国产日韩欧美一区二区三区三州_亚洲少妇熟女av_久久久久亚洲av国产精品_波多野结衣网站一区二区_亚洲欧美色片在线91_国产亚洲精品精品国产优播av_日本一区二区三区波多野结衣 _久久国产av不卡

?

基于融合數(shù)據(jù)和維納建模發(fā)動(dòng)機(jī)余壽預(yù)測(cè)

2021-09-09 02:03趙洪利鄭涅
航空科學(xué)技術(shù) 2021年5期

趙洪利 鄭涅

摘要:性能數(shù)據(jù)是發(fā)動(dòng)機(jī)健康狀態(tài)的重要體現(xiàn),分析性能數(shù)據(jù)可以預(yù)測(cè)發(fā)動(dòng)機(jī)剩余使用壽命,為維修決策提供依據(jù)。發(fā)動(dòng)機(jī)的健康狀態(tài)與多個(gè)監(jiān)測(cè)數(shù)據(jù)密切相關(guān)。基于訓(xùn)練發(fā)動(dòng)機(jī)數(shù)據(jù)和測(cè)試發(fā)動(dòng)機(jī)數(shù)據(jù),采用主成分分析方法融合多元數(shù)據(jù)構(gòu)建了發(fā)動(dòng)機(jī)健康指數(shù)。退化模型構(gòu)建采用維納過程方法,利用EM算法結(jié)合訓(xùn)練發(fā)動(dòng)機(jī)數(shù)據(jù)迭代優(yōu)化離線參數(shù)?;谪惾~斯方法結(jié)合測(cè)試發(fā)動(dòng)機(jī)數(shù)據(jù),在線更新退化模型參數(shù),實(shí)時(shí)計(jì)算測(cè)試發(fā)動(dòng)機(jī)剩余使用壽命概率密度分布及期望。兩種方法對(duì)比結(jié)果顯示,基于單一性能指標(biāo)構(gòu)建的性能模型,對(duì)測(cè)試發(fā)動(dòng)機(jī)最后10循環(huán)的預(yù)測(cè)均方根誤差平均值為12.95,基于融合數(shù)據(jù)構(gòu)建的性能模型的預(yù)測(cè)均方根誤差平均值為5.34,證明數(shù)據(jù)融合發(fā)動(dòng)機(jī)后期預(yù)測(cè)效果更好。

關(guān)鍵詞:綜合健康指數(shù);維納模型;離線參數(shù);在線參數(shù);剩余使用壽命

中圖分類號(hào):V239文獻(xiàn)標(biāo)識(shí)碼:ADOI:10.19452/j.issn1007-5453.2021.05.004

航空發(fā)動(dòng)機(jī)可靠性水平直接影響民航飛機(jī)安全運(yùn)營,適時(shí)維護(hù)修理不僅可以提高飛機(jī)安全水平,還可以降低航空公司運(yùn)營成本,增加行業(yè)競(jìng)爭(zhēng)力,故航空發(fā)動(dòng)機(jī)可靠性研究至關(guān)重要。發(fā)動(dòng)機(jī)運(yùn)行過程中會(huì)生成大量性能數(shù)據(jù),這些數(shù)據(jù)被實(shí)時(shí)監(jiān)測(cè)并上傳至綜合數(shù)據(jù)庫。運(yùn)用可靠性建模方法對(duì)這些過程數(shù)據(jù)進(jìn)行分析處理,可以預(yù)測(cè)航空發(fā)動(dòng)機(jī)退化軌跡,進(jìn)而預(yù)測(cè)發(fā)動(dòng)機(jī)剩余使用壽命,為航空公司制定視情維修決策提供依據(jù)[1-4]。

剩余使用壽命(RUL)預(yù)測(cè)是預(yù)測(cè)與健康管理(PHM)技術(shù)的重要組成部分[5-6],當(dāng)前RUL預(yù)測(cè)主要是基于數(shù)據(jù)驅(qū)動(dòng)的方法,其中包括概率分布、統(tǒng)計(jì)理論模型、機(jī)器學(xué)習(xí)等。劉帥君[7]利用結(jié)合相似性和卡爾曼濾波的方法,對(duì)CMAPSS數(shù)據(jù)集進(jìn)行了RUL預(yù)測(cè)驗(yàn)證;任子強(qiáng)等[8]對(duì)多個(gè)性能數(shù)據(jù)進(jìn)行融合處理,并通過帶線性漂移系數(shù)的維納過程預(yù)測(cè)發(fā)動(dòng)機(jī)RUL;Wang等[9]提出了一種基于健康指數(shù)(HI)進(jìn)行相似性度量的RUL預(yù)測(cè)方法,并對(duì)預(yù)測(cè)過程進(jìn)行了介紹和總結(jié)。

性能數(shù)據(jù)的選取直接影響退化建模和壽命預(yù)測(cè)精度,綜合對(duì)以上各種方法的學(xué)習(xí)研究,提出一種基于融合數(shù)據(jù)和維納建模的發(fā)動(dòng)機(jī)余壽預(yù)測(cè)方法,最后以CMAPSS數(shù)據(jù)集進(jìn)行試驗(yàn)驗(yàn)證。

1數(shù)據(jù)預(yù)處理

由于發(fā)動(dòng)機(jī)工作狀態(tài)復(fù)雜,傳感器測(cè)量存在誤差,監(jiān)測(cè)數(shù)據(jù)帶有噪聲波動(dòng),影響數(shù)據(jù)質(zhì)量,故需要對(duì)數(shù)據(jù)進(jìn)行濾波、歸一化處理。同時(shí)因?yàn)閱我恍阅苤笜?biāo)包含的性能信息非常有限,故將多元數(shù)據(jù)融合為綜合健康指數(shù)(CHI)表征發(fā)動(dòng)機(jī)健康狀態(tài)。

1.1數(shù)據(jù)濾波

卡爾曼濾波是一種結(jié)合已知數(shù)據(jù)對(duì)當(dāng)前數(shù)據(jù)去噪聲處理的方法,其主要原理是利用前一時(shí)刻狀態(tài)估計(jì)值和當(dāng)前時(shí)刻狀態(tài)觀測(cè)值計(jì)算當(dāng)前時(shí)刻狀態(tài)估計(jì)值??柭鼮V波主要包含時(shí)間更新方程和狀態(tài)更新方程兩部分,其中時(shí)間更新方程又稱為數(shù)據(jù)預(yù)估,是利用前一時(shí)刻狀態(tài)估計(jì)量對(duì)當(dāng)前狀態(tài)進(jìn)行先驗(yàn)估計(jì);而狀態(tài)更新方程又稱為數(shù)據(jù)校正,是利用當(dāng)前時(shí)刻狀態(tài)測(cè)量值和狀態(tài)先驗(yàn)估計(jì)值對(duì)當(dāng)前狀態(tài)進(jìn)行后驗(yàn)估計(jì),從而計(jì)算出當(dāng)前狀態(tài)估計(jì)值[10-11]。

卡爾曼濾波方程需要確定狀態(tài)變換矩陣和觀測(cè)模型矩陣等參數(shù),本文試驗(yàn)的濾波參數(shù)是通過將原始數(shù)據(jù)代入卡爾曼濾波并通過EM算法迭代優(yōu)化后確定。通過該方法對(duì)原始性能數(shù)據(jù)進(jìn)行濾波處理,消除外界環(huán)境對(duì)數(shù)據(jù)的隨機(jī)影響,為后續(xù)歸一化和融合提供數(shù)據(jù)基礎(chǔ)。

1.2數(shù)據(jù)歸一化

由于數(shù)據(jù)噪聲波動(dòng)較大,為了規(guī)范化數(shù)據(jù)趨勢(shì),采用卡爾曼濾波方法對(duì)各性能數(shù)據(jù)進(jìn)行濾波,以第1臺(tái)訓(xùn)練發(fā)動(dòng)機(jī)排氣溫度(EGT)數(shù)據(jù)為例,圖3是濾波后EGT增量隨著運(yùn)行時(shí)間的變化趨勢(shì),其中紅色離散點(diǎn)為原始EGT增量值,藍(lán)色曲線為濾波后EGT增量值連接成的曲線。為了進(jìn)一步標(biāo)準(zhǔn)化數(shù)據(jù),提高建模效率和預(yù)測(cè)精度,將濾波后增量數(shù)據(jù)進(jìn)行歸一化處理,結(jié)果如圖4所示。

按照以上濾波和歸一化過程,完成對(duì)11個(gè)性能數(shù)據(jù)預(yù)處理工作,然后再基于PCA分析方法將預(yù)處理后的數(shù)據(jù)融合為綜合健康指數(shù)CHI。以第一臺(tái)訓(xùn)練發(fā)動(dòng)機(jī)為例,融合后的CHI如圖5所示,通過PCA方法分析得到11個(gè)性能數(shù)據(jù)在融合過程中的權(quán)重值,見表2。

3.2離線參數(shù)估計(jì)

為了驗(yàn)證維納退化建模預(yù)測(cè)情況,在CMAPSS的FD001數(shù)據(jù)集中,選取訓(xùn)練集100臺(tái)發(fā)動(dòng)機(jī)數(shù)據(jù)進(jìn)行離線參數(shù)估計(jì),并選取測(cè)試集5臺(tái)發(fā)動(dòng)機(jī)進(jìn)行在線預(yù)測(cè)。

由于EGT可以反映發(fā)動(dòng)機(jī)工作時(shí)最惡劣站位的溫度高低,所以其對(duì)發(fā)動(dòng)機(jī)性能狀態(tài)有較好的表征,為了對(duì)比單一性能數(shù)據(jù)和融合數(shù)據(jù)對(duì)建模精度的影響,將100臺(tái)訓(xùn)練發(fā)動(dòng)機(jī)的EGT數(shù)據(jù)和融合CHI數(shù)據(jù)分別代入EM算法中計(jì)算,得到兩種方法下的三個(gè)參數(shù)離線估計(jì)值見表3。

3.3在線參數(shù)更新

為了對(duì)測(cè)試發(fā)動(dòng)機(jī)進(jìn)行在線預(yù)測(cè),根據(jù)得到的離線估計(jì)參數(shù),分別結(jié)合獲取到的測(cè)試發(fā)動(dòng)機(jī)EGT數(shù)據(jù)和融合CHI數(shù)據(jù),基于貝葉斯更新原理對(duì)模型參數(shù)進(jìn)行實(shí)時(shí)更新,并將更新后的參數(shù)代入剩余使用壽命概率密度函數(shù)以及期望公式中計(jì)算,得到5臺(tái)測(cè)試發(fā)動(dòng)機(jī)在各個(gè)運(yùn)行時(shí)間點(diǎn)的剩余壽命概率密度函數(shù)分布及其期望。以第5臺(tái)測(cè)試發(fā)動(dòng)機(jī)為例,基于融合CHI數(shù)據(jù)建模的預(yù)測(cè)結(jié)果如圖6所示,基于EGT數(shù)據(jù)建模的預(yù)測(cè)結(jié)果如圖7所示。針對(duì)最后10個(gè)運(yùn)行時(shí)間點(diǎn)的RUL預(yù)測(cè),基于融合CHI數(shù)據(jù)建模得到的概率密度分布如圖8所示,基于EGT數(shù)據(jù)建模得到的概率密度分布如圖9所示,兩種方法對(duì)5臺(tái)測(cè)試發(fā)動(dòng)機(jī)的預(yù)測(cè)均方根誤差計(jì)算結(jié)果(RMSE)見表4。

在圖6和圖7中,紫色實(shí)線表示測(cè)試發(fā)動(dòng)機(jī)RUL實(shí)時(shí)預(yù)測(cè)期望值,這是通過概率密度分布積分得到,用以表示RUL實(shí)時(shí)預(yù)測(cè)值;藍(lán)色實(shí)線表示測(cè)試發(fā)動(dòng)機(jī)RUL真實(shí)值,由于性能數(shù)據(jù)是按照固定循環(huán)采樣,剩余壽命變化呈現(xiàn)了線性單調(diào)的特征。在圖8和圖9中,z = 0平面上的圓點(diǎn)連線為運(yùn)行時(shí)間點(diǎn)的RUL真實(shí)值,三角形連線為運(yùn)行時(shí)間點(diǎn)的RUL預(yù)測(cè)值,空間中各條曲線分別為各運(yùn)行時(shí)間點(diǎn)對(duì)應(yīng)的概率密度分布變化。

通過表3可以發(fā)現(xiàn),在最后10、20和30個(gè)時(shí)間點(diǎn)的預(yù)測(cè)過程中,基于CHI數(shù)據(jù)建模方法的預(yù)測(cè)RMSE均小于基于EGT數(shù)據(jù)建模方法的預(yù)測(cè)RMSE,證明基于CHI數(shù)據(jù)的維納建模方法在后期預(yù)測(cè)中精度更高。

4結(jié)論

發(fā)動(dòng)機(jī)健康狀態(tài)與多個(gè)性能數(shù)據(jù)相關(guān),基于PCA分析方法實(shí)現(xiàn)了對(duì)多元數(shù)據(jù)的融合使用,克服了建模過程中面臨的數(shù)據(jù)篩選和使用的問題。隨著不斷獲取測(cè)試發(fā)動(dòng)機(jī)在線數(shù)據(jù),測(cè)試發(fā)動(dòng)機(jī)RUL預(yù)測(cè)誤差逐漸減小,證明該方法對(duì)發(fā)動(dòng)機(jī)運(yùn)行后期RUL預(yù)測(cè)更好。

通過對(duì)比發(fā)現(xiàn),基于融合數(shù)據(jù)建模方法的后期RUL預(yù)測(cè)誤差小于基于單一性能數(shù)據(jù)建模方法的預(yù)測(cè)誤差,證明融合CHI對(duì)發(fā)動(dòng)機(jī)性能狀態(tài)的表征更具代表性,可以更好地應(yīng)用于發(fā)動(dòng)機(jī)退化建模和剩余使用壽命預(yù)測(cè)。

參考文獻(xiàn)

[1]王紅,楊占才,靳小波,等.航空機(jī)電系統(tǒng)先進(jìn)診斷與預(yù)測(cè)技術(shù)研究[J].航空科學(xué)技術(shù), 2020,31(7): 40-47. Wang Hong, Yang Zhancai, Jin Xiaobo, et al. Research on advanced diagnosis and prognosis technology of aeronautical electromechanicalsystem[J].AeronauticalScience& Technology, 2020,31(7): 40-47. (in Chinese)

[2]Buiskikh K P,Kiselevskaya S G,Kravchuk L V,et al. Kinetics and mechanisms of degradation of the surface layers of gas-turbine engine structural components under cyclic thermal loading during initiation and propagation of thermal fatigue cracks[J]. Strength of Materials,2016,48(6):784-790.

[3]Guo Runxia,Liu Zhenghua,Wei Ye. Remaining useful life prediction for the air turbine starter based on empirical mode decomposition and relevance vector machine[J]. Transactions of the Institute of Measurement & Control,2020(1):2578-2588.

[4]Feng Xue,Piero B,Anil V,et al. An instance-based method for remaining useful life estimation for aircraft engines[J]. Journal of FailureAnalysis & Prevention,2008,8(2):199-206.

[5]馬倫,康建設(shè),趙強(qiáng).基于HMM的設(shè)備剩余壽命預(yù)測(cè)框架及其實(shí)現(xiàn)[J].計(jì)算機(jī)仿真, 2010,27(5): 88-91. Ma Lun, Kang Jianshe, Zhao Qiang. Implementation of equipment residual life prediction framework based on hidden markov model [J]. Computer Simulation, 2010,27(5): 88-91.(in Chinese)

[6]黃藍(lán),王景霖,林澤力,等.綜合系統(tǒng)健康管理關(guān)鍵技術(shù)與研究路徑[J].航空科學(xué)技術(shù), 2020,31(7): 12-17. Huang Lan, Wang Jinglin, Lin Zeli, et al. Key technologies and research paths of integrated system health management [J]. Aeronautical Science & Technology, 2020, 31(7): 12-17. (in Chinese)

[7]劉帥君.基于性能退化數(shù)據(jù)的航空發(fā)動(dòng)機(jī)剩余壽命預(yù)測(cè)[D].西安:電子科技大學(xué), 2015. Liu Shuaijun. Remaining useful life prediction of aero-engine based on performance degradation data [D]. Xian:University of Electronic Science and Technology of China, 2015. (in Chinese)

[8]任子強(qiáng),司小勝,胡昌華,等.融合多傳感器數(shù)據(jù)的發(fā)動(dòng)機(jī)剩余壽命預(yù)測(cè)方法[J].航空學(xué)報(bào), 2019,40(12):134-145. Ren Ziqiang, Si Xiaosheng, Hu Changhua, et al. Remaining useful life prediction method for engine combining multisensors data [J]. Acta Aeronautica et Astronautica Sinica, 2019, 40(12):134-145. (in Chinese)

[9]Bo Wang,Li Zhang,He Xuliang. Study on the prediction of remaining useful life of large equipment based on the similarity metric of health index[C]// International Conference on Mechanic Automation & Control Engineering,2010,2255-2258.

[10]陳逸成,何華鋒,鄧會(huì)選,等.基于Kalman濾波和Wiener過程的系統(tǒng)性能退化數(shù)據(jù)建模方法[J].電光與控制, 2015(7): 107-110. Chen Yicheng, He Huafeng, Deng Huixuan, et al. A system degradation modeling method based on Kalman filter and Wiener process [J]. Electronics Optics & Control, 2015(7):107-110. (in Chinese)

[11]彭丁聰.卡爾曼濾波的基本原理及應(yīng)用[J].軟件導(dǎo)刊, 2009,8(11): 32-34. Peng Dingcong. Basic principle and application of Kalman filter [J]. Software Guide, 2009,8(11): 32-34. (in Chinese)

[12]鐘書輝,段麗華,王煒,等.主成分分析和支持向量機(jī)在無刷直流電機(jī)故障診斷中的應(yīng)用研究[J].航空科學(xué)技術(shù), 2019,30(2): 47-52. Zhong Shuhui, Duan Lihua, Wang Wei, et al. Application research of BLDCMs fault diagnosis method based on PCA and SVM [J]. Aeronautical Science & Technology, 2019,30(2): 47-52. (in Chinese)

[13]Ghorbani S,Salahshoor K. Estimating remaining useful life of turbofan engine using data-level fusion and feature-level fusion[J]. Journal of Failure Analysis and Prevention,2020,20(1):323-332.

[14]齊琦.基于Wiener過程的退化數(shù)據(jù)建模與研究[D].西安:西安電子科技大學(xué), 2019. Qi Qi. Modeling and research of degradation data based on Wiener process [D]. Xian:Xidian University, 2019. (in Chinese)

[15]任淑紅.民航發(fā)動(dòng)機(jī)性能可靠性評(píng)估與在翼壽命預(yù)測(cè)方法研究[D].南京:南京航空航天大學(xué), 2010. Ren Shuhong. Research on methods of performance reliability assessments and life on wing prediction for civil aeroengine[D]. Nanjing: Nanjing University of Aeronautics and Astronautics, 2010. (in Chinese)

[16]張先航,李曙林,常飛,等.基于Wiener過程的航空燃油泵壽命預(yù)測(cè)[J].航空科學(xué)技術(shù), 2017,28(11): 47-53. Zhang Xianhang, Li Shulin, Chang Fei, et al. Life prediction of aviation fuel pump based on Wiener process [J]. Aeronautical Science & Technology, 2017,28(11): 47-53. (in Chinese)

[17]莊東辰,茆詩松.退化數(shù)據(jù)統(tǒng)計(jì)分析[M].北京:中國統(tǒng)計(jì)出版社, 2013. Zhuang Dongchen, Mao Shisong. Statistical analysis of degradation data[M]. Beijing:Statistics Press of China, 2013.(in Chinese)

[18]趙洪利,張猛.基于隨機(jī)維納過程的航空發(fā)動(dòng)機(jī)性能衰退研究[J].推進(jìn)技術(shù),2013(2):1-7. ZhaoHongli,ZhangMeng.Researchonperformance degradation of aeroengines based on stochastic Wiener process[J]. Journal of Propulsion Technology, 2013(2):1-7. (in Chinese)

[19]彭才華.衛(wèi)星姿態(tài)控制系統(tǒng)可靠性建模與評(píng)估方法研究[D].西安:西安理工大學(xué), 2017. Peng Caihua. Research on reliability modeling and evaluation of satellite attitude control system [D]. Xian:Xian University of Technology, 2017. (in Chinese)

[20]楊志波,董明.動(dòng)態(tài)貝葉斯網(wǎng)絡(luò)在設(shè)備剩余壽命預(yù)測(cè)中的應(yīng)用研究[J].計(jì)算機(jī)集成制造系統(tǒng), 2007(9): 1811-1815. Yang Zhibo, Dong Ming. Dynamic Bayesian networks for predicting remaining useful life of equipment [J]. Computer Integrated Manufacturing Systems, 2007(9): 1811-1815. (in Chinese)

[21]Heimes F O. Recurrent neural networks for remaining useful life estimation[C]//2008 International Conference on Prognostics and Health Management,2008.

(責(zé)任編輯陳東曉)

作者簡介

趙洪利(1964-)男,碩士,副教授。主要研究方向:發(fā)動(dòng)機(jī)健康管理、發(fā)動(dòng)機(jī)機(jī)隊(duì)管理。

Tel:13920330278

E-mail:henleytrent@163.com

鄭涅(1993-)男,碩士研究生。主要研究方向:發(fā)動(dòng)機(jī)健康評(píng)估和余壽預(yù)測(cè)。

Tel:15662605606

E-mail:nzheng.12@foxmail.com

Engine RUL Prediction Based on the Combination of Fusing Data and Wiener Modeling

Zhao Hongli,Zheng Nie*

Civil Aviation University of China,Tianjin 300300,China

Abstract: The performance data is an important indicator of engines health status, and the analysis of the performance data can be used to predict the engines remaining useful life (RUL), and which will provide basis for making engine maintenance decisions. The health of an engine is closely related to the monitored data. Based on the data of training and testing engines, the principal component analysis (PCA) was used to fuse multivariate data into a comprehensive health index (CHI). Wiener process was used to construct the performance degradation model, and the off-line parameters are optimized iteratively by EM algorithm with the training engines data. Based on Bayesian method, the training engines data are used to upgrade the parameters of the degradation model, and the probability density distribution and expected RUL of the testing engines are calculated at the real time. The comparison shows that the average of root mean square error (RMSE) for the last 10 cycles of the testing engines are 12.95 by using the model with single data and 5.34 by using the model with fusion data respectively, which proves that the modeling method with fusion data is more accurate.

Key Words: composite health index; Wiener model; off-line parameters; on-line parameters; RUL

大田县| 新晃| 衡南县| 滦平县| 怀远县| 平阳县| 岐山县| 延津县| 江山市| 靖边县| 绥滨县| 沙坪坝区| 乌海市| 永寿县| 聊城市| 子洲县| 东平县| 延安市| 额尔古纳市| 伊川县| 谢通门县| 凌海市| 鄂伦春自治旗| 武功县| 遂溪县| 鄄城县| 奉贤区| 永和县| 古田县| 南宫市| 呈贡县| 蓬莱市| 密云县| 安丘市| 沁水县| 绥德县| 启东市| 东方市| 壤塘县| 聂拉木县| 襄垣县|