Remaining useful life prediction of variable-operating turbofan engine based on VMD-CNN-BiLSTM
-
摘要:
为解决传统预测方法变工况涡扇发动机剩余寿命预测精度较低的问题，提出一种变工况涡扇发动机剩余使用寿命(RUL)预测模型。采用归一化处理和变分模态分解(VMD)将数据分解为预定个数的特定区间内的子数据，充分挖掘多维数据中的隐藏时序特征，消除奇异样本和不同量纲的影响；构建基于VMD-CNN-BiLSTM的变工况涡扇发动机RUL预测模型，卷积神经网络(CNN)用于特征提取和融合生成若干映射，将数据映射输入双向长短时神经网络(BiLSTM)进行训练，捕捉时序数据的时间依赖性，输出RUL预测结果；采用麻雀搜索算法(SSA)选取超参数优解，提升模型的预测性能。涡扇发动机RUL预测实验结果表明：VMD-CNN-BiLSTM在变工况涡扇发动机RUL预测中，均方根误差(RMSE)和平均绝对误差(MAE)分别达到13.74±0.51和11.24±0.49，且在噪声环境下仍具有较高的精度和泛化性能。
Abstract:In order to address the issue of low prediction accuracy in traditional forecasting methods for residual life of turbofan engines under variable working conditions, a variational mode decomposition convolutional neural network bidirectional long short term memory (VMD-CNN-BiLSTM) model is proposed. Firstly, variational mode decomposition (VMD) is used to normalize the data and split it into sub-data at predetermined intervals. This allows for the thorough extraction of hidden temporal features in multidimensional data as well as the removal of singular samples and dimensional variations. Secondly, a VMD-CNN-BiLSTM model is constructed for predicting the residual life of turbofan engines under variable working conditions. The convolutional neural network (CNN) is employed for feature extraction and fusion to generate multiple mappings. These mappings are then input into the BiLSTM network to capture time dependencies in the time series data and produce accurate predictions of remaining engine life. Finally, hyperparameter optimization using the Sparrow algorithm enhances the prediction performance of the model. As shown by root mean squared error (RMSE) values of 13.74±0.51 and mean absolute error (MAE) values of 11.24±0.49 when predicting remaining engine life under variable operating conditions, experimental results on the commercial modular aero-propulsion system simulation (C-MAPSS) dataset show that VMD-CNN-BiLSTM achieves high accuracy and generalization performance even with noisy data.
-
图 3 基于VMD-CNN-BiLSTM的航空发动机剩余使用寿命预测模型
Figure 3. RUL prediction model of aero engine based on VMD-CNN-BiLSTM
图 4 基于VMD-CNN-BiLSTM的航空发动机剩余使用寿命预测流程
Figure 4. Prediction process of aeroengine remaining useful life based on VMD-CNN-BiLSTM
图 5 不同序列分解个数K下的RMSE、MAE和总耗时
Figure 5. RMSE, MAE and total time spent under different sequence decomposition numbers K
图 6 不同训练轮数下的RMSE、MAE和总耗时
Figure 6. RMSE, MAE and total time spent under different training rounds
图 7 不同批大小B下的RMSE、MAE和总耗时
Figure 7. RMSE, MAE, and total time spent for different batch sizes B
图 10 FD002中1号发动机的真实寿命和预测寿命
Figure 10. Real and predicted life of engine No. 1 in FD002
表 1 C-MAPSS数据集介绍
Table 1. Description of C-MAPSS datasets
数据集 发动机数量/个 工况种类/个 故障类型 FD001 100 1 1类 FD002 259 6 1类 FD003 100 1 2类 FD004 248 6 2类 
下载: 导出CSV
表 2 涡扇发动机剩余使用寿命预测数据描述
Table 2. Description of remaining useful life prediction data of turbofan engine
特征序号 特征描述 符号 1 飞行高度/km H 2 油门杆角度/% θTRA 3 马赫数 Ma 4 风扇进气温度/℃ T2 5 低压压气机排气温度/℃ T24 6 高压压气机排气温度/℃ T30 7 低压涡轮排气温度/℃ T50 8 风扇入口气压/kPa P2 9 外涵道总压力/kPa P15 10 高压压气机出口总压/kPa P30 11 未修正的风扇转速/(r·min−1) NF 12 未修正的核心机转速/(r·min−1) NC 13 发动机压比 θEPR 14 高压压气机出口静压/kPa PS30 15 燃油流量和高压压气机出口总压比值 θPHI 16 风扇修正转速/(r·min−1) NRF 17 核心机修正转速/(r·min−1) NRC 18 涵道比 θBPR 19 燃烧室燃气比 θFARB 20 引气焓值/kg HT_BLEEd 21 风扇转速命令值/(r·min−1) NF_DMD 22 风扇修正量转速命令值/(r·min−1) θPCNFR_DMD 23 高压涡轮冷却空气流量/(kg·s−1) W31 24 低压涡轮冷却空气流量/(kg·s−1) W32
下载: 导出CSV
表 3 VMD-CNN-BiLSTM模型参数
Table 3. VMD-CNN-BiLSTM model parameters
输入层
尺寸卷积
层数BiLSTM
层数卷积核尺寸 卷积核数量 卷积步长 最大池化层尺寸 池化
步长BiLSTM
隐藏
单元数随机
丢弃
比例序列
分解
个数训练
轮数批大小 第1层 第2层 第1层 第2层 第1层 第2层 第1层 第2层 24 2 1 [2, 1] [2, 1] 27 22 [1, 1] [1, 1] [2, 1] [2, 1] [2, 2] 61 0.3 3 30 160
下载: 导出CSV
表 5 集成模型与个体学习器预测对比
Table 5. Comparison of integrated model and individual learner prediction
预测方法 RMSE MAE CNN 22.56±1.67 19.61±1.70 BiLSTM 22.94±1.87 20.11±1.94 CNN-BiLSTM 21.09±1.92 19.74±1.51 VMD-CNN-BiLSTM 13.74±0.51 11.24±0.49 注：加粗数值表示最优结果。
下载: 导出CSV
-
[1] 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): 447-457.GUO X J, XU X H, GUO J H. Improved GRU-based self-attention optimization algorithm for aero-engine remaining useful life prediction[J]. Journal of Aerospace Power, 2024, 39(12): 447-457(in Chinese). [2] 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.GUO X J, YUN Y J, XU X H. Prediction model of aero-engine remaining useful life based on deep learning method[J]. Journal of Vibration, Measurement & Diagnosis, 2024, 44(2): 330-336(in Chinese). [3] 车鲁阳, 高军伟, 付惠琛. 基于双通道模型的航空发动机剩余寿命预测[J]. 空军工程大学学报, 2023, 24(6): 42-49.CHE L Y, GAO J W, FU H C. Remaining life prediction of aero-engine based on dual-channel model[J]. Journal of Air Force Engineering University, 2023, 24(6): 42-49(in Chinese). [4] 宋慧, 陶冠叶, 曲大义, 等. 基于LSTM网络参数优化的航空发动机寿命预测[J]. 青岛理工大学学报, 2023, 44(5): 112-117.SONG H, TAO G Y, QU D Y, et al. Aeroengine life prediction based on LSTM network parameter optimization[J]. Journal of Qingdao University of Technology, 2023, 44(5): 112-117(in Chinese). [5] LI Y G, NILKITSARANONT P. Gas turbine performance prognostic for condition-based maintenance[J]. Applied Energy, 2009, 86(10): 2152-2161. [6] SON K L, FOULADIRAD M, BARROS A, et al. Remaining useful life estimation based on stochastic deterioration models: a comparative study[J]. Reliability Engineering & System Safety, 2013, 112: 165-175. [7] CHIACHÍO J, JALÓN M L, CHIACHÍO M, et al. A Markov chains prognostics framework for complex degradation processes[J]. Reliability Engineering & System Safety, 2020, 195: 106621. [8] LI X, DING Q, SUN J Q. Remaining useful life estimation in prognostics using deep convolution neural networks[J]. Reliability Engineering & System Safety, 2018, 172: 1-11. [9] ZHANG J S, JIANG Y C, WU S M, et al. Prediction of remaining useful life based on bidirectional gated recurrent unit with temporal self-attention mechanism[J]. Reliability Engineering & System Safety, 2022, 221: 108297. [10] 赵昱宇, 索超, 王雨潇. 基于BSVAR的航空发动机剩余使用寿命预测方法[J]. 北京航空航天大学学报, 2025, 51(11): 3790-3798.ZHAO Y Y, SUO C, WANG Y X. BSVAR-based remaining useful life prediction method for aircraft engines[J]. Journal of Beijing University of Aeronautics and Astronautics, 2025, 51(11): 3790-3798(in Chinese). [11] 林荣祥, 王海瑞, 朱贵富. 一种融合神经网络的航空发动机寿命预测方法研究[J]. 机械科学与技术, 2025, 44(11): 2034-2046.LIN R X, WANG H R, ZHU G F. An aero-engine life prediction method incorporating neural networks[J]. Mechanical Science and Technology for Aerospace Engineering, 2025, 44(11): 2034-2046(in Chinese). [12] 魏震波, 余雷. 基于FFT, DC-HC及LSTM的短期负荷预测方法[J]. 智慧电力, 2022, 50(3): 37-43.WEI Z B, YU L. Short-term load forecasting method based on FFT, DC-HC and LSTM[J]. Smart Power, 2022, 50(3): 37-43(in Chinese). [13] 杨德州, 刘嘉明, 宋汶秦, 等. 基于改进型自适应白噪声完备集成经验模态分解的工业用户负荷预测方法[J]. 电力系统保护与控制, 2022, 50(4): 36-43.YANG D Z, LIU J M, SONG W Q, et al. A load forecasting method for industrial customers based on the ICEEMDAN algorithm[J]. Power System Protection and Control, 2022, 50(4): 36-43(in Chinese). [14] 李练兵, 代亮亮, 李静鹏, 等. 基于改进VMD-CNN的电缆短路电流预测研究[J]. 中国测试, 2025, 51(8): 137-146.LI L B, DAI L L, LI J P, et al. Research on cable short-circuit current prediction based on improved VMD-CNN[J]. China Measurement & Test, 2025, 51(8): 137-146(in Chinese). [15] 龚循强, 邱万锦, 吕开云, 等. 变分模态分解与自适应图卷积门控循环网络的交通流量组合预测模型[J]. 武汉大学学报(信息科学版), 2024, 49(12): 2329-2341.GONG X Q, QIU W J, LYU K Y, et al. A combined traffic flow prediction model based on variational mode decomposition and adaptive graph convolutional gated recurrent network[J]. Geomatics and Information Science of Wuhan University, 2024, 49(12): 2329-2341(in Chinese). [16] 秦志沁, 韩玉环, 张毅, 等. 基于VMD分解和随机矩阵理论的异常用电状态检测[J]. 太原理工大学学报, 2024, 55(1): 66-72.QIN Z Q, HAN Y H, ZHANG Y, et al. Detection of abnormal power consumption state based on VMD decomposition and random matrix theory[J]. Journal of Taiyuan University of Technology, 2024, 55(1): 66-72(in Chinese). [17] 邵必林, 纪丹阳. 基于VMD-SE的电力负荷分量的多特征短期预测[J]. 中国电力, 2024, 57(4): 162-170.SHAO B L, JI D Y. Multi-feature short-term prediction of power load components based on VMD-SE[J]. Electric Power, 2024, 57(4): 162-170(in Chinese). [18] 王勇, 李邑灵, 苗夺谦, 等. 基于Transformer-CNN混合架构的跨模态融合抓取检测[J]. 控制与决策, 2024, 39(11): 3607-3616.WANG Y, LI Y L, MIAO D Q, et al. Cross-modal interaction fusion grasping detection based on Transformer-CNN hybrid architecture[J]. Control and Decision, 2024, 39(11): 3607-3616(in Chinese). [19] 孙兴伟, 杨铜铜, 杨赫然, 等. 基于CNN-GRU组合神经网络的数控机床进给系统热误差研究[J]. 仪器仪表学报, 2023, 44(10): 219-226.SUN X W, YANG T T, YANG H R, et al. Research on thermal error of CNC machine tool feed system based on CNN-GRU combined neural network[J]. Chinese Journal of Scientific Instrument, 2023, 44(10): 219-226(in Chinese). [20] 杨沛东, 黄华, 尉卫卫, 等. 基于注意力机制改进的PSO-BiLSTM刀具磨损预测[J]. 北京航空航天大学学报, 2025, 51(10): 3589-3598.YANG P D, HUANG H, WEI W W. Tool wear prediction based on attention mechanism and PSO-BiLSTM[J]. Journal of Beijing University of Aeronautics and Astronautics, 2025, 51(10): 3589-3598(in Chinese). [21] 方斯顿, 刘龙真, 孔赖强, 等. 基于双向长短期记忆网络含间接健康指标的锂电池SOH估计[J]. 电力系统自动化, 2024, 48(4): 160-168.FANG S D, LIU L Z, KONG L Q. State-of-health estimation for lithium-ion batteries incorporating indirect health indicators based on bi-directional long short-term memory networks[J]. Automation of Electric Power Systems, 2024, 48(4): 160-168(in Chinese). [22] 陶唐飞, 周文洁, 况佳臣, 等. 融合多小波分解的深度卷积神经网络轴承故障诊断方法[J]. 西安交通大学学报, 2024, 58(5): 31-41.TAO T F, ZHOU W J, KUANG J C. Bearing fault diagnosis method of deep convolutional neural network based on multiwavelet decomposition[J]. Journal of Xi’an Jiaotong University, 2024, 58(5): 31-41(in Chinese). [23] 单伽锃, 张茜, 吕西林, 等. 多元特征驱动的超高层建筑变形状态智能学习与预测[J]. 建筑结构学报, 2024, 45(3): 101-112.SHAN J Z, ZHANG X, LYU X L, et al. Multi-feature driven intelligent learning and prediction of deformation state of super-tall buildings[J]. Journal of Building Structures, 2024, 45(3): 101-112(in Chinese). [24] 丁丁, 刘文哲, 盛常冲, 等. 神经网络架构搜索研究进展与展望[J]. 国防科技大学学报, 2023, 45(6): 100-131.DING D, LIU W Z, SHENG C C, et al. State of the art and prospects of neural architecture search[J]. Journal of National University of Defense Technology, 2023, 45(6): 100-131(in Chinese). -
下载:
点击查看大图
计量
- 文章访问数: 115
- HTML全文浏览量: 40
- PDF下载量: 12
- 被引次数: 0
