龔夕霞 李焱鑫 盧琴芬
模塊化永磁直線同步電機考慮制造公差的推力魯棒性優(yōu)化
龔夕霞 李焱鑫 盧琴芬
(浙江大學(xué)電氣工程學(xué)院 杭州 310027)
模塊化永磁直線同步電機(MPMLSM)具有效率與推力密度高、可靠性和可加工性好等優(yōu)點,非常適用于長行程運輸系統(tǒng);缺點是批量生產(chǎn)中性能易受到加工公差的影響。針對這一問題,該文提出了一種考慮制造公差的綜合多目標(biāo)魯棒優(yōu)化設(shè)計方法。首先,基于六西格瑪設(shè)計方法建立了魯棒優(yōu)化模型;其次,采用拉丁超立方采樣方法在尺寸公差范圍內(nèi)根據(jù)正態(tài)分布規(guī)律進(jìn)行抽樣,模擬大規(guī)模生產(chǎn)時電機尺寸受公差影響可能會出現(xiàn)的各種變化;再次,在尺寸優(yōu)化范圍內(nèi)均勻抽樣構(gòu)成足夠的訓(xùn)練樣本,由有限元軟件仿真這些抽樣方案的推力性能,并基于反向傳播神經(jīng)網(wǎng)絡(luò)建立電機設(shè)計代理模型;然后,應(yīng)用該代理模型對拉丁超立方采樣得到的樣本進(jìn)行電機性能的計算,求解樣本整體對應(yīng)的各優(yōu)化目標(biāo)的均值與方差,進(jìn)而計算多目標(biāo)優(yōu)化的適應(yīng)度;最后,采用非支配排序遺傳算法Ⅱ進(jìn)行全局優(yōu)化,得到魯棒優(yōu)化方案,與不考慮公差的確定性優(yōu)化方案相比,驗證了該方法的有效性。魯棒優(yōu)化方案雖然略微增加了電機體積和推力波動,但是其失效概率低,受公差影響小,更加符合產(chǎn)品批量生產(chǎn)過程中的質(zhì)量要求。
模塊化永磁直線同步電機(MPMLSM) 制造公差 魯棒優(yōu)化設(shè)計 六西格瑪設(shè)計 代理模型
永磁直線同步電機(Permanent Magnet Linear Synchronous Motor, PMLSM)作為一種具有高效率、高推力密度、高可靠性的電機,近年來受到了廣泛的關(guān)注與應(yīng)用[1-2]。同時,由于永磁材料含有昂貴的稀土元素,學(xué)者們通過采用初級勵磁型電機結(jié)構(gòu)[3]、模塊化結(jié)構(gòu)以及使用較低最大磁能積永磁材料等手段,實現(xiàn)節(jié)約資源、降低成本的目的。模塊化永磁直線同步電機(Modular Permanent Magnet Linear Synchronous Motor, MPMLSM)在初級插入隔磁間隙(磁隙),不僅有利于簡化繞線制造過程,還能夠提高電機的性能與容錯率[4-5]。
在實際工程應(yīng)用中,電機驅(qū)動系統(tǒng)的最終質(zhì)量很大程度上取決于制造技術(shù)、材料多樣性、加工公差、裝配誤差等因素。因此,為保證低產(chǎn)品報廢率,應(yīng)當(dāng)將這些不確定因素考慮在初期設(shè)計中,以滿足相應(yīng)性能指標(biāo),實現(xiàn)更高的系統(tǒng)可靠性[6-10]。文獻(xiàn)[7]考慮了軟磁材料-特性在制造過程中的劣化對電機性能的影響,并將該制造影響作為優(yōu)化設(shè)計的一部分,保證了優(yōu)化設(shè)計方案對制造影響的低靈敏度,但是軟磁材料的劣化程度隨不同材料、不同加工過程而異,為確定對應(yīng)的數(shù)值變化需進(jìn)行大量的實驗?zāi)M,耗時耗力。文獻(xiàn)[8]考慮了電機尺寸誤差對產(chǎn)品生產(chǎn)缺陷率的影響,基于六西格瑪設(shè)計(Design for Six Sigma, DFSS)與蒙特卡羅模擬對內(nèi)嵌式永磁電機進(jìn)行了魯棒性優(yōu)化,并與確定性優(yōu)化方案對比,驗證了魯棒設(shè)計方案的有效性。文獻(xiàn)[9]通過極限學(xué)習(xí)機求解了PMLSM氣隙對稱度誤差、平行度誤差、直線度誤差與推力性能的非線性關(guān)系,根據(jù)設(shè)計要求,采用遺傳算法求解氣隙的幾何公差范圍,并將公差分配至相關(guān)零件上。文獻(xiàn)[10]同時考慮了電機尺寸參數(shù)與控制參數(shù)對電機性能的影響,保證了系統(tǒng)靜態(tài)與動態(tài)性能的魯棒性,并通過多級框架解決了高維優(yōu)化問題。無論電機推力、尺寸大抑或小,公差對電機性能的影響都不可忽略,文獻(xiàn)[11]對推力可達(dá)數(shù)百牛頓的高溫超導(dǎo)直線同步電機進(jìn)行了魯棒優(yōu)化設(shè)計,同時對傳統(tǒng)田口參數(shù)設(shè)計法進(jìn)行了改進(jìn),在增加平均推力、降低推力波動的同時提升了制造質(zhì)量,提高了求解效率。
由上可見,采用考慮公差的電機優(yōu)化算法成為了一個研究熱點。MPMLSM由于模塊化結(jié)構(gòu)更容易受到加工公差的影響,為了提高電機性能,需要采用魯棒性優(yōu)化方法進(jìn)行優(yōu)化設(shè)計。然而,現(xiàn)在鮮有文獻(xiàn)對此進(jìn)行研究。因此,本文針對18槽20極(18S20P)MPMLSM提出了考慮公差的推力魯棒性優(yōu)化方法。在介紹拓?fù)浣Y(jié)構(gòu)與結(jié)構(gòu)參數(shù)后,建立了考慮公差的魯棒性優(yōu)化模型,包括優(yōu)化目標(biāo)函數(shù)、優(yōu)化變量、約束條件與優(yōu)化方法等。通過全局優(yōu)化獲得了魯棒性優(yōu)化方案,并與不考慮公差的確定性優(yōu)化方案對比,驗證了魯棒優(yōu)化方法的有效性。研究能為MPMLSM優(yōu)化設(shè)計與推廣應(yīng)用提供幫助。
圖1為典型的MPMLSM拓?fù)浣Y(jié)構(gòu),其極槽配合為18槽20極(18S20P),采用端部非重疊式隔齒繞組[12-13],相關(guān)性能指標(biāo)與結(jié)構(gòu)參數(shù)見表1。這種隔齒繞組電機不僅具有線圈端部短、效率高、推力密度大等優(yōu)點,而且平均推力高、容錯能力強[14-15]。模塊化結(jié)構(gòu)下,電機采用單一鐵心就可以實現(xiàn)多種極槽配合,增強了制造靈活性,降低了加工難度。可以看到,在該模塊化結(jié)構(gòu)中模塊與模塊之間存在隔磁間隙,此時相應(yīng)減小槽的寬度,從而保證齒寬及極距不變。研究表明,該間隙不僅方便了安裝,而且在某些結(jié)構(gòu)下還能提高電機的性能[15-16]。
圖1 18S20P MPMLSM拓?fù)浣Y(jié)構(gòu)示意圖
MPMLSM的特點是將多個獨立模塊組裝成一個整體,拼裝的模式使得電機性能容易受到公差的影響,包括加工誤差與安裝誤差,若設(shè)計時沒有考慮公差,優(yōu)化方案的性能可能不是最優(yōu)的,其原因可通過圖2確定性優(yōu)化設(shè)計與魯棒性優(yōu)化設(shè)計的對比來解釋。圖中,A點為確定性優(yōu)化中系統(tǒng)的最優(yōu)解,即范圍內(nèi)輸出最小值對應(yīng)的點,但是當(dāng)公差(噪聲因子)D出現(xiàn)時,可能會引起輸出量較大的波動;B點為魯棒性優(yōu)化解,相比之下,雖然B點對應(yīng)的輸出值不是全局最小,但是B點受到公差的影響較小,具有較高的魯棒性,有利于生產(chǎn)加工。也就是說,在傳統(tǒng)不考慮公差的優(yōu)化(確定性優(yōu)化)過程中最優(yōu)解是A,而若考慮加工誤差,則最優(yōu)解將是B。由此可見,兩種優(yōu)化方法出現(xiàn)了不同,顯然后者更適用于MPMLSM,因此建立魯棒性優(yōu)化設(shè)計模型至關(guān)重要,圖3為詳細(xì)的設(shè)計流程,具體步驟如下所述。
表1 MPMLSM性能指標(biāo)與結(jié)構(gòu)參數(shù)
Tab.1 Performance index and structure parameters of MPMLSM
圖2 確定性優(yōu)化設(shè)計與魯棒性優(yōu)化設(shè)計的區(qū)別
圖3 MPMLSM魯棒性優(yōu)化設(shè)計流程
第一步,設(shè)定西格瑪水平并確定優(yōu)化目標(biāo)函數(shù)。
第二步,為快速、便捷地計算后續(xù)優(yōu)化所需的電機性能,本文先基于有限元軟件仿真計算相應(yīng)的訓(xùn)練樣本(該訓(xùn)練樣本無需呈正態(tài)分布,在尺寸優(yōu)化范圍內(nèi)通過均勻抽樣獲?。?,然后通過反向傳播神經(jīng)網(wǎng)絡(luò)建立代理模型。
第三步,采用非支配排序遺傳算法Ⅱ[17]處理帶約束的多目標(biāo)優(yōu)化問題,優(yōu)化過程主要包括抽樣與適應(yīng)度計算:①抽樣,為計算目標(biāo)函數(shù)所包含的樣本均值與方差,首先需要使樣本點在公差范圍內(nèi)呈正態(tài)分布,本文采用拉丁超立方抽樣方法[18]實現(xiàn)上述目標(biāo),以模擬大規(guī)模生產(chǎn)時受公差影響可能會出現(xiàn)的電機尺寸組合;②適應(yīng)度計算,采用第二步訓(xùn)練完成的代理模型計算各個樣本對應(yīng)的電機性能參數(shù),進(jìn)而計算樣本整體的均值與方差,用于代入目標(biāo)函數(shù)之中,得到適應(yīng)度計算值。
根據(jù)適應(yīng)度計算值得到帕累托前沿,選取最優(yōu)解,并與確定性優(yōu)化方案的失效概率進(jìn)行比較,驗證魯棒性優(yōu)化設(shè)計方案的有效性。
傳統(tǒng)優(yōu)化模型包括了優(yōu)化目標(biāo)以及約束條件,有
式中,為優(yōu)化目標(biāo)函數(shù);為輸入變量;g為部分電機性能的約束條件;為約束條件的個數(shù)。
傳統(tǒng)優(yōu)化模型通常不考慮公差,MPMLSM為了顯著降低批量生產(chǎn)中的缺陷率,需要將不可避免的公差(噪聲因素)考慮在內(nèi),稱為考慮制造公差的魯棒性優(yōu)化設(shè)計,此時優(yōu)化目標(biāo)與約束條件需要采用六西格瑪設(shè)計(DFSS)方法確定[19]。
六西格瑪起初主要針對制造業(yè),通過數(shù)據(jù)收集、研究分布規(guī)律,利用正態(tài)分布分析每百萬次生產(chǎn)中的產(chǎn)品缺陷率(Defects Per Million Opportunities, DPMO),后來在服務(wù)業(yè)也有相關(guān)應(yīng)用。6從統(tǒng)計意義上就是百萬分之三點四,即西格瑪水平由DPMO定義,當(dāng)不合格品率達(dá)到3.4DPMO時,產(chǎn)品已經(jīng)達(dá)到了6水平。表2為不同西格瑪水平與DPMO的轉(zhuǎn)換關(guān)系[20]。由于長期數(shù)據(jù)分布與短期數(shù)據(jù)分布之間存在1.5的水平偏移,4只能符合短期要求,6才能滿足實際工程長期制造的需求。
表2 西格瑪水平、合格率與DPMO的轉(zhuǎn)換
Tab.2 The conversion of Sigma level, qualified rate and DPMO
采用六西格瑪設(shè)計方法考慮公差(擾動)后,就可以得到MPMLSM魯棒性優(yōu)化的優(yōu)化目標(biāo)與約束條件,式(1)轉(zhuǎn)變?yōu)?/p>
對于目標(biāo)函數(shù)最大最小化的情況,目標(biāo)函數(shù)表示為
MPMLSM在保持平均推力基本不變的條件下,需要優(yōu)化的主要是推力波動與體積,因此以這兩者為優(yōu)化目標(biāo),將式(3)轉(zhuǎn)變?yōu)槎嗄繕?biāo)優(yōu)化目標(biāo)函數(shù),即
MPMLSM結(jié)構(gòu)參數(shù)中與性能緊密相關(guān)的是齒寬、齒高、磁隙、背鐵高度、磁鐵高度和初級極距,因此將這6個變量作為優(yōu)化變量,表3為這些優(yōu)化變量的取值范圍。
對于MPMLSM,其優(yōu)化需要滿足以下條件:
(1)在確定性優(yōu)化結(jié)果中次級高度被優(yōu)化到了約束的臨界位置,且六西格瑪設(shè)計使得約束條件增倍,為使得魯棒性優(yōu)化結(jié)果收斂,設(shè)定磁鐵高度以及背鐵高度的公差均為負(fù)公差。
表3 各優(yōu)化變量取值范圍
Tab.3 The range of optimization variables (單位: mm)
(3)在代理模型中,軛部高度設(shè)置為齒寬的一半,為滿足代理模型的計算要求、便于優(yōu)化計算,設(shè)定前者公差同樣為后者的一半。
(4)考慮組合件的公差累計問題時,采用方均根法(Root Sum Square, RSS)作為公差疊加分析模型,將公差二次方和的根作為累計公差。
(5)優(yōu)化時電機槽滿率增加至50%,以改善電機性能。
根據(jù)以上假設(shè)條件,就可以得到MPMLSM魯棒性優(yōu)化設(shè)計的約束條件,有
MPMLSM優(yōu)化目標(biāo)函數(shù)中的推力波動、約束條件中的推力平均值都由電機設(shè)計模型計算得到。傳統(tǒng)優(yōu)化方法中電機設(shè)計模型主要為解析算法、等效磁路法及磁網(wǎng)絡(luò)法等,計算時間短,但是傳統(tǒng)方法會對電機結(jié)構(gòu)進(jìn)行簡化,其計算結(jié)果無法滿足魯棒優(yōu)化的模型精度需求。為了提高計算的快速性與適用性,本文采用了基于代理模型的電機設(shè)計模型,雖然增加了抽取樣本以及訓(xùn)練模型的時間,但訓(xùn)練完成后代理模型的計算速度堪比傳統(tǒng)電機設(shè)計模型,且精度接近于有限元[21-23]。
在電機尺寸允許范圍內(nèi)均勻采樣獲得樣本庫,通過有限元軟件仿真計算相應(yīng)的電機性能參數(shù)。接著,通過機器學(xué)習(xí)的方法對樣本庫進(jìn)行訓(xùn)練與學(xué)習(xí),得到基于代理模型的電機模型。本文選擇反向傳播神經(jīng)網(wǎng)絡(luò)(Back Propagation Neural Network, BPNN)[24]對樣本進(jìn)行訓(xùn)練,其具有高效、高精度的優(yōu)點。
MPMLSM代理模型基于8 100個有限元樣本訓(xùn)練得到,采用了一層隱含層,決定系數(shù)2=0.999 9,方均誤差MSE=1.76×10-3,可見代理模型精度很高。圖4為代理模型與有限元的推力平均值和波動值對比結(jié)果,可以看出所采用的神經(jīng)網(wǎng)絡(luò)對推力及推力波動的高維曲線具有足夠的學(xué)習(xí)和表征能力。
(a)推力平均值
(b)推力波動值
圖4 代理模型與有限元計算結(jié)果的對比
Fig.4 Comparison between surrogate model results and finite element calculation results
(a)磁鋼高度/齒寬制造公差0.1 mm,其余0.05 mm
(b)制造公差均為0.1 mm
表4 不同條件下的魯棒優(yōu)化可行解
Tab.4 The feasible solution of robust optimization under different conditions
在建立的優(yōu)化模型中,如果不考慮公差,優(yōu)化目標(biāo)函數(shù)直接是推力波動與體積,無需進(jìn)行抽樣模擬,無需考慮樣本的均值與方差,全局優(yōu)化后得到的結(jié)果即為確定性優(yōu)化方案。MPMLSM確定性優(yōu)化方案與原樣機的對比見表5。
表5 MPMLSM確定性優(yōu)化方案與原樣機對比
Tab.5 Comparison between deterministic optimization design and original design of MPMLSM
圖6為不同公差條件下確定性優(yōu)化結(jié)果與魯棒性優(yōu)化結(jié)果的推力平均值分布情況。顯然,推力曲線均呈正態(tài)分布規(guī)律,但兩個魯棒性優(yōu)化方案的曲線均值更接近于設(shè)定值。
圖6 不同公差條件下確定性優(yōu)化結(jié)果與魯棒性優(yōu)化結(jié)果的推力平均值分布情況
表6 確定性與魯棒性優(yōu)化結(jié)果的失效概率對比
Tab.6 POF comparison between deterministic and robust optimization results (%)
圖7 實驗樣機
圖8為MPMLSM在0.43 m/s速度下反電動勢波形的實測與有限元結(jié)果對比。如圖8所示,測得的反電動勢基波含量與有限元仿真結(jié)果基本保持一致,而諧波分析結(jié)果略有差異,這可能是磁鋼質(zhì)量差異性與裝配誤差導(dǎo)致的。由于基波幅值遠(yuǎn)遠(yuǎn)高于諧波幅值,因此諧波的差異對反電動勢波形的影響不大。
基于圖9所示的推力測試平臺,測量d=0控制方案下電機推力平均值與q軸電流的關(guān)系。首先,在A軸通入直流電,使得A軸與次級d軸吸合;然后,將電機向前移動1/2極距的距離,以對準(zhǔn)q軸位置;最后,在此基礎(chǔ)上,根據(jù)所測q軸電流,給三相各通入相應(yīng)大小的直流電,模擬交流電的瞬時狀態(tài),并記錄推力計測得的推力值。圖10為推力平均值與q軸電流的對應(yīng)關(guān)系,有限元仿真結(jié)果與實驗測量結(jié)果基本保持一致,且兩者誤差低于5%,除樣機本體外直流電源輸出的不穩(wěn)定性、測力計測量誤差等因素也可能使結(jié)果產(chǎn)生偏差。
(a)單位周期內(nèi)的波形
(b)傅里葉分解結(jié)果
圖8 空載反電動勢的實測與有限元結(jié)果
Fig.8 Comparison between measured and finite element results of no-load back electromotive force
圖9 推力測試平臺
圖10 推力平均值與q軸電流的關(guān)系
考慮到MPMLSM不可避免的公差因素,本文提出了一種考慮公差的魯棒性優(yōu)化設(shè)計模型,其優(yōu)化目標(biāo)函數(shù)與約束條件中都加入了樣本整體對應(yīng)的各優(yōu)化目標(biāo)的均值與方差。為模擬大規(guī)模生產(chǎn)時MPMLSM尺寸受公差影響會發(fā)生的變化,采用LHS建立樣本庫,保證樣本在公差范圍內(nèi)呈正態(tài)分布;通過BPNN建立電機設(shè)計代理模型,經(jīng)過有限元驗證,該代理模型的決定系數(shù)為0.999 9,方均誤差為1.76×10-3,通過該模型計算出電機性能,進(jìn)而計算樣本整體的均值與方差,得到多目標(biāo)優(yōu)化的適應(yīng)度。最后,采用非支配排序遺傳算法Ⅱ全局優(yōu)化算法得到了魯棒優(yōu)化設(shè)計方案,解決了帶約束的多目標(biāo)優(yōu)化問題。
考慮到安裝誤差同樣會對磁隙、氣隙等參數(shù)產(chǎn)生影響,今后可以在本文優(yōu)化結(jié)果的基礎(chǔ)上,進(jìn)一步針對安裝誤差對電機進(jìn)行優(yōu)化,以提升電機性能的魯棒性,提高產(chǎn)品制造質(zhì)量。
[1] 蔣錢, 盧琴芬, 李焱鑫. 雙三相永磁直線同步電機的推力波動及抑制[J]. 電工技術(shù)學(xué)報, 2021, 36(5): 883-892.
Jiang Qian, Lu Qinfen, Li Yanxin. Thrust ripple and depression method of dual three-phase permanent magnet linear synchronous motors[J]. Transactions of China Electrotechnical Society, 2021, 36(5): 883- 892.
[2] Cui Fengrui, Sun Zhaolong, Xu Wei, et al. Com- parative analysis of bilateral permanent magnet linear synchronous motors with different structures[J]. CES Transactions on Electrical Machines and Systems, 2020, 4(2): 142-150.
[3] 沈燚明, 盧琴芬. 初級勵磁型永磁直線電機研究現(xiàn)狀與展望[J]. 電工技術(shù)學(xué)報, 2021, 36(11): 2325- 2343.
Shen Yiming, Lu Qinfen. Overview of permanent magnet linear machines with primary excitation[J]. Transactions of China Electrotechnical Society, 2021, 36(11): 2325-2343.
[4] 丁文, 李可, 付海剛. 一種12/10極模塊化定子混合勵磁開關(guān)磁阻電機分析[J]. 電工技術(shù)學(xué)報, 2022, 37(8): 1948-1958.
Ding Wen, Li Ke, Fu Haigang. Analysis of a 12/10- pole modular-stator hybrid-excited switched relu- ctance machine[J]. Transactions of China Electro- technical Society, 2022, 37(8): 1948-1958.
[5] 王宇, 張成糕, 郝雯娟. 永磁電機及其驅(qū)動系統(tǒng)容錯技術(shù)綜述[J]. 中國電機工程學(xué)報, 2022, 42(1): 351-372.
Wang Yu, Zhang Chenggao, Hao Wenjuan. Overview of fault-tolerant technologies of permanent magnet brushless machine and its control system[J]. Pro- ceedings of the CSEE, 2022, 42(1): 351-372.
[6] Lei Gang, Bramerdorfer G, Liu Chengcheng, et al. Robust design optimization of electrical machines: a comparative study and space reduction strategy[J]. IEEE Transactions on Energy Conversion, 2021, 36(1): 300-313.
[7] Bramerdorfer G. Effect of the manufacturing impact on the optimal electric machine design and perfor- mance[J]. IEEE Transactions on Energy Conversion, 2020, 35(4): 1935-1943.
[8] Xu Gaohong, Jia Zexin, Zhao Wenxiang, et al. Multi- objective optimization design of inset-surface per- manent magnet machine considering deterministic and robust performances[J]. Chinese Journal of Elec- trical Engineering, 2021, 7(3): 73-87.
[9] Yao Zheng, Zhao Jiwen, Song Juncai, et al. Research on selection criterion of design tolerance for air-core permanent magnet synchronous linear motor[J]. IEEE Transactions on Industrial Electronics, 2021, 68(4): 3336-3347.
[10] Lei Gang, Wang Tianshi, Zhu Jianguo, et al. Robust multiobjective and multidisciplinary design optimi- zation of electrical drive systems[J]. CES Transa- ctions on Electrical Machines and Systems, 2018, 2(4): 409-416.
[11] Lei Gang, Liu Chengcheng, Li Yanbin, et al. Robust design optimization of a high-temperature supercon- ducting linear synchronous motor based on taguchi method[J]. IEEE Transactions on Applied Supercon- ductivity, 2019, 29(2): 1-6.
[12] Sun Ke, Tian Shaopeng. Multiobjective optimization of IPMSM with FSCW applying rotor Notch design for torque performance improvement[J]. IEEE Transa- ctions on Magnetics, 2022, 58(5): 1-9.
[13] Skarmoutsos G A, Gyftakis K N, Mueller M. Analyti- cal prediction of the MCSA signatures under dynamic eccentricity in PM machines with concentrated non- overlapping windings[J]. IEEE Transactions on Energy Conversion, 2022, 37(2): 1011-1019.
[14] 丁強, 王曉琳, 鄧智泉, 等. 大氣隙磁通切換無軸承永磁電機徑向力繞組設(shè)計與比較[J]. 電工技術(shù)學(xué)報, 2018, 33(11): 2403-2413.
Ding Qiang, Wang Xiaolin, Deng Zhiquan, et al. Design and comparison of radial force winding configurations for wide air-gap flux-switching bearingless permanent-magnet motor[J]. Transactions of China Electrotechnical Society, 2018, 33(11): 2403-2413.
[15] 隋義, 尹佐生, 鄭萍, 等. 單雙層混合繞組型低互感五相容錯永磁電機的電磁問題研究[J]. 中國電機工程學(xué)報, 2022, 42(1): 329-340.
Sui Yi, Yin Zuosheng, Zheng Ping, et al. Research on electromagnetic problems of low-mutual-inductance five-phase fault-tolerant PMSM with hybrid single/ double-layer FSCW[J]. Proceedings of the CSEE, 2022, 42(1): 329-340.
[16] Mehta S, Kabir M A, Pramod P, et al. Segmented rotor mutually coupled switched reluctance machine for low torque ripple applications[J]. IEEE Transa- ctions on Industry Applications, 2021, 57(4): 3582- 3594.
[17] Deb K, Pratap A, Agarwal S, et al. A fast and elitist multiobjective genetic algorithm: NSGA-Ⅱ[J]. IEEE Transactions on Evolutionary Computation, 2002, 6(2): 182-197.
[18] Ma Yiming, Xiao Yang, Wang Jin, et al. Multicriteria optimal Latin hypercube design-based surrogate- assisted design optimization for a permanent-magnet vernier machine[J]. IEEE Transactions on Magnetics, 2022, 58(2): 1-5.
[19] Koch P N, Yang R, Gu L. Design for six sigma through robust optimization[J]. Structural and Multi- disciplinary Optimization, 2004, 26(3): 235-248.
[20] 唐曉芬主編, 上海質(zhì)量管理科學(xué)研究院編著. 六西格瑪核心教程: 黑帶讀本[M]. 修訂版. 北京: 中國標(biāo)準(zhǔn)出版社, 2006.
[21] 謝冰川, 張岳, 徐振耀, 等. 基于代理模型的電機多學(xué)科優(yōu)化關(guān)鍵技術(shù)綜述[J]. 電工技術(shù)學(xué)報, 2022, 37(20): 5117-5143.
Xie Bingchuan, Zhang Yue, Xu Zhenyao, et al. Review on multidisciplinary optimization key tech- nology of electrical machine based on surrogate models[J]. Transactions of China Electrotechnical Society, 2022, 37(20): 5117-5143.
[22] Wu Jiangling, Sun Xiaodong, Zhu Jianguo. Accurate torque modeling with PSO-based recursive robust LSSVR for a segmented-rotor switched reluctance motor[J]. CES Transactions on Electrical Machines and Systems, 2020, 4(2): 96-104.
[23] 李雄松, 崔鶴松, 胡純福, 等. 平板型永磁直線同步電機推力特性的優(yōu)化設(shè)計[J]. 電工技術(shù)學(xué)報, 2021, 36(5): 916-923.
Li Xiongsong, Cui Hesong, Hu Chunfu, et al. Optimal design of thrust characteristics of flat-type permanent magnet linear synchronous motor[J]. Transactions of China Electrotechnical Society, 2021, 36(5): 916-923.
[24] 何海婷, 柳亦兵, 巴黎明, 等. 基于BP神經(jīng)網(wǎng)絡(luò)的飛輪儲能系統(tǒng)主動磁軸承非線性動力學(xué)模型[J]. 中國電機工程學(xué)報, 2022, 42(3): 1184-1198.
He Haiting, Liu Yibing, Ba Liming, et al. Nonlinear dynamic model of active magnetic bearing in flywheel system based on BP neural network[J]. Proceedings of the CSEE, 2022, 42(3): 1184-1198.
Thrust Robustness Optimization of Modular Permanent Magnet Linear Synchronous Motor Accounting for Manufacture Tolerance
(College of Electrical Engineering Zhejiang University Hangzhou 310027 China)
The modular structure of modular permanent magnet linear synchronous motor (MPMLSM) is beneficial to simplify the winding manufacturing process and improve the motor performance and fault tolerance rate. As a motor with high efficiency, high thrust density, and high reliability, it has received extensive attention and application in recent years. However, in mass production, the modular structure is more susceptible to machining tolerances. In order to improve the motor performance and better meet the quality requirements, this paper proposes a thrust robustness optimization method considering manufacturing tolerance, which provides help for the optimization design and application of MPMLSM.
First, a robust optimization model is established based on design for six Sigma (DFSS), and the optimization objective function is determined according to the sigma level. Then, the Latin Hypercube Sampling (LHS) method is used to sample within the dimensional tolerance range according to the normal distribution law to simulate possible variations in motor dimensions in mass production. A motor design surrogate model based on the Back Propagation Neural Network (BPNN) can calculate the motor performance required for subsequent optimization quickly and conveniently. The training samples of the model are obtained by uniform sampling, and the thrust performance of the samples is simulated by finite element software. Subsequently, the motor performance of the samples obtained by LHS is calculated by the trained surrogate model, and the mean and variance of the whole sample are solved and substituted into the objective function to obtain the fitness value. Finally, Non-dominated Sorting Genetic Algorithm Ⅱ is used for global optimization to obtain the Pareto front and robust optimization schemes. Compared with the deterministic optimization scheme, the effectiveness of the method is verified.
Through finite element verification, the determination coefficient of the surrogate model established by BPNN is 0.999 9, and the mean square error is 1.76×10-3. Thus, the motor performance can be calculated quickly and accurately. Under the harsh condition that the allowable thrust variation range () is 2%, the robust optimization scheme can reduce the probability of failure (POF) from 67.32% to 7.96% under condition 1, and from 53.3% to 1.76% under condition 2. Compared with the deterministic optimization scheme without considering tolerance, although the robust optimization design slightly increases the motor volume and thrust ripple, the robustness is improved. At the same time, it can be inferred that the POF of the robust optimization scheme is 0 when≥7% under condition 1 or≥5% under condition 2. The reduction of failure probability indicates that the MPMLSM robust optimization schemes have higher qualification rates in mass production and are less affected by tolerances.
The following conclusions can be drawn: (1) Under the premise of convergence, reducing manufacturing tolerance can reduce the motor volume and make the motor thrust closer to the set value. However, the manufacturing cost is increased. (2)mainly affects the thrust fluctuation, and little affects the volume. Under the same volume condition, the thrust fluctuation increases with the increase of. (3) In the optimization results, the tooth height and primary polar distance vary with different tolerance conditions, while the remaining variables are optimized to optimum values. (4) The POF of the robust optimization scheme is lower than that of the deterministic optimization scheme, especially whenis small. Therefore, the robust optimization scheme has better robustness and is more in line with the quality requirement in mass production.
Modular permanent magnet linear synchronous motor (MPMLSM), manufacture tolerance, robust optimization design, design for six Sigma, surrogate model
TM351
10.19595/j.cnki.1000-6753.tces.221956
國家自然科學(xué)基金面上資助項目(52177061, 52107060)。
2022-10-14
2022-11-03
龔夕霞 女,1997年生,碩士,研究方向為模塊化永磁直線同步電機設(shè)計與優(yōu)化。E-mail: gongxixia@zju.edu.cn
李焱鑫 男,1988年生,副研究員,碩士生導(dǎo)師,研究方向為新型直線電機與永磁電機的建模分析與優(yōu)化分析。E-mail: eeliyanxin@zju.edu.cn(通信作者)
(編輯 崔文靜)