110柴油機(jī)飛輪殼機(jī)械加工工藝規(guī)程及工藝裝備設(shè)計(jì)

資源目錄里展示的全都有，所見即所得。下載后全都有,請放心下載。原稿可自行編輯修改=【QQ：401339828 或11970985 有疑問可加】 中北大學(xué)分校2007屆本科畢業(yè)設(shè)計(jì)說明書 An Identification Model of Health States of M achine Wear Based on OiL Analysis FU jun-qing ,li han-xiong,suan xin-hua 1. School of Automobile & Mechanical Enginering Changsha Univcrsity of Sciene & Technology Changsha 410076, P.R. China 2.School of Mechanical and Electrical Engineering Ccntral South University Changsha 410083,P.R.China Abstract This paper presents is a modeling procedure for deriving a sing1e value measure based on a resgression and a method for determining a statistical threhoId value as identification criterion of nomal or abnomal states of machine wear A rea1。numerica1 example is examfined by the method and identification criterion presented. The result indicate that the judgments by the presented methods are Basically consistent with the rea1 facts, and therefore the method and identi fication criterion are caluable for judging the no;mal state of machine wea r based on oil analysis. Key words: oil analysis rcgrcssion model ,single value ,measure, and threshold value Introduction Oil analysis has been used worldwide as a method for reducing maintenance cost, improving rcliability and productivity in various industries1.Currently , most oil analyzers use the methods of a tomicemfission spectrometry optical or electronic microscopy and ferrogmph\y etc to conduct the oil analysis. The aim of oil analysis is to evaluate the condition of the lubrication or the equipment from the 1ubricant oil samplce of a machine.,and recommend maintenance actions to the equipment opemting activity. Without disassembling the machine the oil samples of a machine can be acquired according to certa regulations, and through analysis of the oil sample the oil and machine condition can be evaluated Original equipment manufacturers(OEM), lubricant suppliers and oil analysis labomtories provide specific guidelines for vicar metal concentrations in the oil. These limits provide good gencral guide1ines for interpreting oil analysis data. But there are many elements in oil analysis date, it is very difficult to directly judge the wearstate according to the oil analysis data. Forengineering application ,a single value index or measure is needed for identifying the states of the monitored oil samples V. Macia n et al (8) derived。general expression of the rate of vicar from engine oil analysis date and defined the engine wear rate(Zer) as an index Z. The index value(Z) was used to evaluate the wear rate of an engine being normal or abnorm al by reference to a normal wear rate(EMwr).In fact index and reference index should be random variables of oil samples and not exact values , therefore a statistical m odel of the indexes is needed Chunhua Zhao etal developed a model by means of a stepwise pluralistic regression which deletes some insignificant elements or linearly relate elements in oil analysis original date, and transfers the oil analysis data into a single value. The single value was used as a judgment of the wear state, but there was no threshold or critical value used as an identification criterion.. Thus for the values of the samples far from the normal state value 1 or abnormal state value 2, it was not clear whether they belonged to the normal or abnormal state. This paper first improves the modeling p rocedure in Reference(9). for doriving a single value measure，based on a regression mode and then presents a method to determine a statistical threshold value as an identification criterion of a normal or abnormal state, A real numerical example is examined by the method and an identification criterion are presented . The results indicate that the judgments by the presented methods are basically consistent with the true facts, and therefore the method and identification critcrion is valuable for judging the normal or a5normal state of machine wear from oil samples. 2 Modeling procedure 2.1 Experiment design and sample Regularly or irregularly collect and analyze oil samples to obtain the concentration of various elements in oil samples. In meantime, thoroughly inspect and determine the health states of machine weat by means of other methods such as dissembling machine, measuring debris shapes and etc In this paper， the health state is simply of binary-value,ie.norma land abnormal Once sufficient data are collected, the experiment is stopped and single value model will be built 2.2 Modeling The obserbved health states are defined as follows On the other hand, the y-value eventuated by a model will be a real number colse to 1 to 2 。Where y is a function of condition vatiables such as concentrations of wear debris, ie. y=g(x1,x2,…xn) Initially consider the following regression models y= a0+ (1) Where a is the regression coefficient, x the concentration of elements, a the interruption, and n the number of elements in the oil analysis. The above model regression can be completed in Excel, which is a part of Microsoft Office According to the regression model(1), some coefficients are insignificant in the regression model .In order to stress the significant elements of the model as much as possible, some insignificant elements should be deleted from the model . The insignificant elements are indicated by p-values in Excel If the p-values are large, it is likely that the possibility of the related element regression coefficients is zero, and where the p-values are smaller the possibility is less In the paper the p-value 0.1 is taken as a significant criterion of elements, which means that the possibility of regression coefficient of a signigicant element being zero will be less than 10%. The procedure to delete all of the insignificant elements is as follows. Step 1 Regress all of elements of oil analysis , and output the p-values of all elements Check the p-values and select an element related to the maximum p-value. Step 2 Delete the elements related to the maximum p-value. Again regress the left element and output the p-values of the elements Check the p-values and select an element related to the maximum p-value. Repeat step2 until the p-values of the remsining elements are all less than 0.1. At this time, the modeling procedure is ended and the result model is y=a0 + (2) Although the state variavle y of the model(1) is only binary states 1and 2, the values of the output y of the result model (2) will generally not be exactly 1 and 2. If the output values are less than 1, the state y will belong to normal and if the output values are more than 2, the state y will abnomal .But if the values are between 1 and 2, it is vague whether the states are normal or abnormal. Therefore a threshold value is needed to judge the states of the output values. 3 Determination of the threshold value Once the model is built, according to the known normal and abnormal state of variable y, all samples can be divided into the two sub-samples(normal and abnormal), which can be transferred into two single-value samples of y in terms of the resulting model(2). Considering that the two single-value samples are from the resulting model, so it is reasonable that both of the single-value samples obey a normal distribution. Fit them into two distribution functions fA(y) and fN(y), and deternine the means)and standard deviations () of these functions as in Figure 1. Figure 1 The possible distribution functions(PDF) of normal and abnormal groups and the threshold value For any of y values from the resulting model (2), it si a problem to be solved that it belongs to normal or abnormal .For this reason a threshold value needs to be determined which is a critical value of y and denoted as .For any value, there are two types of judgment errors. (a) Normal state si wrongly judged as an abnormal state with the probabilitu. 1- FN(yc) (b) Abnormal state is wrongly judged as a normal state with the probability. FA(yc) The sum of the errors is given by S(yc)=1- FN(yc)+ FA(yc) (3) Where FA(yc)and FN(yc) are, respectively, the probability function of a normal state and the probability function of an abnormal state. For minimizing judgment errors, it is obvious that the value y is optimally determined by minimizing .The existence of the minimal value y has Seen proved in the Appendix . According to the Appendix the threshold value y can be easily determined . Now given an observation, we can calculate a y value using the developed model(2) and compare it with the threshold value. In this way the monitored machine`s state can be determined 4 Numerical example Data of the example from Reference {9} is shown in Table 1, which contains 8 elements(A1, Cu, Si , Pb, Cr, Mn,Ni , Fe).and 1 state variable(State).For the observed data of Table 1 the modeling procedure is described as follows At first using(1),we Can find that insignificant elements are successively Pb Cr Mn Ni and Fe, and that A1 Cu and Si are significant . For the 3 elements the p-values are,respectively,8.99，4.68 and 0.016472 and they are far less than 0.1. Thus we have the regressed model y=0.05166+0.549707 Al –0.19083Cu—0.15495Si (4) Second, now we can use the regressed model (4) to compute the state values of samples and divide these values into two groups by the means of the known normal and abnormal states of samples.Assume that y values for any group follow the normal distribution .We have Once we have two distributions and those parameters, we can optimally find the threshold value referring to the Appendix. The result is =14354 with the wrong judgment probability=397%. The curve of the total wrong-judgment possibility via threshold value y is shown in Figure 2 Now, we can chedk the prediciton power of the model . For the modeling samples, the values of state variable y computed by model (4)are listed in the y column of Table 1. The judged results of comparing y values with the threshold value are listed in the judgment column of Table 1, there is no wrong judgment for all samples .This indicates that the threshold value 1.4354 with the wrong judgment probability=3.97% is reasonable and that the above modeling procedure is also reliable. In order to verify further the correction of the model (4)and its threshold value , we can check the other 5 testing samples, the checked results of the 5 samples are shown in Table 2. From Table 2,we can see that there still are no wrong judgments for all samples.Therefore, we can take advantage of the model (4) and the threshold value to judge whether any new oil samples are normal or abnormal Based on the judgments ,some suggestions or actions of maintenance can be obtained, which will save more costs of maintenance. 5 Conclusions and discussions (a) The above modeling procedure is an improved version of Reference[9], which can effectively delete the insignificant elements of oil analysis data. The regression module of Excel can very simply finish the modeling procedure. The regressed model can transfer the oil samples into single-value state indexes. (b) Considering binary-state outcome for the observations, a method for optimally determining the threshold value has been proved. A numerical example has verified that the judged results of the modeling and testing samples are consistent with the outcome of observations. (c) The above approach has a feature of condition-based maintenance. For example, it can be usde to predict when a monitored item will reach the threshold value and take necessary actions. (d) In case of not enough samples, the judgment correction can be improved by modeling the combination of old samples and new samples, as more new samples are obtained. Thus the approach can be consummated by replenishing more new samples. (e) It is noted that the judgment may be wrong when the y-value is close to the threshold value. To avoid this, an interval including y should be further determined, within which the judgment needs to be confirmed by a further check or other methods . It is the next work to make the approach pergect. References [1] V.M. Martinez and B.T.Martinez, etal Results and benefits of an oil analysis programmer for milway locomotive diesel engines .Insight Vol45, No6,pp.402~406,2003 [2]G Nollet and D.Prince, Rotating equipment reliability for surface operation, Part Oil analysis in a mine CIM Bulletin Vol 96, No 1067, pp.82~86,2003 [3]R.W.Chapman D.J Hodges and T,J Nowell, Micro to macro-wear debris analysis as a condition monitoring tool Insight Vol 44.No,8,pp.498~502, 2002 [4]S. Berg, A study of sample withdrawal for lubricated systems Part 2: practical sample withdrawal and selection of proper sampling methods, Industrial Lubrication and Tribology, Vol 53, No.3, pp.97~107,2001 [5]R,Ong, J.H. Dymond R.D. Findlay and B.Szabados, Systematic practical approach to the study of bearing damage in a large oil-ring-lubricated induction machine .IEEE Transactions on Industry Applications, Vol 36, No 6, pp. 1715~1724,2000 [6[W.Wang P.A. Scarf and M.A.J. Smith, On the application of a model of condition-based maintenance Journal of the Operational Research Sooiety, Vol 51, No.11,pp.1218~1227,2000 [7]G. Fisher, Donalue A Filter debris analysis as a first-line condition monito ring tool Lubrication Engineerng, Vol 56, No 2,pp.18~22，2000 [8]V.Macia’n, B. Tormos, P.Olmeda and L Montoro, Analytical approach to wear rate determination for internal combustion engine condition monitoring based on oil analysis .Tribology International, Vol 36,NO 10, pp.771~776,2003 [9]C.H. Zhao, X.P.Yan etal . The prediction of ware model based on stepwise pluralistic regression 。Proceedings of International Conference on Intelligent Maintenance System, Xi’an, China, pp.66~72,Oct 2003 Brief biographies Fu jun-qing is now an associate professor of changsha university of science and technology, his research field is in mechanical vibration, fault diagnosis, signal analysis and so on. Li han-xiong is now a professor of central south university , his research fields is in fuzzy control, processes and intelligent control, process identification, and so on. Xiao xin-hua is now an associate professor of changsha university of science and technology. His research fields is in combustion engine engineering ,reliability and mechanical design. Appencix Let the probability function of normal state sample group be and the probability function of abnormal state sample group FN(yc)= (A-1) FA(yc)= (A-2) Then the function of wrong judgment probability is S(yc)=1-FN(yc)+FA(yc) S(yc)=1-+ (A-3) In oder to minimize the wrong judgment probability, differentiate the probability (a-3) respect to y, thus =+ (A-4) = (A-5) Simplifying the above equation, we have (A-6) The two sides of equation(a-6) are acted on with a in(*) function and let in ， then the equation(a-6) becomes （A-7） Simplifying and collecting the above equation (a-7), we have (A-8) Generally, the means of normal and abnormal sample groups are different and the mean is more than , that is and < (A-9) Under the conditions (a-9), according to whether the variance(standard deviation) is equal to or not , the equation(a-8) can be classified into the two cases as follows Case 1: = The equation(a-8) can simplified as =0 (A-10) The equation (a-10) has a sole solution of the threshold value of minimum wrong judgment probability = (A-11) (A-12) 形式2 In this case, equation(a-8) can be simply rewritten as (A-13) Where The equation(a-13) is general second order equation with one variable, the solution of roots is (A-14) For the solution(a-14), the judgment condition of existing the real roots is f=>0 (A-15) In fact, the condition from equation (a-13) can be simplified as f== 2 2 (A-16) For the judgmcnt cquation ,if > ，R=ln . It is obvious (A-17) And if < ,R= R=ln,the judgment formula (A一16) can be written can as( A一18) and it is also obvious that f > 0 ( A-18) Until to now , we have the proof that there are only real roots in the solution (a-14). Therefore both and are real roots. They are the two extreme points of the function of wrong judgment probability (a-3). According to the figure 1 of distributions, we can directly observe that one of the roots corresponds to a maximum value of probabiity (a-3), another to a minimum value, and the root to the minimum value should usually be less than and more then thus based on the these roots, we can determine the minimum threshold value of wrong judgment probability as follows If ，then If ，then (A-20) 基于油液分析的機(jī)械磨損狀態(tài)識(shí)別模式 付俊慶 李漢雄 肖新華 1． 長沙科技大學(xué)，汽車與機(jī)械工程學(xué)院，長沙 410076，P.R.china 2． 中南大學(xué)，機(jī)電學(xué)院，長沙 410083，P.R.china 摘要，本文提供了一個(gè)建模過程，這個(gè)過程源于回歸模型基礎(chǔ)上的單值測量方法和用以確定臨界值為正?；虍惓５臉?biāo)準(zhǔn)機(jī)械磨損狀態(tài)的統(tǒng)計(jì)方法。用這種算法和標(biāo)準(zhǔn)驗(yàn)算了一個(gè)實(shí)數(shù)例子。結(jié)果表明,基于油液分析機(jī)械磨損狀態(tài)正常與否的判斷方法和算法基本符合客觀事實(shí)。 關(guān)鍵詞，油液分析 退回模型 單值測量 和臨界值 1引言 石油分析方法已成為各行各業(yè)在世界范圍內(nèi)用于降低維修成本、提高生產(chǎn)率和可靠性的方法。目前，大部分石油分析儀使用發(fā)射光譜、電子顯微鏡、光學(xué)或鐵等方法進(jìn)行石油分析。石油分析的目的是探討從機(jī)械中提取潤滑油樣本所起的滑潤作用或設(shè)備的條件和設(shè)備推薦維修經(jīng)營活動(dòng)的行動(dòng)未拆機(jī)器,按照一定的關(guān)系能夠獲得機(jī)械的油樣樣本，并且通過分析油液的油液樣本和機(jī)械狀態(tài)來評(píng)估原設(shè)備廠商。潤滑油供應(yīng)者和油樣分析實(shí)驗(yàn)室提供具體的具有指導(dǎo)性的在油樣中磨損金屬的含量。這些限制提供了良好的解讀石油分析數(shù)據(jù)的一般準(zhǔn)則。但在石油數(shù)據(jù)分析中還有許多因素，按照石油分析數(shù)據(jù)很難直接判斷出機(jī)械的磨損狀態(tài)。在工程應(yīng)用中，在塞米松或測量中的單值對(duì)于檢測油樣狀態(tài)是必要的。從設(shè)備油液分析數(shù)據(jù)中導(dǎo)出機(jī)械的磨損率，用以確定引擎的磨損率（zwr）記作Z. (z)的指數(shù)值用來評(píng)價(jià)引擎正?；虍惓９ぷ鞯哪p率以參考一個(gè)正常的磨損率(E M)。事實(shí)上, 指數(shù)和參考指數(shù)應(yīng)該是石油樣本的隨機(jī)變量而不是確定的數(shù)值，因此, 該指數(shù)的統(tǒng)計(jì)模型需要發(fā)展一個(gè)依靠多元逐步回歸的模式，這個(gè)模式刪去了一些在油液分析的原始數(shù)據(jù)中無關(guān)緊要的元素和線性相關(guān)的元素，使石油分析數(shù)據(jù)轉(zhuǎn)換成了單值。這個(gè)單值作為磨損狀態(tài)的判斷依據(jù)，但沒有門檻或臨界值,作為這個(gè)值的鑒定標(biāo)準(zhǔn)。因此,樣本的這個(gè)值遠(yuǎn)偏離于正常狀態(tài)的值1或異常狀態(tài)的值2，還不清楚他們是屬于正常狀態(tài)還是異常狀態(tài)。本文首先完善了這樣一個(gè)建模函數(shù)，它是參考了基于回歸模型的單值測量。然后提出了一種確定閾值的統(tǒng)計(jì)標(biāo)準(zhǔn)作為辨識(shí)正?；虍惓顟B(tài)的方法。一個(gè)實(shí)數(shù)例子被所提供的方法和堅(jiān)定標(biāo)準(zhǔn)所檢驗(yàn)。結(jié)果表明有所提供的方法演算出的值基本符合客觀事實(shí)，因此這個(gè)方法和判別標(biāo)準(zhǔn)對(duì)于從油樣中判斷在正常或異常狀態(tài)下機(jī)械的磨損狀態(tài)是有價(jià)值的。 2模擬過程 2.1實(shí)驗(yàn)設(shè)計(jì)與抽樣 通過定期或不定期的收集和分析石油樣本來獲得油樣中的各種元素濃度。在此期間,通過例如拆卸機(jī)械，測量碎片形狀等其他方法來檢查和確定機(jī)械的磨損狀態(tài)。在這篇文章中，健康狀態(tài)用二進(jìn)制數(shù)值來表示，也就是說，正常狀態(tài)和異常狀態(tài)。一旦足夠的數(shù)據(jù)被采集到,這個(gè)實(shí)驗(yàn)就會(huì)中止，單值模型將被建立。 2.2建模 被觀察的健康狀態(tài)定義如下 1，正常狀態(tài) 狀態(tài)= 2 ，異常狀態(tài) 在另一方面,y值經(jīng)過一個(gè)模型的驗(yàn)算將要得到一個(gè)接近于1或2的實(shí)數(shù)。這里y值是各種狀態(tài)的函數(shù)，例如磨粒濃度，即 y=g(x1,x2,…xn) 初步考慮下列回歸模型 y= a0+ (1) 這里是回歸系數(shù)，是元素的濃度，是中斷，是石油分析中的元素?cái)?shù)量。 以上的回歸模型可以在Excel上完成,這是一個(gè)微軟辦公軟件的一個(gè)部分。根據(jù)回歸模型(1), 在回歸模型中的一些系數(shù)是微不足道的。為了盡可能多的壓縮模型中的大量元素，一些無關(guān)緊要的元素應(yīng)該從模型中刪掉.這些無關(guān)緊要的元素在Excel中由p值決定。如果p值很大，很可能是因?yàn)橄嚓P(guān)元素的回歸系數(shù)等于零，并且p值越小這種可能性越小。在這篇文章中，p值等于0.1被作為元素的一個(gè)重要基準(zhǔn)，0.1意味著一個(gè)有意義的元素的回歸系數(shù)為零的可能性不足10%。刪除所有的無關(guān)緊要的元素的步驟如下。 第一步 退回石油分析中的所有元素，輸出所有元素的p值。檢查p值并選出涉及最大p值的元素。 第二步 刪除涉及最大p值的元素。再一次退回到最左邊的元素并且輸出所有元素的p值。檢查p值并選出涉及最大p值的元素。 重復(fù)第二步直到剩余元素的p值全都小于0.1。這時(shí)，建模過程被完成，建模結(jié)果是 y=a0 + (2) 雖然模型1中的狀態(tài)變量y只是狀態(tài)1和2，但結(jié)果模型2的輸出y值一般不是準(zhǔn)確的1和2。 如果輸出值小于1，狀態(tài)值y將屬于正常狀態(tài)。如果輸出值大于2，狀態(tài)值y將是異常的。但如果值介于1和2之間，狀態(tài)是正常還是異常將要是模糊的。因此,需要一個(gè)臨界值來判斷該狀態(tài)的輸出值. 2.3 臨界值的測定 一旦模型被建立，按照已知的正常和異常狀態(tài)下的變量y，所有樣品可分為兩個(gè)小組樣品(正常與異常)，它根據(jù)計(jì)算模型2可以轉(zhuǎn)換成兩個(gè)單樣本的Y值?？紤]到兩個(gè)單樣品值來自計(jì)算模型，所以兩個(gè)單值樣本服從正態(tài)分布是合理的。把它們代入兩個(gè)分布函數(shù)fA(y)和fN(y)，并確定如1圖中這些函數(shù)的系數(shù)和偏差。并確定如圖1中這些函數(shù)的系數(shù))和偏差 () 圖1 正?；虍惓＝M的可能分布函數(shù)(PDF )和臨界值（閾值） 對(duì)于從計(jì)算模式2中得出的任何一個(gè)y值都是用于解決是屬于正常還是異常這個(gè)問題的。為此需要有一個(gè)臨界值來決定關(guān)鍵值y并記作。對(duì)于任何值，都有兩種判斷錯(cuò)誤的類型， (a) 正常的狀態(tài)可能被錯(cuò)誤的判斷為異常的狀態(tài)1-FN(yc)或 (b) 異常的狀態(tài)可能被錯(cuò)誤的判斷為正常的狀態(tài)FA(yc) 錯(cuò)誤的總和被給出