摘要
为解决飞机辅助动力装置(APU)在翼性能参数维度低而无法获得较高准确故障预测结果的问题,提出了一种基于生成对抗网络(GAN)性能参数的扩增方法。首先,在研究GAN原理的基础上,通过网格搜索算法确定生成器与判别器的优化参数;其次,研究面向APU性能退化参数的扩增方法,为APU的剩余寿命预测提供输入参数;最后,基于中国南方航空股份有限公司机队的APU在翼监测参数,全面验证和评估所提方法的性能。基于GAN生成10维的排气温度参数通过欧几里得距离、皮尔森相关系数和KL散度度量方法进行处理,结果表明生成参数与原始参数具有较好的一致性。基于3种寿命预测方法开展的对比实验中,将生成的10维参数与原始参数共同用于APU剩余寿命预测,与仅将原始性能参数用于APU剩余寿命预测相比,平均绝对误差和均方根误差表征的预测结果准确性至少提升了8.55%和3.62%。
The dimension of on-wing performance parameters of the aircraft auxiliary power unit(APU) is low, it is difficult to obtain high accuracy fault prognostics result. To solve this problem, a performance parameter augment method is proposed, which is based on the generative adversarial networks(GAN). Firstly, the principle of GAN is studied, based on which the optimization parameters of the generator and discriminator are determined through the grid search algorithm. Then, the augment method facing to APU performance degradation parameter is studied, which provides the input parameters for remaining useful life(RUL) prediction of APU. Finally, the proposed method was verified and evaluated by utilizing the real on-wing monitoring data of APU from China Southern Airlines fleet. The generated 10 D exhaust temperature parameters based on GAN were processed with Euclidean distance, Pearson correlation coefficient and Kullback-Leibler divergence methods, the results show that the generated data and original data have good consistency. In the comparison experiments based on the three RUL prediction methods, the generated data and original data are both utilized for the RUL prediction, the prediction result accuracies characterized with the mean absolute error and root mean square error are improved by 8.55% and 3.62% at least compared with those using only the original performance parameters for the RUL prediction.
作者
刘连胜
张晗星
刘晓磊
王璐璐
梁军
Liu Liansheng;Zhang Hanxing;Liu Xiaolei;Wang Lulu;Liang Jun(Department of Test and Control Engineering,Harbin Institute of Technology,Harbin 150080,China;Automatic Test and Control Institute,Harbin Institute of Technology,Harbin 150080,China;Shenyang Maintenance Base,China Southern Airlines Co.,Ltd.,Shenyang 110043,China)
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2020年第7期107-116,共10页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(61803121)
中国博士后科学基金(2019M651277)项目资助
关键词
辅助动力装置
生成对抗网络
参数扩增
故障预测
在翼寿命
auxiliary power unit
generative adversarial network
parameter augment
fault prognostics
on-wing life
