基于数域映射的航空发动机剩余寿命预测VIT模型的重参数化改进

郭晓静; 郭佳豪; 徐琛

doi:10.13224/j.cnki.jasp.20240824

基于数域映射的航空发动机剩余寿命预测VIT模型的重参数化改进

doi: 10.13224/j.cnki.jasp.20240824

郭晓静^1,,
郭佳豪^2,*, ,,
徐琛¹

1.
中国民航大学工程技术训练国家级实验教学示范中心，天津 300300
2.
中国民航大学电子信息与自动化学院，天津 300300

详细信息

作者简介:
郭晓静（1980-），女，副教授，硕士，研究方向为智能检测、图像处理。E-mail：13820869553@139.com

通讯作者:
郭佳豪（1999-），男，硕士生，研究方向为航空发动机寿命预测。E-mail：18822575712@163.com

中图分类号: V240.2
计量
- 文章访问数: 685
- HTML浏览量: 453
- PDF量: 40
- 被引次数: 0
出版历程
- 收稿日期: 2024-12-05
- 网络出版日期: 2025-08-02

Aero-engine remaining useful life prediction of VIT model re-parameterized optimization method based on data field mapping

GUO Xiaojing^1
,,
GUO Jiahao^{2,*
, ,},
XU Chen¹

1.
National Demonstration Center for Experimental Engineering Techniques Training Education，Civil Aviation University of China，Tianjin 300300，China
2.
College of Electronic Information and Automation，Civil Aviation University of China，Tianjin 300300，China

摘要

摘要:
航空发动机参数具有高维时序性，参数特征能够用于表征发动机剩余寿命。从起飞到着陆全周期内，不同机型所采集和存储的发动机参数量和数据规模差异明显。为了解决参数维数不同导致的特征提取细粒度不一致，从而影响发动机寿命预测精度的问题，提出一种用重参数化结构改进的vision transformer（VIT）模型。建立多维数域映射算法，将发动机参数源域数据集转换为彩色图像数据集，从数据源端改善了泛化性。改进VIT模型的多头注意力卷积结构，引入重参数化结构及全连接层，将源域数据的时序性转换为图像数据的空间特性，提高了模型寿命预测精度。在公开数据集（CMAPSS）上实验表明，寿命预测方均根误差（RMSE）范围为[10.83，14.68]，预测精度至少提高了4.3%。此外，该方法在公开数据集（N-CMAPSS）测试，RMSE为2.07，进一步验证了模型泛化性能。
- 航空发动机剩余寿命 /
- 数域映射 /
- 重参数化结构 /
- 多头注意力 /
- 图像特征提取
Abstract:
Time-series aero-engine parameters are originated from different data sources. Its feature might be relevant with the remaining useful life prediction of the aero-engine. During the whole periods from the airplane’s taking off to landing, different types of aircrafts have particular engine parameters and data scale. In order to improve the remaining useful life prediction precision and apply it into different dimension engine datasets, the new prediction model was constructed based on the revised VIT （vision transformer） with the re-parameterization. The algorithm of datasets mapping into the RGB image was given, and the model generalization got better. Re-parameterization was combined with multi-dimension attention of VIT, and the precision was improved after the transferring between the time series datasets and space feature. The result of the experiment in datasets （CMAPSS） of the datasets showed that the value of root mean square error （RMSE） lied in [10.83,14.68]. The prediction model and algorithm were better than the traditional method with the smaller RMSE of 4.3% decrease in the least. And also, it could be applicable in another datasets (N-CMAPSS ) with the RMSE value of 2.07.
- aero-engine remaining useful life /
- data field mapping /
- re-parameterized structure /
- multi-dimension attention /
- image feature extraction

HTML

图 1 数域映射流程

Figure 1. Data mapping process

下载: 全尺寸图片幻灯片

图 2 应用 VIT 处理批次图像

Figure 2. VIT processing of batch image

下载: 全尺寸图片幻灯片

图 3 重参数化结构

Figure 3. Re-parameterized structure

下载: 全尺寸图片幻灯片

图 4 改进后的 Rep-VIT 结构

Figure 4. Improved Rep-VIT model structure

下载: 全尺寸图片幻灯片

图 5 各列、类与寿命相关性分析

Figure 5. Correlation analysis

下载: 全尺寸图片幻灯片

图 6 部分数据生成映射图像

Figure 6. Partial-data image mapping

下载: 全尺寸图片幻灯片

图 7 发动机剩余寿命预测曲线（随机选取）

Figure 7. Engine remaining useful life prediction curves （random）

下载: 全尺寸图片幻灯片

表 1 不同训练方式实验结果

Table 1. Dataset experimental results in Group1 & Group2

数据子集	发动机台数	初始训练集图片数	初始测试集图片数	初始训练集、测试集图片数比值	RMSE
数据子集	发动机台数	初始训练集图片数	初始测试集图片数	初始训练集、测试集图片数比值	Group1 （独立训练单独测试）	Group2 （合并训练单独测试）	Group2* （合并扩增训练单独测试）
FD001	100	18331	10796	1.69	16.91	14.75	10.97
FD002	260	47779	28040	1.71	19.27	17.32	14.52
FD003	100	22420	14296	1.57	15.56	13.37	10.83
FD004	249	55522	35532	1.56	20.32	18.64	14.68
注：Group2*训练集与测试集图片数接近2∶1。

下载: 导出CSV

表 2 消融实验结果

Table 2. Ablation experiment results

模型	权重分解映射数据层				e_rms（合并扩增训练单独测试）
模型	原数据映射层	列权重映射层	类权重映射层	状态列映射层	FD001	FD002	FD003	FD004
VIT+全连接	√				22.68	23.76	21.94	24.3
VIT+全连接	√	√			21.46	22.33	19.27	24.19
VIT+全连接	√	√	√		17.38	21.27	18.93	22.38
VIT+全连接	√	√	√	√	15.70	18.76	16.21	19.35
Rep-VIT+全连接	√				19.83	20.88	19.35	22.27
Rep-VIT+全连接	√	√			16.34	18.80	15.92	19.06
Rep-VIT+全连接	√	√	√		11.76	15.87	12.69	15.34
Rep-VIT+全连接（本文模型）	√	√	√	√	10.97	14.52	10.83	14.68

下载: 导出CSV

表 3 对比实验结果

Table 3. Results comparison of different models

模型	e_rms（合并扩增训练单独测试）				S（合并扩增训练单独测试）
模型	FD001	FD002	FD003	FD004	FD001	FD002	FD003	FD004
CNN	20.34	23.64	21.76	25.19	793.21	2876.34	1349.70	2458.29
CNN-GRU	13.23	16.72	12.46	18.33	612.74	1973.58	902.34	1952.24
CNN-Bilstm	12.96	17.03	13.17	18.20	660.48	1853.78	1138.74	2033.51
Rep-VGG	17.29	20.01	18.79	20.83	703.27	2247.57	3768.25	2786.27
Rep-GRU	13.02	15.76	12.37	17.26	634.46	2057.28	983.64	1864.35
Rep-BiLSTM	12.54	16.27	14.30	16.73	504.94	1833.24	1067.53	2357.85
Rep-VIT+全连接（本文模型）	10.97	14.52	10.83	14.68	544.21	1480.22	972.00	1748.32

下载: 导出CSV

表 4 交叉验证实验结果

Table 4. Cross-validation results analysis of datasets

训练集	e_rms（合并扩增训练单独测试）
训练集	FD001	FD002	FD003	FD004	N-CMAPSS-DS08
Group2*	10.97	14.52	10.83	14.68	10.59
N-CMAPSS-DS08	20.76	22.34	23.94	25.78	5.73
CMAPSS & N-CMAPSS-DS08	12.37	14.37	11.62	14.52	2.07

下载: 导出CSV

参考文献(20)

[1]	赵广社, 吴思思, 荣海军. 多源统计数据驱动的航空发动机剩余寿命预测方法[J]. 西安交通大学学报, 2017, 51(11): 150-155, 172. ZHAO Guangshe, WU Sisi, RONG Haijun. A multi-source statistics data-driven method for remaining useful life prediction of aircraft engine[J]. Journal of Xi’an Jiaotong University, 2017, 51(11): 150-155, 172. (in Chinese doi: 10.7652/xjtuxb201711021 ZHAO Guangshe, WU Sisi, RONG Haijun. A multi-source statistics data-driven method for remaining useful life prediction of aircraft engine[J]. Journal of Xi’an Jiaotong University, 2017, 51(11): 150-155, 172. (in Chinese) doi: 10.7652/xjtuxb201711021
[2]	孙见忠, 左洪福, 梁坤. 基于民航发动机状态数据的涡轮叶片剩余寿命评估[J]. 机械工程学报, 2015, 51(23): 53-59. SUN Jianzhong, ZUO Hongfu, LIANG Kun. Remaining useful life estimation method for the turbine blade of a civil aircraft engine based on the QAR and field failure data[J]. Journal of Mechanical Engineering, 2015, 51(23): 53-59. (in Chinese doi: 10.3901/JME.2015.23.053 SUN Jianzhong, ZUO Hongfu, LIANG Kun. Remaining useful life estimation method for the turbine blade of a civil aircraft engine based on the QAR and field failure data[J]. Journal of Mechanical Engineering, 2015, 51(23): 53-59. (in Chinese) doi: 10.3901/JME.2015.23.053
[3]	SHI Yue, ZHU Weihang, XIANG Yisha, et al. Condition-based maintenance optimization for multi-component systems subject to a system reliability requirement[J]. Reliability Engineering & System Safety, 2020, 202: 107042.
[4]	段佳俊, 陆中, 王捷. 基于多尺度CNN和Transformer的航空发动机寿命预测方法[J/OL]. 航空动力学报, (2024-11-19)[2024-11-28]. https://link.cnki.net/doi/10.13224/j.cnki.jasp.20240311. DUAN Jiajun, LU Zhong, WANG Jie. Aero-engine life prediction method based on multi-scale CNN and transformer[J/OL]. Journal of Aerospace Power, (2024-11-19)[2024-11-28]. https://link.cnki.net/doi/10.13224/j.cnki.jasp.20240311. (in Chinese DUAN Jiajun, LU Zhong, WANG Jie. Aero-engine life prediction method based on multi-scale CNN and transformer[J/OL]. Journal of Aerospace Power, (2024-11-19)[2024-11-28]. https://link.cnki.net/doi/10.13224/j.cnki.jasp.20240311. (in Chinese)
[5]	LECUN Y, BOTTOU L, BENGIO Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 2002, 86(11): 2278-2324.
[6]	LI Xiang, DING Qian, SUN Jianqiao. Remaining useful life estimation in prognostics using deep convolution neural networks[J]. Reliability Engineering & System Safety, 2018, 172: 1-11.
[7]	HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735-1780. doi: 10.1162/neco.1997.9.8.1735
[8]	CHO K, VAN MERRIENBOER B, GULCEHRE C, et al. Learning phrase representations using RNN encoder-decoder for statistical machine translation [EB/OL].(2014-09-03) [2024-08-24].https://arxiv.org/abs/1406.1078.
[9]	袁烨, 黄虹, 程骋, 等. 基于特征注意力机制的GRU-GAN航空发动机剩余寿命预测[J]. 中国科学: 技术科学, 2022, 52(1): 198-212. YUAN Ye, HUANG Hong, CHENG Cheng, et al. Remaining useful life prediction of the aircraft engine based on the GRU-GAN network with a feature attention mechanism[J]. Scientia Sinica (Technologica), 2022, 52(1): 198-212. (in Chinese doi: 10.1360/SST-2021-0434 YUAN Ye, HUANG Hong, CHENG Cheng, et al. Remaining useful life prediction of the aircraft engine based on the GRU-GAN network with a feature attention mechanism[J]. Scientia Sinica (Technologica), 2022, 52(1): 198-212. (in Chinese) doi: 10.1360/SST-2021-0434
[10]	王文庆, 郭恒, 范启富. 基于CNN与GRU的航空发动机剩余寿命预测[C]//第37届中国控制会议论文集(F). 武汉: 中国控制学会, 2018: 546-551. WANG Wenqing, GUO Heng, FAN Qifu. Predict remaining useful life of aerospace engine based on CNN and GRU [C]//Proceedings of the 37th Chinese Control Conference. (F). Wuhan: Chinese Control Conference, 2018: 546-551. (in Chinese WANG Wenqing, GUO Heng, FAN Qifu. Predict remaining useful life of aerospace engine based on CNN and GRU [C]//Proceedings of the 37th Chinese Control Conference. (F). Wuhan: Chinese Control Conference, 2018: 546-551. (in Chinese)
[11]	李杰, 贾渊杰, 张志新, 等. 基于融合神经网络的航空发动机剩余寿命预测[J]. 推进技术, 2021, 42(8): 1725-1734. LI Jie, JIA Yuanjie, ZHANG Zhixin, et al. Remaining useful life prediction of aeroengine based on fusion neural network[J]. Journal of Propulsion Technology, 2021, 42(8): 1725-1734. (in Chinese LI Jie, JIA Yuanjie, ZHANG Zhixin, et al. Remaining useful life prediction of aeroengine based on fusion neural network[J]. Journal of Propulsion Technology, 2021, 42(8): 1725-1734. (in Chinese)
[12]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need [C]// Proceedings of the 31st International Conference on Neural Information Processing Systems. New York: Curran Associates Inc. , 2017: 6000-6010.
[13]	郭晓静, 贠玉晶, 徐晓慧. 基于深度学习方法的航空发动机寿命预测模型[J]. 振动、测试与诊断, 2024, 44(2): 330-336, 412. GUO Xiaojing, YUN Yujing, XU Xiaohui. Prediction model of aero-engine remaining useful life based on deep learning method[J]. Journal of Vibration, Measurement & Diagnosis, 2024, 44(2): 330-336, 412. (in Chinese GUO Xiaojing, YUN Yujing, XU Xiaohui. Prediction model of aero-engine remaining useful life based on deep learning method[J]. Journal of Vibration, Measurement & Diagnosis, 2024, 44(2): 330-336, 412. (in Chinese)
[14]	郭晓静, 徐晓慧, 郭佳豪. 基于改进GRU的航空发动机寿命预测自注意力优化算法[J]. 航空动力学报, 2024, 39(12):20220984. GUO Xiaojing, XU Xiaohui, GUO Jiahao. Improved gru-based self-attention optimization algorithm for aero-engine remaining useful life prediction[J]. Journal of Aerospace Power, 2024, 39(12): 20220984. (in Chinese GUO Xiaojing, XU Xiaohui, GUO Jiahao. Improved gru-based self-attention optimization algorithm for aero-engine remaining useful life prediction[J]. Journal of Aerospace Power, 2024, 39(12): 20220984. (in Chinese)
[15]	杨硕, 高成. 基于长短期记忆网络与轻梯度提升机的航空发动机大修期内剩余寿命预测[J]. 航空发动机, 2024, 50(3): 87-92. YANG Shuo, GAO Cheng. Remaining useful life prediction of aeroengine during overhaul based on long short-term memory network and light gradient boosting machine[J]. Aeroengine, 2024, 50(3): 87-92. (in Chinese YANG Shuo, GAO Cheng. Remaining useful life prediction of aeroengine during overhaul based on long short-term memory network and light gradient boosting machine[J]. Aeroengine, 2024, 50(3): 87-92. (in Chinese)
[16]	CHUNGJ, GULCEHREC, CHOKH, et al. Empirical evaluation of gated recurrent neural networks on sequence modeling [EB/OL]. (2014-12-11) [2022-11-02]. https://doi.org/10.48550/arXiv.1412.3555.
[17]	DOSOVITSKIY A, BEYER L, KOLESNIKOV A, et al. An image is worth 16×16 words: transformers for image recognition at scale[EB/OL]. (2020-10-22)[2024-12-02]. https://doi.org/10.48550/arXiv.2010.11929.
[18]	DING X, ZHANG X, MA N, et al. RepVGG: Making VGG-styleConvNets great again[C]//2021 IEEE / CVF Conference on Com-puter Vision and Pattern Recognition (CVPR). New York: IEEE, 2021: 13728-13737.
[19]	丁汕汕, 陈仁文, 黄翊君, 等. 重参数化VGG网络在滚动轴承故障诊断中的应用研究[J], 振动与冲击, 2023, 42(11): 313-323. DING Shanshan, CHEN Renwen, HUANG Yijun, et al. Application study of reparameterized VGG network in rolling bearing fault diagnosis[J]. Journal of Vibration and Shock, 2023, 42(11): 313-323. DING Shanshan, CHEN Renwen, HUANG Yijun, et al. Application study of reparameterized VGG network in rolling bearing fault diagnosis[J]. Journal of Vibration and Shock, 2023, 42(11): 313-323.
[20]	ARIAS CHAO M, KULKARNI C, GOEBEL K, et al. Aircraft engine run-to-failure dataset under real flight conditions for prognostics and diagnostics[J]. Data, 2021, 6(1): 5. doi: 10.3390/data6010005