2009年--電氣工程及其自動化外文翻譯---基于模糊推理系統(tǒng)的失效模式與效應分析（fmea）

1、<p>　　2900單詞，14800英文字符，4740漢字</p><p>　　出處：Kai M T. On Fuzzy Inference System Based Failure Mode and Effect Analysis (FMEA) Methodology[C]// Soft Computing and Pattern Recognition, International Conferen

2、ce of. IEEE, 2009:329-334.</p><p>　　附錄A 外文翻譯原文</p><p>　　On Fuzzy inference system based</p><p>　　Failure Mode and Effect Analysis (FMEA) methodology</p><p>　　Kai Meng

3、Tay</p><p>　　Electronic Engineering Department, Faculty of Engineering,</p><p>　　University Malaysia Sarawak</p><p>　　Sarawak, Malaysia</p><p>　　kmtay@feng.unimas.my<

4、;/p><p>　　Abstract-Filure Mode and Effect Analysis (FMEA) is a popular problem prevention methodology. It utilizes a Risk Priority Number (RPN) model to evaluate the risk associated to each failure mode. The co

5、nventional RPN model is simple, but, its accuracy is argued. A fuzzy RPN model is proposed as an alternative to the conventional RPN. The fuzzy RPN model allows the relation between the RPN score and Severity, Occurrence

6、 and Detect ratings to be of non-linear relationship, and it maybe a more realis</p><p>　　Keywords: Fuzzy inference system, monotonicity property, sufficient conditions, FMEA, manufacturing</p><p&

7、gt;　　Ⅰ NTRODUCTION</p><p>　　Failure Mode and Effect Analysis (FMEA) is an effective problem prevention methodology that can easily interface with many engineering and reliability methods [1]. It can be descr

8、ibed as a systemized group of activities intended to recognize and to evaluate the potential failures of a product/process and its effects [2]. Besides, FMEA identifes actions which can eliminate or reduce the chances of

9、 potential failures from recurring. It also helps users to identify the key design or process charact</p><p>　　Conventional FMEA use a Risk Priority Number (RPN) to evaluate the risk associated to each failu

10、re mode. A RPN is a product of the risk factors, i.e., Severity (S), Occurrence (O) and Detect (D). FMEA assumes that multiple failure modes exist, and each failure mode has a different risk level that have to be evaluat

11、ed, and ranked. In general, S, O and D are of integer 1 to 10, usually defined in scale tables. </p><p>　　From literature, the use of Fuzzy Inference System (FIS) in FMEA is not new. Bowles and Pelez suggest

12、 to replace the conventional RPN model with a FIS (fuzzy RPN model) [3]. The fuzzy RPN model allows the relationship between the RPN score and the three risk factors (S, O, and D) to be of a non-linear relationship, whic

13、h is too complicated to be modeled by the simple conventional RPN model. Motivation of FIS to be chosen in this problem domain can explained with FIS ability to incorporate human/</p><p>　　The fuzzy RPN mode

14、l is a popular method, and has been successfully applied to a number of FMEA problems. For example, it was applied to FMEA of an auxiliary feed water system and a chemical volume control system in a nuclear power plant [

15、6, 7]. It was also used in FMEA of an engine system [8], a semiconductor manufacturing line [9], and a fishing vessel [10]. Over the years, several enhancements have also been proposed to the fuzzy RPN model. Development

16、 of a fuzzy RPN model using the grey relat</p><p>　　However, little attention is paid on the validity and the efficiency of the estimated RPN scores, as available in the literature. Therefore, in this paper,

17、 the efficiency of the fuzzy RPN model, in order to allow valid and meaningful comparisons among different failure modes in FMEA to be made is investigated. The fuzzy RPN model is viewed as an assessment or measurement m

18、odel, which is subjected to some theoretical properties of a length function, e.g. monotonicity, sub-additivity and etc [11].</p><p>　　Investigation in this paper focuses on monotonicity property of the fuzz

19、y RPN model. The fuzzy RPN model is firstly presented. Monotonicity property in FIS and a sufficient condition for a FIS to be monotone is also reviewed. Monotonicity property for the fuzzy RPN model is further defined.

20、 In this piece of work, the sufficient conditions for a FIS to be monotone, as pointed in various sources [12, 13, 14, 15], is applied to the fuzzy RPN model. The sufficient conditions pointed out that for a </p>

21、<p>　　This paper is organized as follow. In section II, the fuzzy RPN model is reviewed. In section III, the sufficient condition for a FIS to be monotone is presented.In section IV, the applicability of the suffi

22、cient condition to the fuzzy RPN model is discussed.Section V reports case studies with data/information collected from a FCBGA plant. Concluding remarks is then presented. </p><p>　　II REVIEW ON THE FUZZY

23、 RPN MODEL</p><p>　　Conventional RPN model is used to evaluate the risk associated with each failure mode in FMEA.Generally,the conventional RPN model takes three factors, i.e., S, O, and the RPN scores is d

24、etermined by the multiplication of these three inputs scores, as shown in (1). </p><p><b>　　(1)</b></p><p>　　In general, these three factors are estimated by experts in accordance wi

25、th a scale from”1”to“10”based on commonly agreed evaluation criteria.Tables 1, 2, and 3 summarize the evaluation criteria for S, O and D ratings,respectively,which is used practically in a semiconductor manufacturing pla

26、nt.</p><p>　　TABLEⅠ. SCALE TABLE FOR SEVERITY</p><p>　　TABLEⅡ. SCALE TABLE FOR OCCURANCE</p><p>　　TABLE Ⅲ. SCALE TABLE FOR DETECT</p><p>　　Even through the tradition

27、al RPN model is simple and has been well accepted for safety analysis, it suffers from several weaknesses. In [3], it is pointed out that the same RPN score can be obtained from a number of different score combinations o

28、f S, O, and D. Although the same RPN is obtained, the risk can be different. Besides, is it argued that the relative importance of S, O and D maybe different.The fuzzy RPN model is proposed in [3], as a solution to these

29、 problems. In the fuzzy RPN, a FIS</p><p>　　Membership functions of S, O, and D can be generated based on the criteria in Tables 1, 2 and 3 respectively. Figures 1, 2, and 3 depict the fuzzy membership funct

30、ion for, O()and D（）, respectively. As an example, referring to Fig. 1, the second membership function of S, with linguistic label of “Law” represents S ratings from 2 to 5, which correspond to “yield hit,cosmetic,impact,

31、special internal handling, effort or annoyance”as in Table1.The same scenario applies to Fig.2,where the “Moderate”me</p><p>　　Output of the fuzzy RPN model, RPN score is varies from 1 to 1000.In this case s

32、tudy, it is divided into five equal partitions, with fuzzy membership functions, B,“Low”, “Low Medium”,“Medium”,“High Medium” and “High”,respectively. The corresponding b scores are assumed to the point where membership

33、value of B is 1.Hence b,is 1,250.75,500.25, and 1000, respectively.</p><p>　　A fuzzy rule base is a collection of knowledge from experts in the If-Then format. Considering S, O, and D, and their linguistic t

34、erms, the fuzzy rule base has 180(5(S) ×6（O）×6（D）) rules in total using the grid partition approach [4,5]. As an example, Fig.4 show two rules that describe a small portion of the fuzzy rules collected from waf

35、er mounting process engineers.</p><p>　　In this paper, a simplified Mamdani FIS [3, 4] is used to evaluate the RPN, as in (2). (2) can be viewed as zero order Sugeno FIS model. </p><p><b>

36、　　(2)</b></p><p>　　Ⅲ REVIEW ON THE SUFFIENT CONDITIONS OF A FIS TO BE OF MONOTONICITY</p><p>　　If for all and such that <, then for a function f to be monotonically increasing or decre

37、asing, the condition ormust be fulfilled, respectively.</p><p>　　From the literature review, there are a lot of investigations on the monotonicity property of FIS models.Developments of FIS models that fulfi

38、l the monotone constraint are also available.Zhao and Zhu examined the condition for an FIS to be monotone, and analyzed the FIS operations step by step [12]. Their findings revealed that as long as the rule base is mono

39、tone,a single-input Mamdani fuzzy model can be monotone, and a two-input Mamdani fuzzy model can be roughly monotone. </p><p>　　Another attempt to study the monotone property is to differentiate the output

40、of an FIS with respect to its input(s). Won [13] derived the sufficient conditions for the first order Sugeno fuzzy model with this approach. From [14,15,16],the sufficient conditions for a zero order Sugeno FIS to be mo

41、notone is reported.</p><p>　　For a FIS to be monotone, the sufficient conditions stated that two conditions are needed. Condition(1) can be viewed as a method how membership function should be tuned in orde

42、r to ensure a FIS to be of monotonicity property. Assume both and are differentiate-able. For <, condition as in (3) has to be fulfilled.</p><p><b>　　(3)</b></p><p>　　Condition (

43、2) highlights the important of having a monotonic rule base in the FIS model. These two conditions are very useful, as it can be directly applied to various FIS related techniques.It was later combined with least squar

44、e learning [17], and evolutionary computation-based learning [18]. These conditions are further extended to a multiple stage FIS [15]. </p><p>　　IV APPLICATION OF THE MONOTONICITY PROPERTY AND THE SUFFICIEN

45、T CONDITIONS TO THE FUZZY RPN MODEL</p><p>　　In this paper, it is proposed that the fuzzy RPN model is of monotonicity property [19]. Similar to the traditional RPN function, S, O, and D of the fuzzy RPN a

46、re defined in such a way that the higher the input scores, the more critical the situation. The output RPN is a measure of the failure risk. </p><p><b>　　Yes</b></p><p>　　Correction

47、required</p><p>　　Figure The proposed fuzzy RPN model and procedure</p><p>　　For example, for two failures with input sets of 5 5 6 and 5 5 7 (S, O, and D), the fuzzy RPN for the second failure

48、 should be higher than that of the first.This can be explained with referring Tables 1, 2 and 3.These two failures are of the same S and O scores, but with D score of 6 and 7 respectively.Failure with D score of 6 (“Cont

49、rols are able to Detect within the same functional area”) represents a better control mechanism than that of D score of 7 (“Controls may not Detect excursion until </p><p>　　The monotonicity property in this

50、 paper suggests that if any of the two scores are static, to allow valid comparison, as the third score increases, the RPN score should not decrease. </p><p>　　To fulfill the monotonicity property, the suffi

51、cient conditions is used in the fuzzy RPN model. Fig.5 depicts the flow chart for the author’s proposed fuzzy RPN model. Condition (1) can be used as a criterion to tune membership function.Membership functions for S, O

52、, and D are tuned with accordance to Condition (1).Figures 1, 2 and 3 illustrate the membership function for S,O and D respectively,which fulfill Condition (1).</p><p>　　Condition (2) can be viewed as criter

53、ia for a set of valid rule base.From Fig.5, Condition (2) can be used to check the validity of the collected rule base.</p><p>　　Ⅴ CASE STUDY AND EXPERIMENTAL RESULTS</p><p>　　To validate the

54、proposed approach, experiments with data/information collected from a semiconductor manufacturing processes of Flip Chip Ball Grid Array (FCBGA) products is conducted.FCBGA is a low cost semiconductor packaging solutio

55、n which utilizes the Controlled Collapse Chip Connect technology, or which is known as Flip Chip (FC) for its die to substrate interconnection.FC was initiated at the early 1960s to eliminate the expanse, unreliability,

56、and low productivity of manual wire bonding</p><p>　　Tables 4 and 5 summarize the failure risk evaluation, ranking, and prioritization results using the traditional and fuzzy RPN models for the wafer mountin

57、g and underfill dispensing process. Columns “Sev”(Severity), “Occ” (Occurrence), and “Det”(Detect) show the three ratings that describes each failure.Failure risk evaluation and prioritization outcomes based on the tradi

58、tional RPN model are shown in columns “RPN” and “RPN rank”respectively. </p><p>　　Column “Fuzzy RPN (FPR)”shows the failures risk evaluation results using the fuzzy RPN model, while sub-columns “FRPN” and “F

59、RPN Rank” show its failure risk evaluation and prioritization outcomes, respectively. Column “Expert’s Knowledge (FPR)” shows the linguistic term assigned by process engineers. </p><p>　　For example, in Tab

60、le 4, failure mode “1” represents “broken wafer” which leads to yield loss, and is given a S score of 3 (refer to Table 1). This failure happens because of “drawing out arm failure”, and because it rarely happens,it is a

61、ssigned an O score of 1 (refer to Table 2). In order to eliminate the cause, software enhancement has been done as action taken. Owing to the action taken is very effective, and can almost eliminate the root cause; a D s

62、core of 1 is given (refer to Table 3).Usi</p><p>　　The monotonicity property is important to allow a valid comparison among 2 failures mode, for example failure mode “1” and “3” in Table 4. Failure model “3”

63、is of S, O, and D of 3, 2, and 1 respectively. Monotonicity property suggests that failure mode “3” should have a higher fuzzy RPN score (fuzzy RPN=19) than that of “1” to allow valid comparison. </p><p>　　T

64、he same scenario can be observed in Table 5, a case study on underfill dispensing process. To allow valid comparison among failures mode “1”, “2” and “3”, fuzzy RPN score for “3” should not be lower than “2” and fuzzy RP

65、N score of “2”should be lower than”1”. With the use of the sufficient condition, Fuzzy RPN score of 1, 1 and 2 is assigned to failure mode “1” “2” and “3”respectively. </p><p>　　From the observation, the fu

66、zzy RPN model is able to fulfill the monotonicity property for all failures. There are no illogical predictions found in both case studies. As summary, as long as Condition (1) and Condition (2) are fulfilled, the mono

67、tonicity property can be ensured. </p><p>　　Ⅵ SUMMARY</p><p>　　In this paper, it is argued that the fuzzy RPN model should be subjected to some theoretical properties of a length function.This

68、is important, as it will ensure the validity of the RPN score, in order to allow comparisons among different failure modes in FMEA. This paper provides a simple and easy approach to construct the fuzzy RPN model in prac

69、tice from FMEA users.It is suggested that the sufficient conditions for a fuzzy inference system to be of monotonicity to be applied to the fuzzy RPN </p><p>　　Experiments have been conducted to evaluate the

70、 proposed approach.Experiment is conducted with data and information from one of the manufacturing processes in a FCBGA manufacturing plant, i.e., wafer mounting and underfill dispensing processes. These experiments give

71、 promising results.</p><p>　　Investigation of a fuzzy inference system based assessment model (fuzzy RPN in particular) to fulfill other properties of a length [11], i.e. sub-additivity can be a good researc

72、h topic. </p><p>　　REFERENCES</p><p>　　[1] W. Grant Ireson, Clyde F. Coombs, JR and R. Y. Moss. Handbook </p><p>　　Reliability Engineering and Management. McGraw-Hill Professional

73、; 2nd edition, 1995. </p><p>　　[2] Chrysler Corporation, Ford Motor Company and General Motors </p><p>　　Corporation (SAE J 1739). Potential Failure Mode And Effect analysis (FMEA) Reference

74、Manual, 1995. </p><p>　　[3] J.B. Bowles, and C.E. Pelez, fuzzy logic prioritization of failures in a system failure mode, effects and criticality analysis Reliability Engineering and System Safety, vol. 50,

75、1995, pp. 203~213.</p><p>　　[4] J.S.R. Jang and C.T. Sun & E. Mizutani. Neural-Fuzzy and soft Computing. Prentice-Hall, 1997. </p><p>　　[5] C.T. Lin and C.S.G.Lee. Neural Fuzzy Systems: A N

76、euro-Fuzzy Synergism to Intelligent systems, 1996. </p><p>　　[6] A.C.F.Guimars, and C.M.F. Lapa,Effects analysis fuzzy inference system in nuclear problems using approximate reasoning?, Annals of nuclear Ene

77、rgy, vol. 31, 2004, pp. 107~115. </p><p>　　[7] A.C.F. Guimars, and C.M.F. Lapa, Fuzzy FMEA applied to PWR chemical and volume control system Progress in Nuclear Energy, vol. 44, 2004, pp. 191~213. </p>

78、<p>　　[8] K. Xu, L.C. Tang, M. Xie, S. L. Ho and M.L. Zhu., Fuzzy assessment of FMEA for engine systems? Reliability Engineering and System Safety, vol. 75, pp. 17~19, 2002. </p><p>　　[9] K

79、.M. Tay, and C.P. Lim, Fuzzy FMEA with Guided Rules Reduction System for Prioritization of Failures? International Journal of Quality Reliability Management, vol. 23, 2006, pp. 1047~1066. </p><p>　　[10]

80、A. Pillay, and J. Wang, Modified failure mode and effects analysis using approximate reasoning, Reliability Engineering & System Safety, vol. 79, 2003, pp. 69~85. </p><p>　　[11] Inder K Rana, An introdu

81、ction to Measure and integration, Alpha Science international, 2005 </p><p>　　[12] H.Zhao and C.Zhu, Monotone fuzzy control method and its control performance, IEEE International Conference on Systems, Man,

82、and Cybernetics, 2000 vol. 5, 2000, pp. 3740-3745. </p><p>　　[13] J.M. Won, S.Y. Park, and J.S. Lee, Parameter conditions formonotonic Takagi-Sugeno-Kang fuzzy system, Fuzzy Sets and Systems, vol.132, 2002,

83、pp. 135~146. </p><p>　　[14] K.M. Tay, and C.P. Lim, On the Use of Fuzzy Inference Techniques in Assessment Models: Part I: Theoretical Properties, Fuzzy Optimization and Decision Making, vol. 7 , Issue.3, 20

84、08, pp. 269~281. </p><p>　　[15] K.M. Tay, and C.P. Lim, On the Use of Fuzzy Inference Techniques in Assessment Models: Part II: Industrial Applications, Fuzzy Optimization and Decision Making, vol. 7 , Issu

85、e.3, 2008, pp. 283.~302. </p><p>　　[16] V.S. Kouikoglou and Y. A. Phillis, On the monotonicity of hierarchical sum-product fuzzy systems, Fuzzy sets and systems,2009. doi:10.1016/j.fss.2009.02.001. </p>

86、;<p>　　[17] K. Koo; J.M. Won, J.S. Lee, Least squares identification of monotonic fuzzy systems, Fuzzy Information, 2004. Processing NAFIPS opos 2004. IEEE Annual Meeting of the vol. 2, Issue , 27~30, 2004, pp. 7

87、45 ¨C 749.</p><p>　　附錄B 外文文獻譯文</p><p>　　基于模糊推理系統(tǒng)的失效模式與效應分析(FMEA)</p><p>　　馬來西亞沙撈越大學工程學院電子工程系</p><p>　　KAI MENG TAY</p><p><b>　　馬來西亞沙撈越</b>

88、;</p><p>　　kmtay@feng.unimas.my</p><p>　　摘要：失效模式與效應分析是一種通用的解決問題的方法。它利用風險優(yōu)先數(RPN)模型來評估每個失效模式相關的風險。傳統(tǒng)的RPN很簡單但是準確性不夠。所以提出用一個模糊的RPN模式來替代傳統(tǒng)RPN模式。模糊的RPN模型允許RPN值和嚴重度，發(fā)生率，檢測評級之間的關系是非線性關系，它或許是更切合實際的表達形式。

89、本文中，為了對模糊的RPN模型的效率進行研究，允許在FMEA的不同故障模式之間進行有效的和有意義的比較。本文建議，模糊理論應受到RPN長度函數如單調性，子可加等些許屬性的制約。在本文中介紹重點是單調性。首先定義模糊RPN的單調性，FIS具有單調性的充分條件應用于模糊的RPN模型。這是一個在實踐中構建模糊RPN方便，可靠的指導。本文將會在半導體制造過程中構建模糊RPN模型。</p><p>　　關鍵字：模糊推理系統(tǒng)

90、(FIS)，單調性屬性，充分條件，FMEA，制造</p><p><b>　　1 簡介</b></p><p>　　失效模式與效應分析方法(FMEA)是一種有效以及可靠的預防故障發(fā)生的方法，它可以方便應用于許多工程問題。這種方法可以描述為一個系統(tǒng)化活動目的組旨在識別并評估產品/過程中潛在的故障和它的影響。此外，FMEA標識可以消除或減少再次出現的潛在故障的可能性的機會

91、。它還可以幫助用戶確定關鍵設計或要求特殊控制的制造工藝，并突出特性控制或性能的改進的部位。</p><p>　　傳統(tǒng)FMEA使用風險優(yōu)先數(RPN)來評估每個故障模式相關的風險。RPN是一種產品的風險因素即嚴重度(S)發(fā)生率(O)和檢測度(D)，FMEA假定多個故障模式存在，并且每個故障模式具有不同的風險級別那么必須對風險進行評估和排名。一般來說，S、O和D等級評分為1-10，通常在分攤比額表中定義。</p

92、><p>　　從文獻中得知，在FMEA中使用模糊推理系統(tǒng)并不是新穎的。Bowles and Pelaze建議用FIS(模糊RPN模型)取代傳統(tǒng)的RPN模型，模糊RPN模型允許RPN分數和三個風險因素(S、O、D)之間的關系是非線性的。實際問題由于過于復雜，難以用傳統(tǒng)的RPN模型建模。在此問題中FIS的選擇動機可以用FIS納入人權/專家知識的能力解釋，在FIS中信息由含糊不清和不精確的語言描述。多年來FIS在各種應用領

93、域例如控制，建模，分類問題中表現出了它的優(yōu)點。</p><p>　　模糊的RPN模型是一種通用的解決問題的方法，而且一直成功的應用于大量FMEA問題。例如，它應用于一個輔助反饋水系統(tǒng)的FMEA和中核化學控制系統(tǒng)[6,7]。他也被用在引擎系統(tǒng)的FMEA，如半導體制造線和漁船系統(tǒng)。多年來提出了模糊RPN模型的幾項增強功能。文獻[10]中提出了使用灰色關聯理論的模糊RPN模型的發(fā)展。文獻[8]提出模糊RPN模型所有故障

94、之間的相互依懶性。文獻[9]中提出了減少模糊RPN模型中的模糊規(guī)則數量的辦法。</p><p>　　然而，在可用的文獻中很少關注RPN估值的有效性和效率。因此本文中，為了對模糊的RPN模型的效率進行研究，允許在FMEA的不同故障模式之間進行有效的和有意義的比較。模糊的RPN模型被認為是評估或測量模型，這是由于長度函數的一些理論屬性例如單調性，子可加等的影響。</p><p>　　本文中重點

95、介紹的是模糊RPN模型屬性中的單調性。首先介紹模糊RPN模型。FIS的單調性和FIS是單調的充分條件也是需要檢驗的。進一步定義模糊RPN模型的單調性，在本文中FIS是單調的充分條件應用于模糊RPN模型。充分條件指出一個FIS是單調的需要兩個數學條件。條件(1)可以被看做一種方法，如何調整隸屬函數以確保模糊RPN模型的單調性屬性，條件(2)突出了模糊模型中具有單調規(guī)則庫的重要性。這可以被看做是一個簡單的準則關于如何在。實踐中構建模糊RPN

96、。為了更進一步評估提出的方法，做了從半導體反轉芯片球柵格陣列(FCBGA)制作過程中收集數據的實驗。</p><p>　　本文的結構如下，第二節(jié)對模糊RPN模型進行檢驗。第三節(jié)提出了FIS單調的充分條件。第四節(jié)對模糊RPN模型的充分條件的適應性進行了討論。第五節(jié)有關案列從FCBGA制作過程手機數據/信息。最后提出結論性意見。</p><p>　　2 模糊RPN模型檢驗</p>

97、<p>　　傳統(tǒng)RPN模型被用來評估與FMEA每個故障模型相關的風險，一般情況下，傳統(tǒng)RPN模型需要三個因素即S、O和D。RPN值是三個輸入值相乘決定，如式(1)中所示。</p><p><b>　　(1)</b></p><p>　　一般情況下，這三個因素由專家估計并按照規(guī)模從1-10共同制定評價標準表。表1、2、3分別總結了S、O、D的評價標準，這被應

98、用到實際的半導體制作工廠。</p><p>　　表1 嚴重度的評價標準表</p><p>　　表2 發(fā)生率的評價標準表</p><p>　　表3 可探測度評價標準表</p><p>　　即使傳統(tǒng)RPN模型是簡單模型而且被用作安全性分析，但它有幾個弱點。3中指出相同的RPN風險系數可以由不同的S、O、D數的組合得到。雖然可以獲得同一RPN

99、值，但是其風險可以是不同的。此外S、O、D允許有不同的相對重要性。提出用模糊RPN模型作為一中解決方案來解決上述提到的問題。在模糊的RPN模型中，使用FIS模型進行代替?zhèn)鹘y(tǒng)RPN模型。它假設RPN值和輸入之間的分數即S、O、D是非線性關系。</p><p>　　分別按表1，2和3的標準可以生成S、O和D的隸屬函數。圖形1，2和3分別描述的是模糊隸屬函數S()，O()和D()。舉個例子，圖1所指，S函數的第二個曲線