基于LMKL和OC-ELM的航空电子部件故障检测方法

Fault detection method for avionics based on LMKL and OC-ELM

针对航空电子部件故障样本获取困难以及检测准确率不高的问题,提出基于局部多核学习(localized multiple kernel learning,LMKL)和一类超限学习机(one-class extreme learning machine,OC-ELM)的故障检测方法。仅运用正常状态的小样本数据,给出了LMK-OC-ELM的数学表达形式,并在不同的门模型下推导了LMK-OC-ELM中局部核权重的优化方法;在获取局部核权重的基础上,定义了离线故障检测所需的统计检验量与阈值,以便工程实现。将所提方法应用于某型接收机,结果表明,在训练时间可控的前提下,与4种常见的一类分类(one-class classification,OCC)算法相比,所提方法可均衡地提高召回率、查准率和特异度,以LMK-OC-ELM-sig为代表,其在F1、曲线下方面积(area under curve,AUC)、G-mean和准确率4个指标上,比最近提出的局部多核异常检测(localized multiple kernel anomaly detection,LMKAD)方法分别提高了1.60%、1.57%、1.53%和2.23%。

In consideration of the difficulty in acquiring real fault samples and the low detection accuracy of the avionics,a fault detection method based on localized multiple kernel learning(LMKL)and one-class extreme learning machine(OC-ELM)is proposed.The mathematical expression of LMK-OC-ELM is given onlyby using small sample data in normal state,and the localized kernel weights in LMK-OC-ELM is deduced under different gating models.On the basis of obtaining the localized kernel weights,the test statistic and threshold required for offline fault detection are defined to facilitate the engineering implementation.The proposed method is applied to the receiver.On the premise of controllable training time,the proposed method can improve recall,precision and specificity equitably compared with the other four common one-class classification algorithms.Taking LMK-OC-ELM-sig as the representative,the indicators of F1,area under curve(AUC),G-mean and accuracy are increased by 1.60%,1.57%,1.53% and 2.23% respectively compared with the localized multiple kernel anomaly detection(LMKAD)which is recently proposed.