基于增强KECA算法的航电电源板层间CAF短路故障智能监测方法研究

Research on intelligent monitoring method of CAF short circuit fault between avionics power boards based on enhanced KECA algorithm

摘要:

飞机巡航阶段遭遇的湍流振动会诱发细丝机械疲劳断裂,而着陆冲击可瞬间改变导电阳极细丝(Conductive Anodic Filament,CAF)接触阻抗,出现短路现象。但CAF短路初期的电信号特征较为微弱,难以获取CAF短路故障的最敏感主元特征,降低了故障监测的准确性。为此,研究基于增强KECA算法的航电电源板层间CAF短路故障智能监测方法。以高精度霍尔效应电流探头捕捉航电电源板层间CAF电气信号,将其分解出多尺度模态分量,计算各电气信号IMF分量的能量熵。鉴于飞机相关状况致CAF短路且初期电信号微弱,单纯依靠能量熵难挖掘最敏感主元特征,传统KECA存在投影方向选择范围窄的缺陷,对号IMF分量的能量熵进行KECA增强下主元特征向量低维映射,筛选出对短路故障最敏感的特征向量并投影到低维空间,使主元向量更聚焦关键特征。引入灰狼算法改进支持向量机,将改进支持向量机作为AdaBoost算法的弱分类器,加权组合为强分类器,得到航电电源板层间CAF短路故障识别模型,实现CAF短路故障智能监测。实验结果表明:所提方法应用下,故障样本与无故障样本点边界清晰且同类样本点之间有很好的聚类性；面对早期故障样本,对数损失相对更低且在面对中期和后期故障样本,对数损失也依然保持了较低水平,由此证明了所研究方法的故障监测准确性更高。

Abstract:

The turbulence vibration encountered during the cruise phase of the aircraft can induce mechanical fatigue fracture of the Filament, while the landing shock can instantaneously change the contact impedance of the Conductive anode filament (Conductive Anodic Filament, CAF), resulting in short circuit. However, the electrical signal characteristics in the early stage of CAF short circuit are relatively weak, making it difficult to obtain the most sensitive principal component characteristics of CAF short circuit faults and reducing the accuracy of fault monitoring. For this purpose, an intelligent monitoring method for interlayer CAF short-circuit faults of avionics power supply boards based on the enhanced KECA algorithm is studied. The electrical signals of the interlayer CAF of the avionics power supply board are captured by a high-precision Hall effect current probe, decomposed into multi-scale modal components, and the energy entropy of the IMF component of each electrical signal is calculated. Given that the CAF short circuit is caused by the airportal-related conditions and the initial electrical signal is weak, it is difficult to discover the most sensitive principal component features solely relying on energy entropy. Traditional KECA has the defect of a narrow selection range of projection directions. The energy entropy of the IMF component is subjected to low-dimensional mapping of principal component feature vectors enhanced by KECA to screen out the feature vectors most sensitive to short circuit faults and project them into the low-dimensional space. Make the principal component vector more focused on the key features. The Grey Wolf algorithm was introduced to improve the support vector machine. The improved support vector machine was used as the weak classifier of the AdaBoost algorithm and weighted and combined into a strong classifier to obtain the inter-layer CAF short-circuit fault recognition model of the avionics power supply board, achieving intelligent monitoring of CAF short-circuit faults. The experimental results show that under the application of the proposed method, the boundaries between faulty samples and non-faulty sample points are clear, and there is good clustering among similar sample points. When dealing with early fault samples, the logarithmic loss is relatively lower, and even when dealing with mid-stage and late-stage fault samples, the logarithmic loss remains at a low level. This proves that the fault monitoring accuracy of the studied method is higher.