Feature extraction is the key technique for fault prediction of avionics , Wavelet transform, Fourier transform, empirical mode decomposition methods can be used to extract fault features of the electronic equipment with few test points.Due to the fact that the avionics is large-scale integrated circuits which includes many test points, fault features extracted based on the method above may be mixed with each other and the number is large, which will seriously affect the accuracy and speed of fault prediction.It is a difficult problem to extract fault features from many fault information.In this paper, we propose the method based on denoising autoencoder and maximum likelihood to extract fault features from a large number of fault information.First of all, maximum likelihood is taken to analyze the high dimensional data comprised of the fault information which were extracted from many test points and historical degradation process, and to estimate the intrinsic dimension of fault features;Then, the high dimensional data is mapped to the specified dimension data space by using denoising autoencoder method.The key fault features are extracted from the data, and the redundant information is removed.Finally, taking the avionics power system as an example, through the fault feature visualization and health assessment demonstrate that the method proposed in the paper which can extract fault features is effective.%故障特征提取是航空电子设备故障预测的关键技术,对于少量测试点的电子设备可以采用小波变换、傅里叶变换、经验模态分解等方法提取故障特征,但是由于航空电子设备属于大规模集成电路,测试点比较多,采用上述方法提取的故障特征可能相互混叠并且数量比较大会严重影响故障预测精度及速度,因此如何从众多故障信息中提取故障特征是一个难题.文章提出基于极大似然和降噪自编码神经网络方法从大量故障信息中提取故障特征.首先,使用极大似然法分析由多个测试点提取的故障信息和历史退化过程的故障信息组成的高维数据集,估计需要提取故障特征的维数;然后使用降噪自编码神经网络方法将高维故障信息映射到指定维数的数据空间,从中提取关键的故障特征,去除冗余信息;最后,以航空电子系统电源模块为例,采用新方法提取故障特征,分别通过将故障特征可视化和使用故障特征进行健康评估来验证其有效性.
展开▼
