劉巧斌,史文庫,陳志勇,商國旭
?
工程車輛車橋位移譜統(tǒng)計分布建模及分步參數(shù)識別
劉巧斌,史文庫,陳志勇※,商國旭
(吉林大學汽車工程學院,汽車仿真與控制國家重點實驗室,長春 130022)
針對非公路用車的車橋實測位移譜統(tǒng)計分布建模中模型選擇、參數(shù)識別的初值選取主觀性大和計算效率低等難題,該文以實測的車橋位移信號為研究對象,分別進行時域分析、頻域功率譜分析,對信號進行分組,統(tǒng)計頻數(shù),獲得統(tǒng)計直方圖和累計概率分布曲線。分別采用正態(tài)分布、雙峰正態(tài)分布、威布爾分布和雙峰威布爾分布模型對位移譜進行建模,提出分步參數(shù)識別方法。引入灰色關聯(lián)度目標函數(shù),以人工魚群算法獲得的參數(shù)作為模型參數(shù)的初始值,采用迭代非線性最小二乘法levenberg-marquardt(LM)算法進行精確參數(shù)識別,使用相關系數(shù)和kolmogorov-smirnov(KS)檢驗對各模型的擬合優(yōu)度進行比較。結果表明,混合威布爾分布與統(tǒng)計直方圖的相關系數(shù)為(0.9800,0.9908,0.9867,0.9665),混合正態(tài)分布為(0.9793,0.9904,0.9783,0.9661),威布爾模型為(0.8613,0.9113, 0.8618,0.8854),正態(tài)模型為(0.8611,0.9127,0.8624,0.8869),混合威布爾模型可以對車橋位移譜進行高精度擬合,而所提出的分步參數(shù)識別法可以高效、準確地進行模型的參數(shù)識別。研究結果可為車輛疲勞載荷譜的編制和臺架試驗提供參考。
模型; 參數(shù)識別;車橋位移譜;灰色關聯(lián);非線性最小二乘法;混合威布爾模型
非公路用車的車橋位移譜統(tǒng)計分布研究是進一步進行載荷譜編制和疲勞可靠性臺架試驗的基礎。正態(tài)分布和威布爾分布是在可靠性工程中應用較為廣泛的2種概率統(tǒng)計分布模型[1-4]。一般單獨的正態(tài)分布或威布爾分布并不完全適用于所有的可靠性數(shù)據(jù)建模,而混合模型通過將實際分布分解為2個或2個以上的獨立分布,采用加權疊加的思想去逼近實際分布,具有很強的實際應用價值,受到了越來越多學者的重視。混合分布模型的引入帶來了諸多挑戰(zhàn),而模型參數(shù)識別難度的增加是其中最為主要的一項,尋找簡單、高效而精確的混合分布模型參數(shù)識別方法已經(jīng)成為可靠性研究領域的一個焦點。
傳統(tǒng)的可靠性模型參數(shù)識別方法有圖解法、非線性最小二乘法、最大似然估計法和貝葉斯估計法等[5-12]。這些算法存在的主要弊端有:①計算效率有待提高,傳統(tǒng)算法大多依賴迭代求解,為提高參數(shù)辨識精度,一定程度上犧牲了計算效率,增加了參數(shù)辨識的時間成本。②參數(shù)識別優(yōu)化目標的選取不當,現(xiàn)有研究,多數(shù)將參數(shù)識別的目標函數(shù)定義為模型與實測數(shù)據(jù)的平方和,這樣不可避免地忽略了樣本點和仿真點之間橫坐標的仿真誤差,而只考慮了縱坐標的仿真誤差。③參數(shù)識別的經(jīng)驗依賴度高,參數(shù)識別的初始值對結果影響大。
智能算法在多維非線性、初始值不易確定的模型的參數(shù)識別問題上展現(xiàn)出了巨大的潛力,許多學者采用智能優(yōu)化算法對模型參數(shù)識別問題進行了大量的研究[13-27]。智能算法應用于參數(shù)識別存在的問題是,不同的智能算法對解決同一問題的效率和適用性不盡相同,有必要針對具體問題選取一種最適合的智能算法或針對問題對算法進行改進,保證算法在具有較高精度的同時提高算法效率。
針對以上問題,本文以實測礦用車輛的車橋位移譜統(tǒng)計分布建模及參數(shù)識別為例,提出一種新型參數(shù)識別方法,引入灰色關聯(lián)度目標函數(shù),應用了分步識別的思想,采用人工魚群算法進行參數(shù)的初步估計,在此基礎上,使用levenberg-marquardt(LM)算法進行參數(shù)的精確識別,分別建立正態(tài)分布、威布爾分布、混合正態(tài)分布和混合威布爾分布模型對實測數(shù)據(jù)擬合逼近,采用相關系數(shù)和kolmogorov-smirnov(KS)檢驗評價指標對各模型的擬合優(yōu)度進行評價。
為獲取某礦用車輛的實際位移譜,在車輛的實際使用道路上進行車橋位移譜的采集,圖1a所示是試驗所用車輛和試驗場地的具體情況。試驗在某采礦區(qū)進行。試驗車在山下裝載,載貨約40 t,然后運至山頂某卸料廠卸載,再空車原路返回裝載處,往返距離約3 km,平均車速為10 km/h,完成1個完整的采集循環(huán),1個周期的信號采集時間為1 200 s。試驗路面為未鋪裝砂石路,平均坡度10%,最大坡度17%。試驗中采用IMC多通道數(shù)據(jù)采集設備采集試驗數(shù)據(jù)。圖1b所示是車橋傳感器的安裝圖,所使用的傳感器為CELESCO拉繩式位移傳感器,試驗所測得的位移為輪胎與車架縱梁之間的垂向相對位移。試驗時,數(shù)據(jù)采集頻率為200 Hz。試驗共采集5個循環(huán)的載荷數(shù)據(jù)。由于實測的試驗數(shù)據(jù)存在干擾,應對其進行濾波、剔除奇異值和消除趨勢項等預處理。最終選取1組穩(wěn)定的數(shù)據(jù)作為后續(xù)處理的樣本。
圖1 試驗車輛和傳感器的安裝
圖2所示是采集到的位移信號的時間歷程曲線,由圖2可知,中橋、后橋上布置的4個測點的位移呈現(xiàn)同樣的時程規(guī)律,即信號存在2個不同幅值,在200~800 s的時間內(nèi),位移幅值為200 mm左右,在900~1 150 s的時間內(nèi),位移幅值大于第一段時間內(nèi)的幅值。
圖2 實測位移譜時間歷程
為觀察位移信號的頻域特性,對采集到的時域信號求自相關函數(shù)的傅里葉變換,獲得功率譜密度(power spectral density,PSD),圖3所示是4個位移測點獲得信號的功率譜密度曲線,由圖3可知,位移信號呈現(xiàn)低頻聚集性,隨著頻率增加,功率譜密度減小,不存在明顯的共振峰,說明路面的位移激勵是寬頻的隨機信號。因此,為了定量分析位移信號的宏觀規(guī)律,并進行高精度的可靠性仿真分析和臺架試驗研究,采用統(tǒng)計學方法對位移譜進行統(tǒng)計分布研究具有較大的理論意義和實際工程應用價值。
圖3 位移信號的功率譜密度
為研究位移譜統(tǒng)計規(guī)律,在對數(shù)據(jù)進行初步的統(tǒng)計直方圖分析后,選取工程中常用的正態(tài)分布、威布爾分布及這2種分布對應的混合分布作為位移譜的統(tǒng)計分布模型,以下分別介紹正態(tài)分布、威布爾分布和混合分布的數(shù)學模型。
正態(tài)分布的密度函數(shù)為
正態(tài)分布的分布函數(shù)為
三參數(shù)威布爾分布的密度函數(shù)為
三參數(shù)威布爾分布的分布函數(shù)為
混合分布是由若干個單一分布線性加權疊加形成?;旌戏植嫉拿芏群瘮?shù)和分布函數(shù)分別如式(5)和式(6)所示。
合理的分布模型選取是統(tǒng)計建模的基礎,而對于確定的某一分布模型能否準確描述客觀事物的規(guī)律,對分布模型的參數(shù)進行精確識別是關鍵所在。為進行車橋位移譜統(tǒng)計分布的參數(shù)識別,本文提出分步參數(shù)識別方法,在人工魚群算法獲得的粗略的參數(shù)值的基礎上,進一步采用LM算法進行參數(shù)的精確識別。以下分別對人工魚群算法和LM算法進行介紹。
人工魚群算法(artificial fish school algorithm, AFSA)是一種基于動物行為的群體智能優(yōu)化算法。該算法通過模擬魚類的覓食、聚群、追尾、隨機等行為在搜索域內(nèi)進行迭代尋優(yōu),是集群思想的一個成功應用[28-30]。
人工魚群算法的主要行為有魚群初始化、覓食行為、聚群行為、追尾行為和隨機行為。算法主要步驟如下
本文采用人工魚群算法進行參數(shù)的初步估計,經(jīng)反復測試并參考文獻資料[29],將算法的運行參數(shù)設置如表1所示時,參數(shù)估計結果較為理想。
表1 人工魚群算法運行參數(shù)
Levenberg-marquardt(LM)算法是梯度優(yōu)化迭代求解方法的一種。LM算法采用目標函數(shù)的二階微分,并采用了方向矢量的方法對收斂方向進行動態(tài)調整,以此增加收斂性能,同時保證較好的收斂速度。式(13)所示為LM算法的變量迭代公式[11]。
LM算法的主要步驟如下
3.3.1 目標函數(shù)
選定一個合適的目標函數(shù),是進行分布模型參數(shù)識別的前提?,F(xiàn)有的參數(shù)識別目標函數(shù)絕大部分都是以模型和實測曲線的誤差平方和作為優(yōu)化目標,不可避免的存在只考慮縱軸方向誤差而忽略橫軸方向誤差,一些學者提出的全最小二乘法方法一定程度上緩解了最小二乘誤差計算的固有弊端[14]。本文引入灰色關聯(lián)度目標函數(shù),以實測和模型之間的灰色關聯(lián)度最大化作為第一步參數(shù)識別的目標函數(shù)進行參數(shù)的初步識別。
灰色關聯(lián)度分析是灰色系統(tǒng)理論的重要組成部分[31-32]。采用灰色關聯(lián)度評價模型和實測曲線的接近程度,可以實現(xiàn)所辨識模型和實測曲線的宏觀幾何相似程度的最大化,從而在保證了參數(shù)識別結果的精確性。
灰色關聯(lián)度目標函數(shù)的計算過程如下
1)求實測數(shù)據(jù)和模型計算結果的歸一化序列,如式(16)。
2)求歸一化后的2個數(shù)據(jù)序列之間的絕對差序列,如式(17)。
3)求絕對差序列的最值,如式(18)。
4)計算關聯(lián)系數(shù),如式(19)。
3.3.2 參數(shù)識別流程
本文所提出的分步參數(shù)識別方法流程如圖4所示。
圖4 分步參數(shù)識別流程
根據(jù)統(tǒng)計學中直方圖的分組經(jīng)驗,將實測位移信號從小到大分為50個區(qū)間,統(tǒng)計每個區(qū)間的頻數(shù)和累計頻數(shù),作為位移譜統(tǒng)計分布實測的概率密度和分布函數(shù)值。采用本文提出的分步參數(shù)識別方法分別對4個測點獲得的位移譜進行4種不同模型的參數(shù)識別,表2為各模型的參數(shù)識別結果。
圖5~圖8是4個測點位移譜的擬合結果,由圖5~圖8可知,威布爾模型的擬合效果優(yōu)于正態(tài)分布模型,而混合模型的擬合效果優(yōu)于單一分布模型,4種模型之中,混合威布爾模型的逼近精度最高。以中橋左側位移譜為例,單一分布模型的概率密度誤差最大值為0.013,而混合分布模型的概率密度誤差為0.006;單一分布模型的累計概率誤差最大值為0.091,而混合分布模型的累計概率誤差最大值為0.052??梢?,混合分布的擬合精度普遍大于單一分布,單一分布將數(shù)據(jù)的實際分布“均勻化”,從而抹去了數(shù)據(jù)中相對較小的峰值,而只保留了最大峰值。在2種混合分布中,混合威布爾模型擬合精度高于混合正態(tài)分布,混合正態(tài)分布的概率密度曲線和累計概率分布圖都與頻數(shù)統(tǒng)計的趨勢十分的吻合。
4.2.1 相關系數(shù)
表2 4種不同模型的參數(shù)識別結果
Tabel 2 Parameter estimation results of 4 models
模型 Model中橋左側 Left side of middle bridge中橋右側 Right side of middle bridge后橋左側 Left side of rear bridge后橋右側 Right side of rear bridge 正態(tài)分布 Normal distribution 混合正態(tài)分布 Mixed normal distribution 威布爾分布 Weibull distribution 混合威布爾分布 Mixed Weibull distribution
注:為正態(tài)分布的位置參數(shù);為正態(tài)分布的形狀參數(shù);為威布爾分布的尺度參數(shù)為威布爾分布的形狀參數(shù);為威布爾分布的位置參數(shù);為子分布的權重比例。
Note:is the positional parameter of normal distribution;is the shape parameter of normal distribution;is the scale parameter of Weibull distribution;is the shape parameter of Weibull distribution;is the positional parameter of Weibull distribution;is the weight ratio of the sub-distributions.
圖5 中橋左側位移譜分布曲線
圖6 中橋右側位移譜分布曲線
圖7 后橋左側位移譜分布曲線
圖8 后橋右側位移譜分布曲線
4.2.2 KS檢驗
4.2.3 擬合優(yōu)度檢驗結果
以上所述的相關系數(shù)和KS檢驗統(tǒng)計量分別從概率密度函數(shù)的曲線擬合效果和概率分布函數(shù)曲線的擬合效果上對擬合優(yōu)度進行了檢驗。表3所示是各模型的擬合優(yōu)度指標,從表3可知,混合分布的相關系數(shù)均在0.95以上,最大KS值不大于0.25。因此,混合分布的擬合效果優(yōu)于單一分布,而威布爾分布的擬合效果優(yōu)于正態(tài)分布。
表3 4種不同模型的擬合優(yōu)度對比
注:為相關系數(shù);為KS值。
Note:is the correlation coefficient andis the KS value.
為了對混合模型擬合精度高的原因及其內(nèi)在組成規(guī)律進行具體分析,以中橋左側位移譜的混合威布爾分布為例,對混合模型進行分解。圖9a和圖9b分別是混合模型分解出的分布密度函數(shù)曲線和累計分布概率曲線,由圖9可知,混合分布實現(xiàn)了獨立分布的加權線性疊加,曲線的形狀由組成子分布共同決定,且在不同的區(qū)間內(nèi),各個子分布的影響程度不同。由圖9a的分布密度函數(shù)曲線可知,在第一個子分布的峰值鄰域內(nèi),第一個子分布對曲線的影響占主要地位,而在第二個子分布的峰值鄰域內(nèi),第二個子分布對曲線的影響是主要的。由圖9b的累計分布概率函數(shù)曲線可知,累計分布呈現(xiàn)2個不同的上升斜率,第一段斜率主要由第一個子分布決定,而第二段上升斜率主要由第二個子分布決定。
注:f(x)為混合概率密度;f1(x)為子分布1的概率密度;f2(x)為子分布2的概率密度;為混合累計概率;為子分布1的累計概率;為子分布2的累計概率。
本文研究的主要是雙重混合分布,而對于多重混合分布,其規(guī)律可由二重混合分布推廣。
1)所采集到的車橋位移譜呈現(xiàn)出雙峰規(guī)律,采用混合威布爾模型可以較好的進行描述,且混合威布爾分布的各項擬合優(yōu)度指標均優(yōu)于正態(tài)模型、威布爾模型和混合正態(tài)模型, 混合分布的相關系數(shù)均在0.95以上,最大KS值不大于0.25;
2)以灰色關聯(lián)系數(shù)作為初步參數(shù)識別的目標函數(shù),可以保證擬合曲線和原曲線的幾何相似程度最大,以全最小二乘誤差作為LM算法的優(yōu)化目標,解決了混合可靠性模型參數(shù)識別中目標函數(shù)的選取問題;
3)提出的分步參數(shù)識別方法,綜合了人工魚群算法這種智能優(yōu)化算法和傳統(tǒng)迭代算法的優(yōu)點,以人工魚群算法優(yōu)化解作為LM算法的初始值,解決了非線性最小二乘參數(shù)識別法的初值選取困難問題??蔀橄嚓P的參數(shù)識別問題提供參考。
[1] 薛廣進,李強,王斌杰,等. 軌道車輛結構動應力譜分布的估計[J]. 機械工程學報,2013,49(4):102-105. Xue Guangjin, Li Qiang, Wang Binjie, et al. Estimation of dynamic stress spectrum distribution of rail vehicle structure [J]. Journal of Mechanical Engineering, 2013, 49 (4): 102-105. (in Chinese with English abstract)
[2] 翟新婷,張曉晨,江柱錦,等. 基于混合分布的輪式裝載機半軸載荷譜編制[J]. 農(nóng)業(yè)工程學報,2018,34(8):78—84. Zhai Xinting, Zhang Xiaochen, Jiang Zhujin, et al. Compilation of half-axle load spectrum of wheel loader based on mixed distribution[J]. Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE), 2018, 34 (8): 78—84. (in Chinese with English abstract)
[3] T Bu?ar, M Nagode, M Fajdiga. Reliability approximation using finite Weibull mixture distributions[J]. Reliability Engineering & System Safety, 2004, 84(3): 241-251.
[4] Hagey T J, Puthoff J B, Crandell K E, et al. Modeling observed animal performance using the Weibull distribution [J]. Journal of Experimental Biology, 2016, 219(11): 1603-1607.
[5] Davies I J. Unbiased estimation of the Weibull scale parameter using linear least squares analysis[J]. Journal of the European Ceramic Society, 2017, 37(8): 2973-2981.
[6] Wais P. Two and three-parameter Weibull distribution in available wind power analysis[J]. Renewable Energy, 2016, 103: 15-29.
[7] 許偉,程剛,黃林,等. 基于混沌模擬退火PSO算法的威布爾分布參數(shù)估計應用研究[J]. 振動與沖擊,2017,36(12):134-139. Xu Wei, Cheng Gang , Huang Lin , et al . Research on Weibull distribution parameter estimation and application research based on chaotic simulated annealing PSO algorithm[J]. Vibration and Shock , 2017 , 36(12): 134-139. (in Chinese with English abstract)
[8] 鄭銳. 三參數(shù)威布爾分布參數(shù)估計及在可靠性分析中的應用[J]. 振動與沖擊, 2015(5): 78-81. Zheng Rui. Parameter estimation of three-parameter Weibull distribution and its application in reliability analysis[J]. Vibration and shock, 2015(5): 78-81. (in Chinese with English abstract)
[9] 叢楠,陳俊達,任焱晞,等. 一種指定功率譜密度與峭度值的對稱威布爾分布道路譜重構方法[J]. 振動與沖擊,2018(2):1-5. Cong Nan, Chen Junda, Ren Yanxi, et al. A road spectral reconstruction method for symmetric Weibull distribution with specified power spectral density and kurtosis [J]. Vibration and Shock, 2018 (2): 1-5. (in Chinese with English abstract)
[10] Krifa M. A mixed Weibull model for size reduction of particulate and fibrous materials [J]. Powder Technology, 2009, 194(3): 233-238.
[11] 凌丹,黃洪鐘,張小玲,等. 混合威布爾分布參數(shù)估計的L-M算法[J]. 電子科技大學學報,2008,37(4):634-636. Ling Dan, Huang Hongzhong, Zhang Xiaoling, et al. L-M algorithm for parameter estimation of mixed Weibull distribution[J]. Journal of the University of Electronic Science and Technology, 2008, 37(4): 634-636. (in Chinese with English abstract)
[12] Nagode M, Fajdiga M. An improved algorithm for parameter estimation suitable for mixed Weibull distributions[J]. International Journal of Fatigue, 2000, 22(1): 75-80.
[13] 王繼利,楊兆軍,李國發(fā),等. 基于改進EM算法的多重威布爾可靠性建模[J]. 吉林大學學報:工學版,2014,44(4):1010-1015. Wang Jili, Yang Zhaojun, Li Guofa, et al. Multiple Weibull reliability modeling based on improved EM algorithm [J]. Journal of Jilin University:Engineering Edition, 2014, 44(4): 1010-1015. (in Chinese with English abstract)
[14] 王曉峰,張英芝,申桂香,等. 基于ITLS和DE的加工中心三參數(shù)威布爾分布[J]. 華南理工大學學報:自然科學版,2015,43(6):84-88. Wang Xiaofeng, Zhang Yingzhi, Shen Guixiang, et al. Three parameter Weibull distribution of machining center based on ITLS and DE [J]. Journal of South China University of Technology :Natural Science Edition, 2015, 43 (6): 84-88. (in Chinese with English abstract)
[15] 張根保,李冬英,劉杰,等. 面向不完全維修的數(shù)控機床可靠性評估[J]. 機械工程學報,2013,49(23):136-141. Li Genbao, Li Dongying, Liu Jie, et al. Reliability Evaluation of NC Machine tool for incomplete maintenance [J]. Journal of Mechanical Engineering, 2013, 49(23): 136-141. (in Chinese with English abstract)
[16] 楊兆軍,楊川貴,陳菲,等. 基于PSO算法和SVR模型的加工中心可靠性模型參數(shù)估計[J]. 吉林大學學報:工學版,2015,45(3):829-836. Yang Zhaojun, Yang Chuangui, Chen Fei, et al. Parameter estimation of Machining Center Reliability Model based on PSO algorithm and SVR Model [J]. Journal of Jilin University: Engineering Edition, 2015, 45 (3): 829-836. (in Chinese with English abstract)
[17] Datsiou K C, Overend M. Weibull parameter estimation and goodness-of-fit for glass strength data[J]. Structural Safety, 2018, 73: 29-41.
[18] 魏艷華,王丙參,邢永忠. 指數(shù)-威布爾分布參數(shù)貝葉斯估計的混合Gibbs算法[J]. 統(tǒng)計與決策,2017(16):70-73. Wei Yanhua, Wang Bingcan, Xing Yongzhong. Mixed Gibbs algorithm for Bayesian estimation of exponent- Weibull Distribution parameters [J]. Statistics and Decision Making, 2017(16): 70-73. (in Chinese with English abstract)
[19] Ignacio M, Chubynsky M V, Slater G W. Interpreting the Weibull fitting parameters for diffusion-controlled release data[J]. Physica A Statistical Mechanics & Its Applications, 2017, 486: 1-24.
[20] 吳龍濤,王鐵寧,楊帆. 基于貝葉斯法和蒙特卡洛仿真的威布爾型裝備器材需求預測[J]. 兵工學報,2017,38(12):2447-2454. Wu Longtao, Wang Tiening, Yang Fan. Prediction of Weibull equipment demand based on Bayesian method and Monte Carlo simulation [J]. Journal of Military Engineering, 2017, 38(12): 2447-2454. (in Chinese with English abstract)
[21] Lan C, Bai N, Yang H, et al. Weibull modeling of the fatigue life for steel rebar considering corrosion effects[J]. International Journal of Fatigue, 2018, 111:134-143.
[22] Voronov S, Frisk E, Krysander M. Data-driven battery lifetime prediction and confidence estimation for heavy-duty trucks[J]. IEEE Transactions on Reliability, 2018, 67(2): 1-17.
[23] 劉大維,蔣榮超,朱龍龍,等. 基于遺傳算法的路面有理函數(shù)功率譜密度參數(shù)識別[J]. 農(nóng)業(yè)工程學報,2012, 28(8):128—133. Liu Dawei, Jiang Rongchao, Zhu Longlong, et al. Identification of pavement rational function power spectrum density parameters based on genetic algorithms[J]. Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE), 2012, 28 (8): 128—133. (in Chinese with English abstract)
[24] ?rkcü H H, ?zsoy V S, Aksoy E, et al. Estimating the parameters of 3-p Weibull distribution using particle swarm optimization: A comprehensive experimental comparison[J]. Applied Mathematics & Computation, 2015, 268(9): 201-226.
[25] Abbasi B, Niaki S T A, Khalife M A, et al. A hybrid variable neighborhood search and simulated annealing algorithm to estimate the three parameters of the Weibull distribution[J]. Expert Systems with Applications, 2011, 38(1): 700-708.
[26] Abbasi B, Hosseinifard S Z, Coit D W. A neural network applied to estimate Burr XII distribution parameters[J]. Reliability Engineering & System Safety, 2010, 95(6): 647-654.
[27] Xu M, Droguett E L, Lins I D, et al. On the q-Weibull distribution for reliability applications: An adaptive hybrid artificial bee colony algorithm for parameter estimation[J]. Reliability Engineering & System Safety, 2017, 158: 93-105.
[28] Cai Y. Artificial fish school algorithm applied in a combinatorial optimization problem[J]. International Journal of Intelligent Systems & Applications, 2010, 2(1): 37-43.
[29] 李曉磊. 一種新型的智能優(yōu)化方法-人工魚群算法[D]. 杭州:浙江大學,2003. Li Xiaolei. A New Intelligent Optimization method-artificial Fish Swarm algorithm [D]. Hangzhou: Zhejiang University, 2003. (in Chinese with English abstract)
[30] 劉志君,高亞奎,章衛(wèi)國,等. 混合人工魚群算法在約束非線性優(yōu)化中的應用[J]. 哈爾濱工業(yè)大學學報,2014,46(9):55-60. Liu Zhijun, Gao Yakui, Zhang Weiguo, et al. Application of hybrid artificial fish swarm algorithm in constrained nonlinear optimization [J]. Journal of Harbin University of Technology, 2014, 46 (9): 55-60. (in Chinese with English abstract)
[31] 劉思峰,楊英杰,吳利豐. 灰色系統(tǒng)理論及其應用. [M]. 第7版北京:科學出版社,2014.
[32] Kayacan E, Ulutas B, Kaynak O. Grey system theory-based models in time series prediction[J]. Expert Systems with Applications, 2010, 37(2): 1784-1789.
Statistical distribution modeling and two-step parameter identification of vehicle bridge displacement spectrum
Liu Qiaobin, Shi Wenku, Chen Zhiyong※, Shang Guoxu
(130022,)
The study of the statistical distribution is the basis for further loading spectrum and fatigue reliability platform test. Normal distribution and weibull distribution are 2 kinds of probability statistical distribution models widely used in reliability engineering. The idea of weighted superposition is used to approximate the actual distribution by so-called mixed model, and it has a strong practical application value, so it has been paid increasing attention by many scholars. The introduction of mixed distribution model brings many challenges in model parameter identification. Finding a simple, efficient and accurate mixed distribution model parameter estimation method has become a focus in the field of reliability research. The traditional reliability model parameter identification methods include graphic method, nonlinear least square method, maximum likelihood estimation and bias estimation, and so on. The main disadvantages of these algorithms are as follows: (1) The calculation efficiency needs to be improved, and the traditional algorithms mostly rely on iterative solution. Requirement to improve the accuracy of parameter estimation distinct increases the time cost. (2) The selection of parameter identification and optimization targets is improper. Most of the existing studies have defined the objective function of parameter identification as the square sum of the model and the measured data, which inevitably ignores the simulation error of the transverse coordinates between the sample points and the simulation points, that only considers the simulation error of the ordinate. (3) The empirical dependence of parameter identification is high, and the initial value of parameter identification has a great influence on the results. However, the intelligent algorithm shows great potential in the problem of parameter identification of the model with multidimensional nonlinearity and uneasy initial value. In view of this, the measured vehicle bridge displacement signal was taken as the research object in this paper, the time domain analysis and frequency domain power spectrum analysis were carried out respectively. In order to further study the statistical law of the displacement signals, the signal was grouped and the frequency was counted, the statistical histogram and the cumulative probability distribution curve were obtained. The normal distribution, mixed normal distribution, weibull distribution and mixed weibull distribution were employed respectively. A novel two-step parameter identification method was proposed, and the grey correlation degree objective function was introduced. The grey correlation coefficient objective function could ensure the maximum geometric similarity between the fitting curve and the original curve. By doing this, the inherent malpractice of the optimization process with the square sum of error as the fitness was overcome to some extent. The proposed parameter estimation method's tep was as following: Firstly, the parameters obtained by the artificial fish swarm algorithm were applied as the initial values of the model parameters. Secondly, the iterative nonlinear least square method, namely, levenberg-marquardt (LM) algorithm was used to identify the parameters accurately. Thirdly, the goodness of fit for each model were calculated by using the kolmogorov-smirnov test index and correlation coefficient. The result showed that the mixed weibull model could be used to describe the tested displacement signal best. The correlation coefficient between the mixed Weibull distribution and the statistical histogram was (0.9800, 0.9908,0.9867,0.9665), whereas, the mixed normal distribution was (0.9793,0.9904,0.9783,0.9661), the weibull model was (0.8613,0.9113,0.8618,0.8854), and the normal model was (0.8611,0.9127,0.8624,0.8869). The proposed two-step parameter identification method combined the advantages of the artificial fish swarm optimization algorithm and the traditional iterative algorithm, and used the artificial fish swarm optimization result as the initial value of the LM algorithm. It solved the problem of the difficulty in selecting the initial value of the nonlinear least square method and improved the efficiency of the parameter identification. This study can provide reference for the fatigue load spectrum and the bench test of off-road vehicles.
model; parameter identification; rehicle bridge displacement spectrum; grey relation; nonlinear least square method; mixed Weibull model
劉巧斌,史文庫,陳志勇,商國旭. 工程車輛車橋位移譜統(tǒng)計分布建模及分步參數(shù)識別[J]. 農(nóng)業(yè)工程學報,2018,34(23):67-75. doi:10.11975/j.issn.1002-6819.2018.23.008 http://www.tcsae.org
Liu Qiaobin, Shi Wenku, Chen Zhiyong, Shang Guoxu.Statistical distribution modeling and two-step parameter identification of vehicle bridge displacement spectrum[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2018, 34(23): 67-75. (in Chinese with English abstract) doi:10.11975/j.issn.1002-6819.2018.23.008 http://www.tcsae.org
2018-06-05
2018-10-27
吉林省科技發(fā)展計劃項目基金(20150307034GX);吉林省重大科技攻關項目基金(20170204063GX)
劉巧斌,博士生,主要研究方向為汽車系統(tǒng)動力學。Email:liuqb17@mails.jlu.edu.cn
陳志勇,副教授,主要研究方向為汽車振動噪聲控制。Email:chen_zy@jlu.edu.cn
10.11975/j.issn.1002-6819.2018.23.008
U463.2
A
1002-6819(2018)-23-0067-09