毛文貴 李建華 劉桂萍
摘要: 軸承轉(zhuǎn)子系統(tǒng)不平衡量識(shí)別過(guò)程中,在輸出響應(yīng)和模型中存在的不確定性參數(shù)一般采用概率法描述,通過(guò)貝葉斯理論獲得不平衡量的聯(lián)合后驗(yàn)概率密度分布時(shí)涉及大量采樣。針對(duì)采樣效率,提出了基于遺傳智能采樣技術(shù)改進(jìn)貝葉斯理論。首先,以代價(jià)函數(shù)作為指示因子通過(guò)信賴(lài)域模型管理方法不斷更新先驗(yàn)空間使其覆蓋高密度后驗(yàn)空間,然后通過(guò)智能布點(diǎn)技術(shù)和樣本遺傳策略以有限的樣本點(diǎn)集中呈現(xiàn)在聯(lián)合后驗(yàn)概率密度分布的高密度區(qū)域,提高信賴(lài)域上關(guān)鍵區(qū)域的精度,從而加快收斂速度,減小耗時(shí)的正問(wèn)題調(diào)用次數(shù)。最后將其應(yīng)用于識(shí)別具有不平衡量先驗(yàn)信息和帶有隨機(jī)噪聲的測(cè)試響應(yīng)的滑動(dòng)軸承-轉(zhuǎn)子系統(tǒng)的不平衡量,獲得不平衡量的均值、置信區(qū)間。案例顯示能準(zhǔn)確快速地抽樣,提高了貝葉斯識(shí)別的計(jì)算效率。
關(guān)鍵詞: 滑動(dòng)軸承-轉(zhuǎn)子系統(tǒng); 不平衡量; 貝葉斯理論; MCMC法; 遺傳智能采樣技術(shù)
中圖分類(lèi)號(hào): TH133.31; O347.6 ?文獻(xiàn)標(biāo)志碼: A ?文章編號(hào): 1004-4523(2019)04-0660-08
DOI:10.16385/j.cnki.issn.1004-4523.2019.04.013
引 言
滑動(dòng)軸承-轉(zhuǎn)子系統(tǒng)材質(zhì)的不均勻性、制造和安裝過(guò)程引起的變形以及工作中的磨損等都會(huì)引起軸承轉(zhuǎn)子系統(tǒng)的不平衡振動(dòng)。利用不平衡量識(shí)別技術(shù)進(jìn)行軸系動(dòng)平衡來(lái)提高轉(zhuǎn)子及其構(gòu)成的產(chǎn)品質(zhì)量,減小噪聲和振動(dòng),提高軸承的使用壽命,以保證軸系運(yùn)行的長(zhǎng)期性和穩(wěn)定性,是軸承轉(zhuǎn)子系統(tǒng)經(jīng)常使用的一種校準(zhǔn)方式。但由于影響因素的多樣性和復(fù)雜性,不平衡量識(shí)別結(jié)果存在一定的誤差。對(duì)于滑動(dòng)軸承-轉(zhuǎn)子系統(tǒng),轉(zhuǎn)子幾何特征、滑動(dòng)軸承油膜特性系數(shù)和測(cè)量響應(yīng)的隨機(jī)性這些不確定性因素即使是在較小情況下,也很有可能導(dǎo)致不平衡量識(shí)別結(jié)果產(chǎn)生較大的偏差。而工程實(shí)際中,工程師能根據(jù)經(jīng)驗(yàn)和知識(shí)對(duì)不平衡量參數(shù)在尚未獲取實(shí)驗(yàn)測(cè)量信息之前有一定的預(yù)先估計(jì)。如何利用這些先驗(yàn)信息減小不確定因素對(duì)待識(shí)別參數(shù)的影響成為不確定性反問(wèn)題領(lǐng)域的研究熱點(diǎn)[1-3]。姜雪等[4]利用最大似然法對(duì)影響齒輪傳動(dòng)疲勞壽命的分布參數(shù)進(jìn)行不確定性識(shí)別分析,求得分布參數(shù)的最大似然估計(jì)值。對(duì)于工程不確定性反問(wèn)題,最大似然法[5-6]在處理隨機(jī)性不確定問(wèn)題時(shí)考慮了測(cè)量數(shù)據(jù)的隨機(jī)性對(duì)待識(shí)別參數(shù)結(jié)果的影響,通過(guò)獲得待識(shí)別參數(shù)的最大似然估計(jì)值來(lái)計(jì)算相應(yīng)的置信區(qū)間,充分利用了已知不確定性參數(shù)的樣本信息,但未考慮待識(shí)別參數(shù)的先驗(yàn)信息。貝葉斯理論[7-11]對(duì)結(jié)構(gòu)模型參數(shù)進(jìn)行反求分析,同時(shí)考慮了已知參數(shù)概率密度樣本信息和未知參數(shù)先驗(yàn)信息。但貝葉斯理論求解工程不確定性反問(wèn)題常常涉及非常耗時(shí)的正問(wèn)題計(jì)算,難以滿(mǎn)足實(shí)際工程對(duì)計(jì)算效率的要求。Zhang等[12]提出基于自適應(yīng)近似加密技術(shù)的馬爾科夫鏈蒙特卡羅(MCMC)法,其主要思想是構(gòu)建高精度的未知參數(shù)后驗(yàn)空間自適應(yīng)近似模型,進(jìn)而避免傳統(tǒng) MCMC 法調(diào)用耗時(shí)的仿真模型。但近似模型的構(gòu)建精度也會(huì)影響參數(shù)識(shí)別精度。本文針對(duì)貝葉斯理論構(gòu)建不平衡量聯(lián)合后驗(yàn)概率分布時(shí)要生成大量的抽樣點(diǎn),并要大量調(diào)用耗時(shí)的正問(wèn)題計(jì)算,造成效率低的問(wèn)題。提出基于遺傳智能采樣技術(shù)[13]使采集樣本點(diǎn)集中呈現(xiàn)在真實(shí)聯(lián)合后驗(yàn)概率密度分布的高密度區(qū)域,提高采樣效率,并遺傳有效樣本點(diǎn),減少正問(wèn)題調(diào)用次數(shù)。同時(shí),采用快速的傳遞矩陣法計(jì)算滑動(dòng)軸承轉(zhuǎn)子系統(tǒng)不平衡響應(yīng)以提高計(jì)算效率,從而改進(jìn)貝葉斯理論識(shí)別滑動(dòng)軸承-轉(zhuǎn)子系統(tǒng)的不平衡量。
由公式(3)獲得未知參數(shù)不平衡量的聯(lián)合后驗(yàn)概率密度分布,而質(zhì)量m,偏心距e,相位φ各自的后驗(yàn)概率密度分布才是關(guān)注的重點(diǎn),即邊緣后驗(yàn)概率密度分布。則要對(duì)公式(3)進(jìn)行2階積分處理。由于工程實(shí)際的復(fù)雜性,積分會(huì)產(chǎn)生維數(shù)災(zāi)難,計(jì)算量會(huì)隨未知參數(shù)維數(shù)的增加成指數(shù)倍的增長(zhǎng)。常用數(shù)值抽樣法統(tǒng)計(jì)未知參數(shù)聯(lián)合概率密度分布空間中的樣本信息來(lái)獲得邊緣后驗(yàn)概率密度分布的近似解。數(shù)值抽樣法中馬爾科夫鏈蒙特卡羅(MCMC法)是較為實(shí)用的算法,通過(guò)隨機(jī)游走(馬爾科夫鏈),利用抽樣點(diǎn)之間的相關(guān)性來(lái)產(chǎn)生大量有用的樣本點(diǎn),保證更多樣本點(diǎn)落在最重要的區(qū)域。一般鏈長(zhǎng)即抽樣點(diǎn)數(shù)目取為105量級(jí)時(shí),MCMC法才可能得到邊緣后驗(yàn)概率密度分布較為準(zhǔn)確的近似解。鏈長(zhǎng)越長(zhǎng)對(duì)于本文中不平衡量識(shí)別問(wèn)題,意味著調(diào)用不平衡響應(yīng)計(jì)算次數(shù)越多。為了提高效率,本文采用不平衡響應(yīng)計(jì)算效率比較高的傳遞矩陣法[15]。
1.2 遺傳智能采樣技術(shù)
貝葉斯理論構(gòu)建未知參數(shù)聯(lián)合概率密度獲得代價(jià)函數(shù)時(shí),為了要獲得高精度的聯(lián)合后驗(yàn)概率密度分布空間要進(jìn)行大量的采樣。一種有效的做法是使在先驗(yàn)分布空間產(chǎn)生的用于構(gòu)建代價(jià)函數(shù)所需的有限樣本點(diǎn)集中呈現(xiàn)在真實(shí)聯(lián)合后驗(yàn)概率密度分布的高密度區(qū)域。實(shí)際工程中,通過(guò)給定的不平衡量參數(shù)先驗(yàn)分布是相對(duì)真實(shí)區(qū)間較大的空間,因此,采樣時(shí)要不斷調(diào)整不平衡量先驗(yàn)分布空間以便最大限度地集中有限的樣本點(diǎn)反映真實(shí)后驗(yàn)空間高密度區(qū)域,進(jìn)而保證不平衡量聯(lián)合后驗(yàn)概率分布的精度。本文基于遺傳智能采樣技術(shù)采樣,通過(guò)信賴(lài)域模型管理方法探測(cè)非支配解區(qū)域,不斷更新不平衡量先驗(yàn)分布空間,來(lái)保證獲得與真實(shí)解接近的區(qū)間。通過(guò)智能布點(diǎn)技術(shù)和樣本遺傳策略使各個(gè)信賴(lài)域上的樣本均勻分布,通過(guò)遺傳部分樣本落入下代信賴(lài)域的點(diǎn)作為智能布點(diǎn),減小正問(wèn)題計(jì)算次數(shù)而提高信賴(lài)域上關(guān)鍵區(qū)域的精度,從而加快收斂速度。其關(guān)鍵步驟如下:
1.2.1 根據(jù)信賴(lài)域更新方法確定下代信賴(lài)域區(qū)域
1.2.2 樣本遺傳智能布點(diǎn)策略
下代信賴(lài)域與當(dāng)代的信賴(lài)域會(huì)有重合的區(qū)域。當(dāng)代的樣本點(diǎn)可能會(huì)落入下代信賴(lài)域里。將遺傳的舊樣本和遺傳拉丁超立方實(shí)驗(yàn)設(shè)計(jì)(ILHD)產(chǎn)生的新樣本組合起來(lái)作為下代代價(jià)函數(shù)構(gòu)建的樣本點(diǎn),這樣可以大大減少需要做仿真計(jì)算的不平衡響應(yīng)的總樣本的個(gè)數(shù),提高計(jì)算效率。如果將其全部遺傳給下一步,會(huì)產(chǎn)生一部分區(qū)域樣本過(guò)于緊湊而不利于智能布點(diǎn)的樣本均勻分布。樣本遺傳策略在盡可能充分利用遺傳的舊樣本的原則下根據(jù)極大極小距離準(zhǔn)則對(duì)實(shí)驗(yàn)設(shè)計(jì)進(jìn)行優(yōu)化,篩選部分落入到下代信賴(lài)域的樣本,采用模擬退火優(yōu)化求解器使新樣本在各個(gè)設(shè)計(jì)變量上投影均勻且產(chǎn)生的新樣本到遺傳樣本的距離最大,遺傳樣本和新樣本在下代信賴(lài)域區(qū)域保持空間均布性、投影均勻性。
[2] Zhang Kai, Wang Zengfei, Zhang Liming. A hybrid optimization method for solving Bayesian inverse problems under uncertainty[J]. PloS One, 2015, 10(8): e0132418.
[3] Wang Hu, Zeng Yang, Yu Xiancheng, et al. Surrogate-assisted Bayesian inference inverse material identification method and application to advanced high strength steel[J]. Inverse Problems in Science and Engineering, 2016, 24(7): 1133-1161.
[4] 姜 雪, 趙繼俊. 汽車(chē)變速箱齒輪可靠性試驗(yàn)數(shù)據(jù)分布參數(shù)估計(jì)[J].機(jī)械設(shè)計(jì)與制造, 2009,(4):21-23 .
Jiang Xue, Zhao Jijun. Distribution parameter estimation for the reliability test data of transmission gears [J]. Machinery Design & Manufacture, 2009,(4):21-23.
[5] Kian Jafari .A parameter estimation approach based on binary measurements using Maximum Likelihood analysis—Application to MEMS[J]. International Journal of Control, Automation and Systems, 2017, 15(2):716-721.
[6] Ba Bin, Cui Weijia,Wang Daming,et al. Maximum likelihood time delay estimation based on Monte Carlo importance sampling in multipath environment[J]. International Journal of Antennas and Propagation, 2017, 2017:1-7.
[7] 王榮橋,劉 飛,胡殿印,等.基于貝葉斯理論的低循環(huán)疲勞壽命模型不確定性量化[J]. 航空學(xué)報(bào), 2017, (9):244-253.
Wang Rongqiao, Liu Fei, Hu Dianyin. et al. Uncertainty quanti-fication in low cycle fatigue life model based on Bayesian theory[J].Acta Aeronautica et Astronautica Sinica, 2017, (9):244-253.
[8] 尹 彬,胡祥云. 非線(xiàn)性反演的貝葉斯方法研究綜述[J].地球物理學(xué)進(jìn)展, 2016,(3):1027-1032.
Yin Bin,Hu Xiangyun. Overview of nonlinear inversion using Bayesian method [J]. Progress in Geophysics, 2016, (3):1027-1032.
[9] Wei Yaobing, Wang Xin, Kong Yuanyuan, et al. A probability uncertainty method of fault classification for steam turbine generator set based on Bayes and Holospectrum[J]. Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science, 2016,230(10):3767-3776.
[10] Tyminski N C,Tuckmantel F W S, Cavalca K L, et al. Bayesian inference applied to journal bearing parameter identification [J]. Journal of the Brazilian Society of Mechanical Sciences and Engineering, 2017, 39(8): 2983-3004.
[11] Shao Jiye, Xu Minqiang, Wang Rixing, et al. Bayesian model-based fault diagnosis for the rotor [J]. Aircraft Engineering and Aerospace Technology, 2009, 81(1): 19-24.
[12] Zhang W, Liu J,Cho C, et al. A fast Bayesian approach using adaptive densifying approximation technique accelerated MCMC. Inverse Problems in Science and Engineering, 2016, 24(2): 247-264.
[13] 韓 旭, 姜 潮, 陳國(guó)棟, 等.基于代理模型的汽車(chē)結(jié)構(gòu)安全多目標(biāo)優(yōu)化的研究報(bào)告[J].科技創(chuàng)新導(dǎo)報(bào), 2016, 13(19): 179-180.
Han Xu, Jiang Chao, Chen Guodong, et al. Multi-objective optimization method based on metamodel for vehicle structural safety[J].Science and Technology Innovation Herald,2016, 13(19): 179-180.
[14] Friswell M I, Perny J E T, Garvey S D, et al. Dynamics of Rotating Machines[M].Cambridge: Cambridge University Press, 2010.
[15] Mao Wengui, Han Xu, Liu Guiping, et al.Bearing dynamic parameters identification of a flexible rotor-bearing system based on transfer matrix method[J]. Inverse Problems in Science & Engineering, 2016, 24(3): 372-392.
[16] 聞邦椿,顧家柳,夏松波,等.高等轉(zhuǎn)子動(dòng)力學(xué)[M].北京:機(jī)械工業(yè)出版社,2000:182-197.
Wen Bangchun, Gu Jialiu, Xia Songbo, et al. Higher Rotor Dynamics[M].Beijing: Mechanical Industry Press, 2000:182-197.
[17] 陳國(guó)棟.基于代理模型的多目標(biāo)優(yōu)化方法及其在車(chē)身設(shè)計(jì)中的應(yīng)用[D].長(zhǎng)沙:湖南大學(xué), 2012.
Chen Guodong. Multi-objective optimization method based on metamodel and its applications in vehicle body design[D]. Changsha: Hunan University, 2012.
[18] Tarantola A. Inverse Problem Theory and Methods for Model Parameter Estimation[M]. Philadelphia: SIAM, 2005.
[19] 毛文貴.高速電主軸滑動(dòng)軸承-轉(zhuǎn)子系統(tǒng)動(dòng)力學(xué)關(guān)鍵特性參數(shù)識(shí)別[D] .長(zhǎng)沙: 湖南大學(xué), 2015.
Mao Wengui. Identification of the key character-istic parameters in dynamics for a sliding bearing-rotor system of high speed spindle[D].Changsha: Hunan University, 2015.
Abstract: Probability method is used to describe the uncertainty of the output and model in the unbalanced identification process of the sliding bearing-rotor system, the Bayesian theory is used to obtain the joint posterior probability density distribution of the unbalance parameters, which involves massive sampling. A novel algorithm based on genetic intelligent sampling technique is presented to promote the efficiency. In this algorithm, Trust region model management method is firstly used to update the prior space to cover the high-density posterior space by calling the cost function as an indicator. Then the finite sample points are concentrated in the high-density region of joint posterior probability density distribution by intelligent placement technology and sample genetic strategy in order to improve the accuracy of critical areas on trust which can speed up convergence and reduce the number of calling time-consuming positive problem. Finally, the presented method is applied to identify the mean value and confidence interval of the unbalance parameters of the sliding bearing-rotor system, which has unbalanced prior information, and test response with random noise. In the work, the sampling algorithm based on the genetic intelligent sampling technique can promote the efficiency of Bayesian approach for fast identifying the unbalance parameters.
Key words: sliding bearing-rotor system; unbalance parameters; Bayesian theory; Markov Chain Monte Carlo; genetic intelligent sampling technique
作者簡(jiǎn)介: 毛文貴(1975-),女,博士,副教授。電話(huà):(0731)58688521;E-mail:maowengui@hnie.edu.cnZ ··y^