论文部分内容阅读
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。
Aiming at the problem that the measurement data can not effectively identify the degree of performance degradation of each component because of the coupling between the components, a performance degradation identification method of engine components based on the performance correction factor kernel mode analysis is proposed and can be distinguished from the sensor measurement deviation. The sensor measurements are first entered into the adaptive model to generate a set of correction factors for identifying component degradation. The correction factor reference model is mapped into the high-dimensional feature space through the kernel mode, and recognition can be completed in the separable (substantially separable) space. Considering that the map of correction factor reference model is zonal distribution in the high-dimensional space, the geometric distance can not be effectively identified. Based on this, the neural network method is used to identify the pattern. The recognition success rate reached 94.34%. The effect of input dimension of feature reduction on recognition effect and the generalization ability of the proposed method are further analyzed. Examined the impact of noise on pattern recognition, the noise within 3% of the amplitude of no significant impact on the recognition results. It is proved that “adaptive model + kernel mode analysis + neural network” identification method is feasible.