Computer Integrated Manufacturing System ›› 2022, Vol. 28 ›› Issue (2): 417-425.DOI: 10.13196/j.cims.2022.02.008
Previous Articles Next Articles
Online:
Published:
徐硕,侯贵生
Abstract: To improve the prediction accuracy of Remaining Useful Life (RUL) of turbofan engines,a pre-training feature extraction model combining Variational Autoencoder (VAE) with Dual Discriminator Generative Adversarial Nets (D2GAN) was proposed.As the generator of D2GAN,VAE participated in the model training to form a double nested generation structure to improve the quality of intermediate feature extraction.Long Short-Term Memory Networks (LSTM) was designed to further capture the time-series degradation information from the extracted features to predict the engine RUL.To verify the efficiency of the proposed method,the proposed model was tested on a common dataset and compared with several current state-of-the-art studies.The results showed that the proposed model had achieved better prediction performance,which greatly improved the safety of engine system.
Key words: deep learning, remaining useful life prediction, variational autoencoder, dual discriminator generative adversarial nets, turbofan engine
摘要: 为了提高涡扇发动机剩余使用寿命的预测精度,提出一种将变分自编码器(VAE)和双判别器对抗式生成网络(D2GAN)相结合的预训练特征提取模型。在该模型中,VAE作为D2GAN的生成器参与模型训练,形成双重嵌套生成结构,以提高中间特征的提取质量;利用长短时记忆网络进一步挖掘所提取特征的时序退化信息,预测发动机剩余使用寿命。为了验证所提模型的高效性,将模型在通用数据集上进行测试,并与当前最先进的研究比较,结果显示所提模型具有更优秀的预测表现,极大提高了发动机系统的安全性。
关键词: 深度学习, 剩余使用寿命预测, 变分自编码器, 双判别器对抗式生成网络, 涡扇发动机
CLC Number:
TP183
V23
徐硕, 侯贵生. 基于VAE-D2GAN的涡扇发动机剩余使用寿命预测[J]. 计算机集成制造系统, 2022, 28(2): 417-425.
0 / Recommend
Add to citation manager EndNote|Ris|BibTeX
URL: http://www.cims-journal.cn/EN/10.13196/j.cims.2022.02.008
http://www.cims-journal.cn/EN/Y2022/V28/I2/417