Journal of Mechanics Engineering and Automation 5(20 1 5)362-369 doi:1O.17265/2159—5275/2015.06.005 …N6 Comparison of Procedures to Evaluate the Cyclic Stress-Strain Curve from Incremental Step Test Giovanni Zonfrillo and Duccio Nappini Department ofIndustrial Engineering,University ofFlorence,Florence 50139,italy Abstract The design of structural components in low-cycle fatigue field often requires the knowledge of the cyclic properties of the material,which are commonly described by the classical relation of Ramberg-Osgood.In order to obtain the cyclic curve using experimental data from incremental step tests,four methodologjes are described and critically discussed.Three methods differ in the procedure of evaluation of the elastic modulus.while in the last one the experimental data are interpolated with a single non—linear regression.The various techniques were applied to data obtained from tests camed out on stainless steel specimens,and the resulting differences were analysed and quantiied.An average behavifour was evaluated considering the totaI set of data obtained from experimenta1 tests.The choice ofthe most suitable method is related to both the strain range of interest and the goal for which the results are used. Key words:Cyclic curve,incremental step test,Ramberg-Osgood,stainless steel 1.IntrOductiOn hysteresis loops of metals and alloys and,in During service,structural components can be subject to stress that exceeds the elastic limit,especially when local stress concentration is involved.For these applications,fatigue design requires a special attention conjunction with Neuber’s rule,to evaluate the local notch tip stresses and strains values[2].A practical drawback of Eq.f 1)is that the relation is not analytically invertible and it is necessary to obtain a solution of this problem using numerical procedures or tO the evaluation of stress and strain fields in the critica1 areas.The stress・strain responses of many materials change significantly when elasto—plastic cyclic loading conditions are applied,and therefore,it is necessary to refer to stable conditions described by cyclic stress—strain curve.From the analytical point of view, the cyclic curve can be expressed in the classic form of approximate expressions[3]. Therefore,the cyclic curve is characterized by the values of the three materiaI coe所cients:/C.,z and In literature,various approaches are presented in order to estimate the materials cyclic properties from experimental monotonous data,using analytical a Ramberg—Osgood relation: formulations[4—6】or neural network interpolation methodologies [7, 8】. Unfortunately,these formulations have a statistical nature and should be (1) where, and Ca are the stress and the strain ampliude,t applied only to material classes(usually stee1)used for their validation.From a theoretical point of view,it is possible to estimate and力’from compatibility respectively,E is the elastic modulus,/C is the cyclic strength coeficient and ifv/’is the cyclic strain equation[9]using oligocyclic fatigue data,but this approach leads to a roughly evaluation of true values, with a significant discordance respect to experimental results. Numerous correlations between these hardening exponent[1].Admitting the Masing hypothesis,this relationship is used to describe the COrresp0nding author: Giovanni Zonfrillo, associate parameters and in particular between the exponents professor,Ph.D.,research field:behaviour ofmaterials,fatigue, life prediction.E—mail:giovanni.zonfrillo@uniif.it. were proposed in other studies【10,l l】. 364 Comparison of Procedures to Evaluate the Cyclic Stress-Strain Curve from Incremental Step Test loop in the block.These values,plotted in a stress-strain diagram,represent the data which must be interpolated to obtain the analytic expression of the cyclic curve(referred to the chosen block of cycles), as shown inFig.2. The determination of the three parameters E, ,n’ could be performed with two different approaches. The ifrst approach needs frstly the determination of Young’s modulus to decompose the total strain in the elastic and plastic components;the remaining two parameters could be assessed with a linear regression 0f versus plastic strain amplitude in the bi—logarithmic scale.It is clear that to obtain a reliable prediction of the material behaviour an accurate estimation of Young’s modulus is required.The alternative approach requires a complex numerical procedure:the three values which involve the best fit of the experimental data are evaluated at the same time by means of a non—linear regression analysis.In this case,the Young’s modulus is the“cyclic’’value of the parameter. 3.1 Method 1 In order to estimate the Young’s modulus three different procedures have been applied,namely l a,l b and 1c. 3.1.1 Procedure 1a Only the tips ofthe stable elastic hysteresis loops are considered and the elastic modulus is estimated as the angular coefifcient ofthe best ifting straight line using least squares regression analysis.The data in the elastic ifeld are identiifed by an iterative procedure: ・The best fit line for a increasingly number of data starting from zero is estimated; ・The value of the Euclidean norm of the regression residuals is calculated; ・The largest set of data for which the Euclidean norm is lower than a threshold value.assumed equal to 1 0 MPa.is determined. 3.1.2 Procedure 1b The values of the unloading modulus(both in tension and in compression)for each hysteresis loop are estimated and an arithmetic mean is carried out. The range of values on which the linear regression is performed is based on two parameters,expressed as percentage of the stress and strain peak values[1 6】;in detail,the range considered starts from a point for which the strain varies 2%compared to the peak value and it has an amplitude set to 60%ofpeak stress. 3.1.3 Procedure 1c This methodology requires all the experimental points recorded in the terminal part of the block(in particular in the last 6 cycles),surely in the elastic region.The value of Young’s modulus is obtained as the angular coefifcient of the straight line that better interpolate these data set. 3.1.4 Evaluation of the Points in the Plastic Region Once estimated the elastic modulus value,it is necessary to determine the points belonging to the plastic strain field in order to perform a linear regression in bi-logarithmic scale and to obtain the n’ and K’values. From a theoretical point of view would be enough to identify the ifrst point(with the lowest stress value)for which the plastic strain is greater than zero. Unfortunately,in the elastic field the acquired experimental values deviate a bit from the straight line that describe their trend,since an interpolation error occurs.To obtain the first point inside the plastic range, ifve consecutive recorded points must be overcome,on average,a positive threshold value of plastic strain, assumed equal to 0.01. 3.2Method2 In this method,the three characteristic material parameters of the analytic expression of the cyclic curve are evaluated at the same time using a least squares root method.The Ramberg Osgood relationship is the nonlinear model to be used. Therefore,the dependent variable s口is function of both the independent variable oa and the trhee parameters E, ’,K’:ga=/’( ,E, ’, ). Comparison of Procedures to Evaluate the Cyclic Stress-Strain Curve from Incremental Step Test 365 With reference to a generic experimental point with Table 1 Values of modulus of elasticity evaluated with variOUS methods. coordinates(aai,gai),the relationship Si:[Cai一 f,£, n’,K’)] defines the square ofthe deviation between the Specimen Modulus of elasticity(GPa) Method 1a Method 1b Method 1c Method 2 measured strain at i-th point and the value assumed by the function f when dn=8ni.The cyclic parameters values will be those which minimize the sum of Si. evaluated on the whole set of points. Whereas the linear regression model perforlTIS a closed—form solution,this method requires an iterative procedure to converge to a solution.For this purpose, was developed a Matlab code which is able to estimate the material coefifcients E,n’,K’,starting from the values obtained by the procedure 1 c.In order to ensure the reliability of the solution,it was verified that the result obtained was not a function of starting values.Indeed starting values affect only the number of iterations to obtain the convergence. 4.Application to X22CrMoV12.1 Stainless Steel The methodologies were applied to data obtained from tests carried out on X22CrMoV 1 2—1 stainless stee1.The alloy showed cyclic softening. 4.,Young Modulus The Young’s modulus values,calculated with the various procedures,are shown in Table 1. The scatter of data(related to the average value) obtained on five samples,regardless of the applied procedure,shows a mean values 1ess than 2.5%.with the highest value related to method 2. The analysis of the results obtained on each sample, as the method changes,shows values of dispersion slightly higher;approach l a gives on average lowest values,instead the hi ghest values are given by approach 2. However,it should be noted that the value of E obtained with procedure 2 js not the true value of material Young’s modulus;in this case,E is only one of three variables suitable to describe the material behaviour.The Ramberg-Osgood relation satisfactorily describes the plastic part of the curve Ca- ,but it does not consider an initial linear behaviour.because the plastic component of the strain is always present for anyvalue ofaa. The value of the elastic modulus obtained with the method l a may be signiifcantly different from the one evaluated in a tensile test[1 6].Moreover,an additional dififculty is due by the fact that the points used ofr the regression analysis may be insufifcient to an accurate estimation,since they depend on the number of cycles inside the imposed】oad block. The evaluation performed with procedure 1 b could be present a variability factor due to the possible diference in the traction.compression material behaviour.If the values achieved in the two different phases are different,the cyclic curve evaluation should be carried out separately for traction and compression stresses.In the case studied the difference is of the order of 3%.with highest value in traction. 4.2 Cyclic Strain Hardening Exponent and Strength Coeifcient The values of hte parameters obtained ofr the plastic component of strain amplitude are shown in Table 2. It is observed a greater dispersion than the one obtained for the Young’s modulus,especially as regards the value of . Again,the scatter among the results obtained applying the four methodologies to a single sample are hi曲er than that obtained using the same methodology on the various samples. On the other hand,these considerations valid rfom a mathematical point of view.are not directly linked to 366 Comparison of Procedures to Evaluate the Cyclic Stress-Strain Curve from Incremental Step Test Table 2 Values of n’and evaluated with various methods. the reliability of the material behaviour evaluation, since it is important to evaluate the dispersion in the cyclic curve,given by the set of the three parameters, and not the scatter of the single coeficients.f the curves in the range 0< <O.02 are reported in Fig. 4.Table 3 shows the maximum,minimum and mean values of the difference between these areas,estimated on each sample.These data are related to the dispersion of results attributable to the materia1. .3 Dispersion ofResults .Different values for E.,z’and K’could lead to cyclic 4 Methodologies Comparison I^ -互 喜 鼍lI舞ll“l=l: curves that from an engineering point of view are suficiently approximate;in order to estimate the data fdispersion,it is useful to calculate the differences in a well-defined strain range.It is necessary to set a Fig.5(related to sample 1)shows a compari son 0 Ifl 0 ∞ 0 ” 0 950 瘩 maximum threshold value,because increasing the strain amplitude the cyclic curves tend to diverge progressively from each other.The threshold value was chosen equal to 0.02.because for highest values the life { = 一: … } } i of a component would be only few cycles and so not interesting for engineering applications. ●一4 To perform the comparison,it is necessary to define a suitable parameter proportional to the deviation between the various curves,so as to be able to carry out a quantitative evaluation of the dispersion. various specimens. :三.S 0 00 0 0 0 0: 0 0: Strain amplltude Fig.3 Cyclic curves obtained by applying the method 2 to An obvious choice would be to determine the stress value obtained in the various cases for the same strain; extending this parameter to a range,the area under the curves evaluated in the strain range considered was 00. Sample U: used.As a first step the cyclic curves,obtained by applying the same methodology to the five tested samples are compared.Fig.3 shows for example the 臻2 羹 一 curves obtained with the procedure 2;similar representations are obtained also with the other methodologies.The highest differences,as expected, occur essentially in the plastic field.In order to quantify these differences,the values of the area under i2.巷 :2.2 lethodla ̄kthodlb Xkthodlc X ̄thod2 Fig.4 Area under cyclic curves evaluated with the four methods. 一 一薹)【睾Ⅱl 鲁耋 篁荔 Comparison of Procedures to Evaluate the Cyclic Stress-Strain Curve from Incremental Step Test 367 Table 3 Diflferences between the areas under the curves in Section 4.3 and attributable to the material behaviour. Table 4 presents the average and the extreme values of obtained using the same method on all specimens(MPa). the differences among the areas under the curves. To estimate a mean behaviour.it is possible to use the cyclic curve obtained by averaging arithmetically spedro憾 |_- the parameters calculated in the five tests.If the _ 650 450 / 一 3。。 / 一 : 0 7 一一 , 一 一。 0 0.0e §.0 0. §: StraiR amplitede Fig.5 Cyclic curves evaluated with different methods on sample 1. among the cyclic stress—strain curves calculated applying the different techniques on a single specimen. The situation is similar for all samples.As it can be observed,the differences increase with Ea.Moreover, method 2 always gives curves with the highest values, while the lowest values are obtained with method 1 a. Ifthe Euclidian norm ofthe residual vectors between the experimental data and the correspondent calculated values is considered,it is clear that,for the way used to obtain it,the best approximation is associated to method 2. A way to enhance the differences is to evaluate the value of corresponding to,ga=0.02(Fig.6).The obtained values are significantly different,especially using the procedure l a.This is probably due to lower number of experimental values used in calculating the Young’s modulus(about 20 against 100 or more associated to procedures l b and l c),with an higher sensitivity of the result to the speciifc values acquired. The differences among the curves calculated using the four methodologies are comparable to htose described number of the recorded data for each test is the same. an alternative way to obtain a description of the mean material response is to gather all the experimental data acquired in only one set and to carry out hte following elaborations on this set.This 1ast procedure is preferable,since in this way the best interpretation of the whole set of acquired values is obt暮 ained.吾基dn_I-薹兰 j;0 0 0 0 Table 5 shows the values ofthe coemcients E,,z’,K’ evaluated both by arithmetic average and by the elaboration of hte complete data set.Fig.7 shows hte ratio between the mean material parameters evaluated with the two procedures;in this case the observed S e ?∞ 2 3 Spee 啦 Fig.6 Values of stress amplitude for ga equal to 0.02. Table 4 Diflferences between the areas under the curves obtained with the set of methods on a single specimen (MPa1. 0 rJ 368 Comparison of Procedures to Evaluate the Cyclic Stress-Strain Curve from Incremental Step Test Table 5 Average values of the coefifcients of the cyclic CUnre. 纛芝 鞠K 瓣五 I【 一= 0 Clll1 l|II I 毒==-I-筮一 {_ 、★ ,-. —, v 广i 1 、-; differences are minima1.The corresponding cyclic curves,calculated with the four procedures,do not deviate from the behaviour described above,even if they show some small variations in their relative position.This conclusion is also confirmed by the results of the analysis of the areas under the curves,as shown in Fig.8 and Table 6. 5.Conclusions Four different methOdOlOgies were reviewed and used to evaluate the parameters useful to analytically modelling the cyclic stress--strain curve.The various techniques were applied to data obtained from tests carried out on X22CrMoV 1 2.1 stainless steel specimens,and the resulting differences were analyzed and quantified.Young’S modulus evaluation has a 詈 lI.JI-.1二竺兽Ill署-f 嚣骞lI 。孽 : ;-^ 0 ,t t ^ 一 Method la Method lb Method le Method 2 Fig.8 Average values of the area under the cyclic curve. Table 6 Differenees between the areas under the mean curves evaluated with the set of methods(MPa). great influence on the obtained results:a correctly assessment of this parameter is necessary to reliable represent the behaviour of the materia1.To reach this goa1.the best method is the 1 C.The method 1 a is influenced by the number of cycles in elastic field present in a block and it could provide not enough accurate results if this number iS lOW.Method 1 b rfequently shows different results as the portion of the hysteresis loop considered changes. The four methodologies allow building cyclic curves which mainly differ in the plastic field.Applying these procedures to a single specimen the results are more dispersed than those obtained using a single method on a set of nominally identical samples.However,the differences are of the same order of magnitude,as quantified by the values of area under the curves.An average behaviour was evaluated considering the total set of data obtained from experimental tests.In order to assess the results.it iS important to consider the strain range of interest;it is clear that increasing the value of strain amplitude results in a greater scatter ofthe curves, because they were built from experimental data in which the maximum strain iS equal to 1%.In this field Comparison of Procedures to Evaluate the Cyclic Stress-Strain Curve from Incremental Step Test 369 of strain,no signiifcant differences among results were observed. The choice of the most suitable method is related to both the strain range of interest and the purpose for which the results will be used.The method 2 provides surely the best correspondence to the experimental data, but its use is only appropriate to describe the cyclic curve,because the values of the three parameters obtained should not be used separately.If,on the other hand,it is useful to have a single value of the modulus of elasticity to use also for other purpose,the use of methodology l c is more appropriate. References Stephens,R.I.,Fatemi,A.,Stephens,R.R.,and Fuchs,H. O.200 1.Metal Fatigue in Engineering.2nd edition.New York:John Wiley&Sons. [2] Milella.P.P.2O13. igue and Corrosion in Metals. Milan:Springer Verlang ltalia. [3] Mostaghel,N.,and Byrd,R.A.2002.“Inversion of Ramberg-Osgood Equation and Description of Hysteresis Loops.”Journal of Non-Linear Mechanics 37(8): 1319—35. [4】 Lopez,Z.,and Fatemi,A.2012.“A Method ofPredicting Cyclic Stress--Strain Curve from Tensile Properties for Steels.”Materials Science&Engineering A 556:540—50. [5】 Zhang,Z.,Li,J.,Sun,Q.,Qiao,Y.,and Li,C.2009.“Two Parameters Describing Cyclic HardenirIg/So ̄ening Behaviors of Metallic Materials.’’Journal of Materials Engineering andPerformance 1 8(3):237—44. [6】 Zhang,Z.,Qiao,Y.,Sun,Q.,Li,C.,and Li,J.2009. “Theoretica1 Estimation to the Cyclic Strength Coefficlent and the Cyclic Strain—Hardening Exponent for Metallic Materials:Preliminary Study.’’Journal of Materials Engineering and Performance 1 8(3):245—54. 【7】Tomasella,A.,El Dsoki,C.,Hanselka,H.,and Kaufmann, H.201 1.“A Computationa1 Estimation ofCyclic Material Properties Using Artiifcia1 Neural Networks.”Procedia Engineering 1 O:439-45. [8】 Ghajar,R.,Naserifar,N.,Sadati,H.,and Alizadeh,K.J. 20 1 1.“A Neural Network Approach for Predicting Steel Properties Characterizing Cyclic Ramberg-Osgood Equation.”Fatiuge&Fracture foEngineering Materials &Structures 34(7):534-44. [9】Dowling,N.E.1 993.Mechanical Behavior foMaterials. Englewood Cliffs:Prentice.Hal1. [10】Lukas,P.,and Klesnil,M.1973.“Cyclic Plastic Reaction and Fatigue Process in Metals.”Kovove Materialy l1: 340—57. 【1 1】Raman,S.G.S.,and Radhakrishnan,V.M.2002.“On Cyclic Stress—Strain Behaviour and Low Cycle Fatigue Life.”Materials and Design 23 f31:249-54. 【12】Mo ̄ow,J.D.1965.“Cyclic Plastic Energy and Fatigue of Metals In Internal Friction and Cyclic Plasticity,AsTM 7T尸3 78,edited by Lazan,B.J.Philadelphia:ASTM Internationa1. 【1 3】Landgraf,R.W.,Mo ̄ow,J.D.,and Endo,T.1 969. “Determination of the Cyclic Stress Strain Curve.” Journal ofMaterials 4 f11:176.88. 【14】Pickard,A.C.,and Knott,J.F.1987.“Effects ofTesting Method on Cyclic Hardening Behavior in Face-Centered—Cubic Alloys.’’In Low Cycle Fatiuge, ASTM STP 942,edited by Solomon,H.D.,Halofrd,G.R., Kaisand,L.R.,and Leis,B.N.Philadelphia:ASTM Internationa1. 【15】Schoeler,K.,and Christ,H.J.2001.“Influence of Prestraining on Cyclic Deformation Behaviour and Microstructure ofa Single-Phase Ni-Base Superalloy.”/nt. Journal ofFatigue 23(9):767-75. 【16】Hales,R.,Holdsworth,S.R.,O’Donnell,M.P.,Perrin,I. J.,and Skelton,R.P.2002.“A Code of Practice for the Determination of Cyclic Stress-Strain Data.”Materials at High Temperatures 19(4):165—85.