeroengine Component Performance Deterioration Recognition

关键词:  自适应模型  核模式分析  性能衰退  修正因子 

Key words:  Self-adaptive model  Kernel pattern analysis  Performance deterioration  Modified factor

摘要:

针对测量数据因其部件之间的耦合不能有效识别各个部件性能衰退程度的问题,提出一种基于性能修正因子核模式分析的发动机部件性能衰退识别方法,并能与传感器测量偏差区分开。首先将传感器测量数据输入到自适应模型中去,产生一组用于识别部件性能衰退的修正因子。将修正因子参考模式通过核模式映射到高维特征空间中去,在此可分(基本可分)空间中完成识别。考虑到修正因子参考模式在高维空间中映射的像呈带状分布,几何距离不能有效识别,基于此采用神经网络方法对模式进行识别。识别成功率达到94.34%。进一步分析了特征约简的输入维数对识别效果的影响以及所提方法的泛化能力。考查了噪声对模式识别的影响,得到幅值3%以内的噪声对识别结果无明显影响。证明了“自适应模型+核模式分析+神经网络”识别方法是可行的。 

Abstract:

Aiming at component performance deterioration not being recognized because of coupling with components in measured parameter, a recognition method of component performance deterioration based on performance modified factor kernel pattern analysis was presented and deviation of sensor was distinguished. Measured parameters were fed to self-adaptive model, a group of component performance modified factor used as pattern recognition was generated. Modified factor reference mode was mapped to high-dimension character space and recognition was accomplished in this separable (basically separable) space. Considering that map of modified factor reference mode has shape of strap, geometry distance is not recognized well. Therefore neural network is adopted. Success ratio of recognition reaches 94.34%. Furtherly the effects of input dimension after simplification on recognition and generalizability were analyzed. The effects of noise on pattern recognition was examined and noise within 3% had little influence on recognition. It verifies the pattern recognition method ‘self-adaptive model+kernel pattern analysis+neural network’ is feasible.