Materials and Struct 5, 327-334 Measurement of the fracture energy using three-point «bend tests: Part 3 ~ Influence of cutting the P-6 tail M. ELICES, G. V. GUINEA, J. PLANAS Departamento de Ciencia de Materiales, Bscuela de Ingenieros de Caminos, Universidad Politécnica, Ciulad Universitaria, 28040 Madrid, Spain “Available measures of the fracture energy Gy obtained with the procedure proposed by RILEM TC-50 provide valves that appear to change with sample size, calling ino question the possibilty of considering G as & material parameter. In previous papers several sources iy energy dissipation were analysed and it was concluded that, although important, they were thot enough to account for the measured size effect. Here, the diss of the test is analysed. It is shown that this energy ed energy at the very end proposal. ; ni model previously developed by the authors 1, INTRODUCTION ‘The specific fracture energy Gy has proved to be a useful parameter for design with concrete and cementitious materials [1] and for modelling the fracture behaviour 2,3]. Nevertheless, available inetease with specimen size ‘As already pointed out in a previous paper [8}, this apparent sizo-dependence of Gy calls into question the possibility of considering Gp as a material parameter. The only way to remove this doubt is to find some shortcomings in the determination of Gp that would explain this size effect as an artelact rather than an intrinsic feature of the material. {systematic search for possible sources of experimental errors was undertaken by the authors. In tvo previo papers [8,9] the role of the testing equipment, the experimental set-up and the energy dissipation in the “men bulk were analysed. In all cases, some energy equipment was less o% y dissipation, generated at regions of high tensile stresses, contributes at most 2%, The major contribution comes from the energy dissipation at the supports due to crushing and friction. This energy dissipation can ‘account, al most, for a 15% increase. Anyhow, putting all these values together gives a size-dependence which is not enough to account for the observed size dependence, about 50%, in our experiments ‘The final key to solve this problem seems to be tha in bending ~ the test cannot be controlled up to absolute 0025-5432/92 © RILEM. breakage of the specimen. Rather, the test must be stopped somewhere before that point and some energy, that corresponding to the unrecorded tail of the P-6 curve, is not accounted for in the measurements. A careful c of this analysis, itis found that the P-6 tail correction is consistent with the perturbed ligament model (PJM) previously developed by the authors on phenomeno- fogical grounds [10-12]. The PLM provides, then, a practical procedure to extract a ‘true’ Gp value from a set of apparently size-dependent Gr measurements. Tn the second part of this paper all the adjustments for Gy — those previously analysed [8,9] and that used in this paper ~ are applied to a set testing concrete beams of four si the energy correspond 2. THE P-6 TAIL IN A BENDING BEAM TEST ‘To estimate the energy dissipated at very low loads ~ when the load tends to zero in the descending branch of the P-d diagram ~ one needs to model the beam behaviour when the cohesive erack closely approaches the free surface. For cohesive materials and beams where weight is compensated, the last phase of a stabl328 Elices, Guinea and Planas conesive "ONE Fig. 1 (a) Rigid-body kinematics for the final stages of the bending te pieces which are connected only through the fracture zone, as sketched in Fig. 1. Also, itis assumed that Iways exists a compression zone at the top of the beam, which has however been proved to be asymptotically exact in the case of rectangular softening Dugdale model with stress cut-off) [14], ‘The crack opening w at a certain distance x from A, as shown in Fig. 1a, wil be given by vw = 2y sin(0/2) = Ox a Assuming a known softening function, the central bending moment may be written as w= [coon is=0 [awards e 0 acw)w di @ where M is the bending moment, o(w) is the softening function (as sketched in Fig. 1b), B is the beam thickness ‘and x, the point at which the softening is complete, hence W(x.) = wy. This result may be written as BG} = Sr Te 8) because the integral in Equation 2 is the first-order ‘moment of o(w) and can be expressed as the area enclosed between the positive axes and the softening eurve, G, times the abscissa, é, of the centroid of that area. Th distance can always be written as aGg/f,, where tensile softening -xponential softening «= Equation 3 may be rewritten as and abscissa of the centroid aGh M an“ ® ao? B & which indicates that in a plot of M/B versus 6? the trend for large @ values is that of a straight line of size-independent slope A. The asymptotic P-3 curve is obtained from the equations above by setting PS 46 maPS g4 5) 4 5 ee so that L Po BATS o . Jn a weight-compensated testing set-up, the zero load is, not known with great precision because the exe weight distribution is not known. If the beam were completely broken at test stop, this final reading would be, by construction, the "zero load” [10]. However, the beam approaches this pointed out, and the test stop point B must be taken as DEFLECTION . & Fig. 2 Neglected energy when the test is interrupted at poi B (point B is taken to correspond to zero load),Materials and Structures ‘Table I Concrete properties Proportional mixing, by weight Cement e Water iongth results at 28 days. the ‘practical zero’. The area that one can determine with bounded error is the area enclosed by the curve over the segment AD, which ia token as the best approximation of the fotal work of fracture WV. The uncounted work of fractures then the dashed area in Fig, 2, which we denote 45 Ala. (A more rigotous treatment of this approxima tion based on the use of the complementary work instead of the ordinary work may be found elsewhere [10]}) To estimate AM¥;,y we neglect the small triangular area OAA\, and letting dy be the deflection at the point of test stop, B, the shaded aren may be written, with the hel of the approximate P-6 relation (Equation 6), - BSA, BSA_BSA Wea = Pe oe rae Pa) a oy © and the total measured fracture work # will be W = GpBb— Ain = Geb A) where b is the initial beam ligament (b = D— a). This can be rewritten as Ww Gea» ale ) = G,Bib—bo) 0) 2G ede where SA 9 10) bo" 58, eed ‘According to the RILEM procedure [2] the measured fracture energy would be evaluated as w bo) inten = Be = Gr( 1 — >> G a ( a) ay while the true fracture energy is, from Equation 9, Ww Ga 12) B(b — bo) o When the rotation angle at which the testis stopped is the same forall the specimen sizes (whieh is the case for ‘our tests and may be very usual) the fraction S/d, is ‘constant and so is be. In such case Equation 12 coincides with that derived by the authors on phenomenological grounds in what was called the perturbed ligament model 329 Concrete strength (MPa) Compressive (PLM) [10-12]. The above equation may be read, then, as saying that premature interruption of the tests at some fixed rotation angle produces the same effect as having portion of ligament of length bo not contributing to the overall dissipation. bo coincides, then, with what ‘was called the perturbed length. The above results provide an alternative physical support for the PLM equation, ving, essentially, for the measurement of Gy [2]. Details of materials and experimental procedures have already been published [12] and therefore only the relevant aspects will be briefly summarized here. ‘The concrete was a standard RILEM concrete, as described by Dutron [15]. Natural rounded aggregates, classified as siliceous, were used. Table I summarizes the characteristics of the conerete mix and some concrete properties measured according to ASTM standards, The aggregate maximum size was 10 mm, fea specimens were notched beams, as sketched in Fig. 3: their dimensions are listed in’ Table 2. All specimens were cast in steel moulds and compacted with a vibrating \ ca = ts Fig. 3 Specimen dimensions L(mm) 6 (mm) Specimen (mm) FTt so 15 300100 FT2 100-250 33580100 FT3) 200500 67-580, 100 Fra 300750, 00, 800100330 Elices, Guinea and Planas tored under srvohydraulic ‘machine (Instr run in, mode. Loads were 25/50 KN Toad cell With a resolution ind 0.5% accuracy. MOD" was mea: ip-on gauge MTS 632.03C-51, with 0.2,4m resolution and +2 ym accuracy, Deflection was measured as the relative displacement of the central loading head and the line defined by the points on the upper surface af the specimen located on the verticals of the lower supports. The displacement was ‘measured by an extensometer located in a tranaverse hole in the loading head. The accuracy of the extensometer 4. RESULTS 41 Experimental values Tests were performed on gcometricaly similar notched beams of the same thickness but of different sizes, as according to the RILEM procedure [2] by dividing the measured work of fracture IV by the ligament area: (3) where the ligament b as already stated, is b =D —a, according to Fig. 3 ‘These Grrrexe results show a definite effect of the specimen size on the values of the fracture energy. When all possible sources of energy dissipation discussed in previous papers [8,9] and the effect of interrupting the {est at some fixed rotation angle considered here are taken, into account, an almost size-independent Gy emerges, as will be shown later on ke 3 Concrete fracture eneray Gp (RILEM) Sample Dimm) Ry(KN) fc (MPR) Geng (No! FTL 50 2.80f0.05) 431 37) FT2 100 5210.05) 4.15 75013] FTS 200 -9.37f006] 3.92 82 (2) FT¢ 30011297058] 3.08 94 {5} Pyis the maximum load and je = 1.52 S/B(D — a) (ee Fig, 3). 108) ~ it z 2 1 as 108 a S| 7 108 igh tot — o.0010 0.010040 0 =48 (rd) Fig. 4 Log-tog plot of the experimental M-@ curve showing the M oc 8? dependence at large rotations for all sizes 400 200 100 Ge etoF a) Fig. $ M-0¥ plot of the experimental M-0 curve showing the Mac 0°? dependence at large rotations for all the sizes of the asymptotic equation for the P-3 ,e seen, for jarge 0 all the plots merge into a straight line of slope 2, as expected, because Equation 4 may be written as tog Ms og A ~ 21080 aay ‘The value of A ean be computed by extrapolation in Fig. 4 or, simply by adjusting a straight line to the M versus 0? plots for large values of 0, as shown in Fig. 5.Materials and Structures 331 Table 4 Corrections for Gy in Nm! Source Sie t Size 2 Grstear S712] 75 [13], ~AGy (hysteresis) 00 02 = AG (botk) =05 (02) 1003) “AG (ateral supports) 340) 18.05) ‘AG (central support) —39(18) 4845) Gr, 1 oa Grr 26) Be G, (corrected) 7113) 85 (23) Size 3 Size 4 8212) 98 [5] 02 -05 15 04) -17 03) =18 0.7) 20(0.2) 68 (1.6) -72(14) m7 826 7@) 6m) Pin 89 (11) Values in square brackets indicate half-range, Values ‘Values in braces { } are estimated variation intervals. All these results support the suitability of the model for ‘our purposes. From the knowledge of A, Gy and Ji it is possible to estimate the value of « from Equations 4. For this concrete, where Ge one obtains a = 2.1 This value of «is rather larger than those that can be directly derived from the usual analytical expressions for the softening curve, such as the bilinear curve proposed by Petersson [13], the pure exponential curve or the modified exponential of Reinhardt et al. [16] which deliver values of z ranging from 1 to 1.3. The fundamental reason for this discrepancy seems to be that the analytical proposals fit very well the experimental results for small and medium erack openings while the fit for large ‘openingsis rather poor, the experimental results displaying a much longer tail than the analytical fits. This causes tle or no effect on the predicted results for small deflections of the beams (for usual laboratory sizes), because then only the initial portion of the softening curve comes into play. The divergence arises, however, when analysing a large deflection range as we do here. ‘That the discrepancy between analytical fits and experimental results may be large in the softening far end is obvious from the fits displayed in the above references [13,16] but it is clearer in the results of Rokugo et al [17] which display a critical crack opening about three times larger than the Petersson fit. From Fig, $ of the work of Rokugo er al. an estimate of the position of the centroid of the area enclosed by the experimental softening curve may be given, The centroid turns out to be around w = 0.084 mm, which in dimensionless terms corresponds to a ~ 1.9, value much closer to our result than to those of the analytical fits. This gives further ‘support fo our theoretical analysis, although further work is necessary to enhance the experimental support. Once 4 is known, the ‘perturbed ligament length’, bp, defined in Equation 10, can be evalwated from the value of the rotation angle 45,/S at which the tail of the experimental curve was cut. Using the average value from the test records, 45,/S = 0.012, an estimated perturbed length bo = 12mm is obtained, in agreement with the result delivered by the empirical procedure used previously parentheses indicate standard deviations, [12]. This consistency provides further support for the B $97 indicated that several sources of spurious energy dissipation exist which are not usually taken into account when computing the specific fracture energy Gp. In the present work, it hhas been found that a non-negligible amount of energy may be neglected when cutting the tail of the P-5 curve The next step is to analyse whether there is a significant improvement when all these corrections for the measured Gy-are taken into account. The best estimate of Gp can then be written as Ge = Gentes ~ LAGE + AGerai, (15) where the sum is extended to all extra dissipation sources. Tn Part [ of this work [8] it was concluded that some ‘energy dissipation due to hysteresis ofthe testing machine may occur and that this was specimen size-dependent The energy dissipated for each beam size may be evaluated from Fig. 6 (Fig, 2a from [8}) with the knowledge of the average maximum: load for each beam size reported in Table 3. The values of dissipated energy appear in Table 4, os oo 0 os 02 4G, (N/m) ob pacertessereseeetc So 100 150 200 250 300 350 BEAM DEPTH (mm) Fig, 6 Correction due to hysteresis in the measuring system (measured in [8)).332, Elices, Guinea and Planas 20 AG, (N/m) os oof 1 ee 06> 100 150 200 250 000 950 BEAM DEPTH (mm) Fig. 7 Correction due to ik dissipation (computed in £97). In Part 2 [9] it was shown that, apart from the surface energy, some energy was dissipated in the bulk of the material as the crack propagates. This bulk energy is dissipated al regions of high tensile stresses and the result, asa function of the beam depth, is shown in Fig. 7 (Fig. 8a from [9}). This dissipated energy is also reported in Table 4. Supports are the major source of energy dissipation, Dissipation in the rolling supports, due to friction and to crushing, was discussed and measured in Patt 1 [8] For our tests, only fietion must be accounted for, because the displaccment measurement already exchided the crushing component, Dissipation at the central support, due (o crushing, was considered and measured in Part 2 {9]. The result of both sources of spurious energy waste is represented in Fig. 8, Again, from the knowledge of the maximum loadsfor the different beam sizes, the values ed energy at the supports were computed. To ‘emphasize the different contributions from the lateral and central supports, both results appear in Table 4. When ail these sources of energy dissipation ~ not essential for fracturing — are taken into account and of dissi 12 10 eo38 Zz 6 e a 4 2 50 100 160 200 250 900 060 BEAM DEPTH (mm) Fig. 8 Correction due to dissipation at the supports (measured in [8] and [9)). deducted from the measured energy, an intermediate cstimate of Gr Gps, is oblained. Such results. also appear in Table 4 and, unfortunately, still exhibit a non-negligible size effect (about 60% for the sizes considered), As is shown in this paper, there is some amount of energy consumed in breaking the beam that has not ‘been accounted for when testing is prematurely stopped. This energy is, from Equation 7, Wy = PSA 26, a6) and depends only on the rotation angle 45,/8 at which the tail of the P-3 curve is cut, apart obviously from fixed material parameters, When all tests are stopped at the same rotation angle, this energy will be size- independent, and when divided by the specimen ligament the specific fracture energy correction will decrease with specimen size, The value of AT;,; for each concrete beam size was ‘obtained in the following way. Thickness B, loading span S and the deflection at which tests were cut, 3,, were computed as the average values forthe specimens of that size, A was obtained as the average slope of the M-0-? ‘curves forall the sizes shown in Fig. 5. Finally, the values ‘of the correction terms AGrx,y Were computed as AWua Geran o bB an The values of this correction are shown, for each size, in Table 4, together with the final corrected valucs of Gp, which appear to have a much milder size dependence, if any at all, To illustrate this, such results are plotted in Fig. 9 as a function of specimen size, As can be casily almost size-independent. ‘The average ind no definite trend may be guessed scatter band which comes from the we scatter of all the measurements and corrections. 120) 100 20) 0} G, (N/m) 20] 0 50 100 150 200 260 300 80is and Structures per, and two previous ones [89], was to ascertain possible sources of energy dissipation in addition to that essential for the fracture process which are not taken into account when dleter- mining Gp following the RILEM recommendation based on the work of fracture, Several sources were detected and the corresponding dissipated energy was 1. In this paper, attention was focused on the dissipated energy, often negiected, in weight compensated tests at the very end of the test. In practice, tests are stopped at a coriain value of the displacement, or bonding angle, and the remaining dissipated energy neglected. Here it was shown that this energy cannot be neglected for small specimens if the tests are interrupted at « reasonably low rotation. 2. When this energy is taken into account, the final values of Gp appear to be. izesindependent for the experimental resus. of the guihors TMB Tel supports considering Gy as a material parameter for design purposes and provides further confidence in modelling concrete and rocks as cohesive materials. 3. The energy enclosed in the P-d tail justifies. and provides a physical explanation for the perturbed ligament model, previously developed by the authors ‘on phenomenological grounds. The work of fracture obtained when this energy is neglected tums out to be equal to the work of fracture one would obtain if a por- tion of the ligament ~ the perturbed length — were destroyed (or perturbed) prior to the start of the test ACKNOWLEDGEMENTS ‘The authors gratefully acknowledge financial support for this research provided by the Comisién Interministerial de Ciencia y Tecnologia CICYT, Spain, under grant PBR6-0494, “and by the Polytechnical University of Madrid under grant A-91 0020 02-31 REFERENCES 1 Hilllerborg, A., ‘The theoretical basis of a method to determine the fracture energy Gy of concrete’, Mater ‘Struct, 18(106) (1985) 291-296, 2, RILEM TC.50 FMC (Draft Recommendation), ‘Determi- nation of the fracture energy of mortar and conerete by ‘means of three-point bend tests on notched beams’, ibid 18(106) (1985) 285-290, 333 Elices, M. and Planas, J, ‘Material models’, in ‘Fracture Mechanics of Concrete Structures" (Chapman & Hall, London, 1989) pp. 16-6 Idem, "Size effect and experimental validation of available fracture models’, in “Analysis of Concrete Structures by Fracture Mechanics’ (Chapman & Hall, London, 1990 pp. 99-127), Hillerborg, A, “Results of thee comparative test series for determining the fracture energy Gy of concrete’ Mater. Struct. 18(107) (1985) 407-413, Wittmann, F. H. (Ed), ‘Fracture Toughness and Fracture Energy of Concrete’ (Elsevier, Amsterdam, 1986), Mihashi, H., Takahashi, H. and Wittman, F. H. (Eds), “Fractuce Toughnest and Fracture Energy: Test Methods for Conerete and Rock’ (Balkema, Rotterdam, 1989). nea, G. V., Planas, J. and Elices, M., “Measurement of the fracture energy using three-point bend tests: 1 — Influence of experimental procedures’, Mater. Struct 25(148) (1992) 212-218. Planas, J, Blices, M. and Guinea, G. VY, “Measurement of the fracture energy using three-point bend tests: 2— Influence of bulk exergy dissipation’ ibid 28(149) (1992) 305-312. Planas, 5, and Blices, M, *Conceptual and experimental problems in the determination of the fracture energy of concrete’, in ‘Fracture Toughness and Fracture nergy: Test Methods for Conerete and Rock’ (Balkema, Rotterdam, 1989) pp. 165-181 Planas, J, Maturana, P,, Guinea, G. V. and Elices, M, “Bractie energy of water saturated and partially dry conerete at room and at cryogenic temperatures’, in "Advances in Fracture Research’ (Pergamon, Oxford, 1989) pp. 1809-1817. Malurana, P., Planas, J, and Elices, M, "Evolution of fracture behaviour of saturated concrete in the low temperature range’, Engng Fract. Mech 38 (1990) 827-834, Petersson, P. E, ‘Crack Growth and Development of Fracture Zones in Plain Conerete and Similar Materials’, Lund Institute of Technology Report TVBM-1006 (ase). Panas, J. and Blices, M, G “The influence of specimen size ‘and material characteristic size on the applicability of elective crack models’, in "Fracture Processes. in Conerete, Rock and Ceramics’ (Chapman & Hall, London, 1991 pp. 375-385). Dutron, P,, "Mise au point Pune composition de béton de référence pour recherches et essais en laboratoire’, ‘Mater, Steuet, 139) (1974) 207-224. eae eet eee eet 1D. A, "Tensile tests and faire analysis of coneret Struet. Engng ASCE 112(11 (1986) 2462-2477. Rokugo, K., Iwasa, M., Suzuki, T,, and Koyanagi, W., “Testing methods to determine tensile strain soften ‘curve and fracture energy of conerete’, in ‘Fracture ‘Toughness and Fraciure Energy’ (Balkema, Rotterdam, 1989) pp. 153-163,334 Blices, Guinea and Planas RESUME, de energie de rupture par les essais de flexion si on retranche le segment Les mesures de énergie de rupture Gy obtenues selon la ‘néthode précontsée par la Commission Technique 50, dont ‘on dispose, fournissent des valeurs qui se trowvent changer avec Ia taille de Péprowette, ce qui met en question ta possibilité de consilérer Gp comme un paramétre du ‘matériau. Dans les articles: précédents, on a examiné plusieurs sources de dissipation de lénergie, et on a conclu ‘que, rout en ayant de Pimportance, elles we suffisent pas & expliquer Veffet d'échelle mesuré. Une solution semble résider dans le fait que, en flexion, Pessai ne peut ire controle jusqu’'d rupture complete de Héprowette. I! dott étre stoppé quelque part avant ce point, et la quantité d'énergie qui correspond au segment non enregistré de la courbe P~5 n'est pas prise en compte dans les mesures. Quand cette énergie est prise en compte, dans les résultats expérimentaux des auteurs, les valeurs, finales de Gy semblent presque indépendantes de échelle Ce résultat permet de considérer Gy comme un parametre du matérian a des fins de calcul, et constitue un encouragement pour la modélisation dia béron en tant que matériau eohévent. Leénergie comprise dans le segment Pd justifie une explication physique du concept phénoménolagique dit *Perturbed Ligament Model’ que tes auteurs ont précédemment élaboré,