Large‐scale experiments on concrete hinges under general loading

The structural behavior of concrete hinges under general loading is far from being properly understood, which is to a large extent due to the lack of pertinent experimental data. This paper contributes to filling this knowledge gap by presenting and discussing the results of an experimental campaign on one‐way Freyssinet concrete hinges. Seven large‐scale concrete hinges were tested in the Large Universal Shell Element Tester, which allowed the investigation of their behavior under general loading by all six stress resultants. A combination of digital image correlation and distributed fiber optic measurements allowed a deeper insight into the structural behavior of the specimens. The tested hinges could sustain very high axial stresses exceeding 4.5 times the uniaxial concrete compressive strength, large rotations of over 60 mrad, and shear stresses in transverse and longitudinal directions up to 2.24 times the axial compressive stress. The resistance to bending moments about the strong axis and torques also proved to be significant. A moderate amount of reinforcement crossing the hinge throat considerably increased the shear resistance at low axial stresses and produced a ductile shear behavior.


| INTRODUCTION
Concrete hinges are monolithic articulations in structural concrete. By suitably tapering the cross-section of a structural element, that is, providing a throat, its bending stiffness can be strongly reduced locally, resulting in the localization of curvature in the throat, which effectively acts as an imperfect hinge. Primarily applied in linear elements under high axial compressive loads, such as bridge piers and arches, concrete hinges are shaped to carry large axial forces and undergo large rotational movement while opposing little moment resistance. Nowadays, concrete hinges are mainly used to reduce restraint stresses in monolithic structures and to prevent the transfer of large bending moments to the foundations (thus reducing their costs). Over the past century, concrete hinges have been successfully applied in many structures and have often proven to be a superior alternative to mechanical bearings, especially in terms of maintenance costs, durability, robustness, and sustainability. 1,2 However, despite these advantages and the good experience, engineers are often reluctant to opt for concrete hinges. This is mainly due to a considerable degree of uncertainty in their design caused by the scarcity of experimental data 3 and the resulting lack of adequately substantiated design rules. Most experimental campaigns were conducted more than half a century ago, [4][5][6][7][8][9][10][11][12][13][14][15] and only a few more recent experimental investigations are available. [16][17][18][19][20][21] Moreover, these past experimental investigations clearly focused on the behavior under the standard load case, that is, axial compressive load and hinge rotations. Tests with more general loadingincluding shear forces, bending moments about the strong axis or torque-are thus scarce, 7,10,[16][17][18]21 which can be attributed to the complexity of test setups and load control, which are already demanding for standard loading.
However, pertinent experimental data is essential to understand the complex behavior of concrete hinges and develop suitable design models, eliminate the design uncertainties and ultimately foster the use of concrete hinges. To this end, the authors started a broad research project investigating analytically and experimentally the structural behavior of one-way (also referred to as linear) Freyssinet concrete hinges, which permit the rotation about one horizontal axis while being moment-resistant in the perpendicular direction. In contrast to Mesnager hinges, 4 Freyssinet hinges contain little to no reinforcement crossing the hinge throat and resist the loading mainly by the concrete of the throat. This paper focuses on the experimental part of the research project. Seven large-scale concrete hinges were tested under general loading in the Large Universal Shell Element Tester (LUSET) at the Structures Laboratory of ETH Zurich. 22 The test specimens were subjected to a multistage load history, and different failure types were targeted: failure due to axial compression, shear force in transverse and longitudinal directions, torque, or bending moment about the strong axis perpendicular to the throat axis. To the best of our knowledge, these are the first large-scale tests on concrete hinges under such a broad spectrum of loadings.
In the following sections, the experiential campaign is described in detail, the obtained results are presented and discussed, and the findings are finally summarized in the conclusions. The authors are currently establishing mechanical models for the observed behavior, which are compared to the experimental data and current design rules in a separate publication. 23 The experimental and theoretical outcomes resulting from this research project will not only improve the understanding and modeling of concrete hinges, but also of other related structural members such as longitudinal joints between segmental tunnel linings. 24

| EXPERIMENTAL CAMPAIGN
This section gives an overview of the experimental campaign, describing (i) the specimen geometry, materials and production; (ii) test setup and the loading sequence; Refers to the four layers above and below the throat. b Maximum applied action, failure did not occur. and (iii) the measurement systems. Table 1 summarizes the main parameters and results.

| Specimens, materials, and casting
In the frame of this study, seven large-scale, one-way Freyssinet concrete hinges were tested in the LUSET. The seven specimens, denominated CH1-CH7, simulated a part of a linear element (e.g., bridge pier) with a concrete hinge in the center. The specimen geometry and nomenclature are illustrated in Figure 1. The specimens measured 2 m in axial (x-) direction and had a rectangular cross-section measuring 0.38 m in transverse (y-) and 1.2 m in longitudinal (z-) direction. At mid-height, the cross-section of the specimen was tapered to form a oneway concrete hinge. The specimen parts above and below the hinge throat are referred to as (top and bottom) blocks or adjacent blocks. The hinge geometry, shown in Figure 1b,c, was chosen following Leonhardt's guidelines for concrete hinges. 2,25 Two throat geometries were investigated to explore the possibilities of the articulation, with a focus on narrow throats since these have been scarcely investigated in past experiments. Six of the specimens, CH1-CH6, had a very narrow hinge throat, d 1 = 0.1d = 38 mm wide and h t = 8 mm tall. Specimen CH7, on the other hand, had a wider (d 1 = 0.3d = 114 mm) and taller (h t = 20 mm) throat.
In order to prevent their premature failure before the hinge throat, the adjacent blocks were heavily reinforced, F I G U R E 2 Specimen reinforcement layout: (a) axonometry (only bottom block) and (b) top views of the reinforcement of the adjacent blocks; (c) axonometry of the reinforcement crossing the throat; (d) reinforcement list (centerline dimensions in mm). See Table 1 for additional information as shown in Figure 2a. Based on the findings of a previous research project on strip-loaded concrete blocks, 26,27 the confinement reinforcement was placed as close as possible to the hinge and sufficient longitudinal reinforcement was provided. Ten equally spaced horizontal reinforcement layers ensured the confinement and the transfer of splitting and shear forces in each adjacent block. Each reinforcement layer comprised several reinforcing ties, all with a diameter Ø16 mm. The horizontal reinforcement layers were supported by 12 vertical bars Ø30 mm. These bars constituted the flexural reinforcement of the adjacent blocks, which toward the specimen ends had to withstand significant bending moments (parallel and perpendicular to the throat axis) in the shear tests. In order to connect the specimens with the testing machine and introduce the loads (or impose deformations in displacement control), three load introduction blocks were placed each at the top and bottom of the specimen (Figure 1a). The blocks were connected to the flexural reinforcement by means of reinforcing bar couplers (BARTEC ® Type X18-24) fixed to the blocks with M24 high-strength bolts. These couplers and the shear keys on the load introduction blocks ensured a slip-free moment and force introduction. Except for CH3, virtually no reinforcement crossed the hinge throat: merely two plain bars Ø8 mm were placed vertically ( Figure 2c) to facilitate the extraction of the specimen from the testing machine in case of severe damage. On the other hand, Specimen CH3 was provided with five pairs of ribbed bars Ø12 mm positioned in an x-shape across the throat (as common for Mesnager hinges 4 ), see Figure 2c, corresponding to 2.8% reinforcement content referred to the throat area (or 0.25% refereed to the block cross-section). All reinforcing bars, except for the bars crossing the throat, were of Class B500B. The steel properties are given in Appendix A.
The specimens were produced in collaboration with the company DSE systems at their prefabrication plant. Although vertical casting would have been more representative of typical concrete hinges in practice, it was opted to cast the specimens in a horizontal orientation in order to facilitate the connection of the load introduction blocks to the flexural reinforcement before casting. The specimens were cast on a vibrating table in lubricated wooden formwork with 3D-milled, hard plastic inserts to shape the hinge. A vibration needle was also used to achieve good concrete compaction, particularly important in the throat region. The specimens were cast one at a time (as only one formwork was available). The top casting surface (specimen front) was carefully trowelled, and the specimens were covered with plastic sheets for at least 2 days before demoulding. Two steel plates were heated to roughly 60 C and bolted on the north and south sides of the demoulded specimens, resulting in a slight prestressing to stabilize the hinge during transport and installation in the testing machine, ensuring an initially uncracked hinge throat. The same concrete recipe with maximum aggregate size of 8 mm was used for all specimens. The measured uniaxial cylinder compressive strengths on the day of tensing were around 40 MPa for all specimens except CH1, which had 33 MPa as it was tested at an earlier age, see Table 1. Further concrete properties (E-modulus, tensile strength, and cube compressive strength) are reported in Appendix A.

| Test setup
The LUSET 22 was used in a non-standard configuration for this test series, as the load introduction blocks at the specimen ends were connected to the middle three yokes of the top and bottom actuator groups of LUSET (Figure 3), without using the remaining actuator groups. By reconnecting the 30 hydraulic actuators used with the 20 control channels available, all six stress resultants (Section 2.3) could be introduced and controlled at the top and bottom ends of the specimens independently, with some limitations merely for the bending moment M y about the strong axis, which could be applied only in symmetric or antisymmetric configurations.

| Loading
As concrete hinges are mainly applied in linear elements, six stress resultants (internal actions) generally act on them. In the global Cartesian coordinate system defined in Figure 4, these stress resultants are: axial (normal) force N x (defined positive in compression), transverse and longitudinal shear forces V y and V z , respectively, torque M x , hinge resisting moment M z (about the longitudinal axis), moment M y about the strong axis (perpendicular to the hinge rotation axis). In the case of one-way concrete hinges, it is convenient to refer to the rotation r z about the longitudinal axis. For convenience, the term action is also used for this rotation, despite that it is actually an action effect. Depending on the type of structure and the loading situation, concrete hinges can experience arbitrary combinations of the six stress resultants mentioned above, noting that the bending moments and shear forces need to satisfy equilibrium, that is, dM z /dx = ÀV y and dM y /dx = V z . In bridge structures, which are the main application field of one-way concrete hinges, axial compressive forces N x and rotation r z are the predominant F I G U R E 4 Actions investigated in this experimental campaign (drawn as positive) with corresponding state diagrams: Axial force N x , hinge resisting moment M z , and hinge rotation r z (about the longitudinal axis z), torque M x , transverse shear V y , longitudinal shear V z , moment M y about strong axis y perpendicular to the hinge rotation axis actions. However, due to traffic loads, lateral horizontal loads (e.g., wind, earthquake, and impact), and curved or skewed horizontal alignments, significant shear forces, moments, and torques can also be induced in the hinges. This was the reason to subject the specimens to all six stress resultants, as shown in Figure 4. However, to make the results easier to interpret and within the capabilities of the LUSET, the combinations were limited to meaningful combinations of up to three simultaneously applied stress resultants and excluding significant axial tensile forces. The actions were applied on the top and bottom ends of the specimens and are indicated by the subscripts top and bot, respectively. The resulting actions in the hinge at the specimen center are referred to without any subscript. In the case of the rotation r z , equal rotations ( Figure 4b) were applied at the top and bottom ends of the specimen, r z,top = r z,bottom in cases without transverse shear (V y = 0), resulting in a relative rotation angle r z ≈ r z,top + r z,bottom between the two specimen halves since the curvatures localize in the hinge. In the specimens with V y ≠ 0, opposite rotations corresponding to the integral of curvatures due to M z in the top and bottom block were superimposed to the specimen end rotations (without affecting r z due to the antisymmetry with respect to the hinge). The exact rotation angle r z of the hinge was computed during the post-processing based on the measurements from the front and back digital image correlation (DIC) systems (Section 2.4).
The actions were applied quasi-statically with rates of 2 kN/s for the forces, 1 kNm/s for M z and M x , and 0.002 mrad/s for r z . As the applied actions refer to the global coordinate system, the applied actions included components in other directions as the specimen deformed. These second-order effects were relatively small and can be neglected in most of the analyses reported in this paper, with the exception of the hinge resisting moment analyses in Section 3.1.3. Figure 5a-g show the multi-stage load history of each test. Initially, all specimens were subjected to axial compression and several rotation cycles to mimic the typical situation in real concrete hinges at the time of application of significant additional actions. As it not feasible to apply a corresponding, realistically large number of small rotation cycles with durations of days (daily cycles) and months or even years (seasonal cycles)-allowing for creep and relaxation effects-without disproportionate effort, it was opted to instead perform a small number of large rotation cycles, similar as in the experiments of Base. 10 These large rotations were applied over the period of minutes, and presumably resulted in a more severe initial damage than to be expected in real structures. 8 Subsequently, the specimens were loaded with different action combinations targeting various failure types. Table 1 summarizes the occurred failure types, which will be discussed in detail in Section 3.

| Measurement systems
The LUSET is equipped with various sensors (including force and displacement sensors on each actuator, oil pressure sensors per pressure line, and inclination sensors per yoke) that precisely determine the imposed actions and deformations. 22 A Lagrangian optimization on the logged sensor data was used to reduce possible noise. 28 In addition, several independent optical measurement systems were applied to monitor the strains and deformations of the specimens, see Figure 6. Two 3-dimensional DIC systems monitored the deformation of the speckled front and back surfaces of the specimens. An additional 3D DIC system monitored the south end of the throat (z = À525 mm), see Figure 6b. Due to the very challenging conditions-that is, very small and curved surface, difficult to speckle and lit and prone to flaking-valuable data from this throat DIC system could be obtained only in the first load stages of a few tests. Distributed fiber optic (FO) sensing was applied to measure the axial strain distribution along the specimen height. In all specimens with unreinforced throats, six fiber optic sensors were placed vertically at different z-coordinates along the hinge axis (y = 0), as shown in Figure 6a. The FO sensors used (BRUsens DSS 3.2 mm V9 grip) are very robust and provided with a structured sheath for better bond transfer with the surrounding material. The sensors were positioned by fixing their ends to the reinforcement cages, and springs were mounted on one end of the sensors to slightly tension them and ensure their straightness before casting. During the concrete pouring, care was taken to minimize the impact of the concrete flow on the sensors. More details on the DIC and FO measurement system are provided in Appendix B, including the post-processing procedures and the achieved measurement accuracies.
During the experiments, SLR cameras with long-focus lens were used to photograph details such as cracks and flaking in the throat. For Specimen CH5, the examination of these photos after the test allows estimating the length and longitudinal crack opening in the throat (Section 3.2).

| PRESENTATION AND DISCUSSION OF TEST RESULTS
This section presents and discusses the experimental results, starting with the observed structural behavior under axial compression and transverse rotation, followed by bending moment about the strong axis, longitudinal and transverse shear forces, and finally torque.

| Axial compression N x and rotation r z
Axial compression N x and transverse rotation r z is the predominant load combination that every concrete hinge in practice will typically undergo. The following subsections describe the general behavior under axial compression and rotation, discuss the axial strain distribution along the specimens, and analyze the rotation resisting moment.

| General observations on load-bearing and cracking behavior
The top and bottom half of a one-way concrete hinge under axial compression essentially corresponds to a reinforced concrete block subjected to strip loading. 29 The deviation of the compressive stress trajectories through the hinge throat causes transverse splitting stresses-often referred to as bursting stresses-in the top and bottom  blocks and compressive confining stresses in the throat region. These confining stresses, together with deformation constraints by the top and bottom blocks, cause a triaxial compressive stress state in the throat region, which can consequently sustain axial stress several times greater than the uniaxial compressive strength. 29 In the experiments, the first fine splitting cracks on the north and south faces were visually detected at N x ≈ 4.5 MN in Specimen CH4 with the narrow throat, and N x ≈ 5.3 MN in Specimen CH7 with the wide throat. As the axial load increased, the opening of these splitting cracks also increased, and new cracks formed. At high axial loads, vertical cracks also appeared on the longitudinal (front and back) surfaces of the specimen, indicating an engagement of the longitudinal confining reinforcement. Generally, no excessive distress was observed in the adjacent blocks, indicating that the adopted reinforcement amount and layout were appropriate and efficient. This confirms the findings of a previous research project on partially loaded blocks 26,27 on the importance of placing sufficient longitudinal and transverse confinement reinforcement immediately above and below the loaded surface, corresponding to the throat in concrete hinges. In all specimens, flaking (i.e., superficial spalling of the cement skin outside the coarse aggregates) of the throat fillets was observed at a relatively early stage of the tests, after the axial load had been applied and a few rotation cycles were performed. This flaking is presumably due to the high strains in the throat and the different stiffness of concrete aggregates and cement matrix. 13 The resulting stress concentrations cause the thin cement layer to flake off, making the concrete aggregates visible, see Figure 7. Such flaking can likely be reduced by choosing a smaller throat height (h t ), fillet radii, and maximum aggregate size, but is most probably unavoidable; it was also observed in other experiments on concrete hinges (e.g., 10,13,14 ). No noticeable influence of this flaking phenomenon on the load-bearing behavior of the hinge was observed neither in this nor in past experimental campaigns. Superficial throat flaking is, therefore, an aesthetic issue at most. 30 A hinge rotation r z causes the axial stresses to increase on one side of the throat and decrease on the opposite side. With increasing rotation, a horizontal bending crack penetrates the throat starting from the tensile side; consequently, the throat width effectively carrying the axial load decreases, and the compressive stresses increase to values that can exceed by several times the uniaxial concrete strength. Because of the very challenging geometric conditions and the cement skin flaking, it was challenging to quantify the crack development in the throat. For this reason, primarily qualitative observations are possible for most tests. Reliable quantitative data could be obtained mainly from Specimen CH7, which had a wider and higher throat enabling to track the crack development with the local DIC system for several rotation cycles. Figure 8a reports the length of the throat bending crack at the south end of CH7 during four rotation cycles at two magnitudes of N x . Already at relatively small rotations, the bending crack started propagating across the throat; note that crack lengths at small rotations are missing since the short cracks forming on the edges of the throat could not be detected by the DIC system. Figure 8b,c show the throat at two time instances with the overlaid axial strain field (ε x ), which allows detecting the cracks. At both instances, the magnitudes of the axial force (N x = 1 MN) and rotation (jr z j ¼ 4:2 mrad) were the same, but the hinge was rotated in the opposite direction. In Figure 8b, the crack propagates from the back side of the throat and extends across more than half the throat width, with a projected length on the y-axis of roughly 70 mm. As the rotation was decreased, the crack progressively closed, and its visible length reduced. At a rotation of r z = À4.2 mrad (Figure 8c), a crack propagating from the front side of the throat also extended beyond the throat centerline, with its tip apparently coinciding with the tip of the crack visible in Figure 8b. The throat is therefore cracked side-toside with the left part of the crack being completely closed and not discernible at this stage. The results indicate that the crack grows with increasing r z and decreasing N x , and closes with decreasing r z and increasing N x . The length and opening of the bending crack induced by a rotation r z likely depend also on the material properties and the hinge geometry: subjected to similar N x and r z , Specimen CH7 having a wider throat appeared to exhibit longer and wider cracks than the specimens with a narrow throat. Note that time dependent effects, such as plastic deformations and creep and relaxation processes in the highly stressed throat region of concrete hinges, as investigated, for example, by, 10,13,20 are not the focus of this experimental campaign, and will merely be discussed briefly at appropriate locations in the following sections.
Similar bending cracks as in Specimen CH7 formed in the throat of all specimens during the initial test phase (consisting of cyclic rotations applied at constant axial load), typically accompanied by fine splitting cracks in the adjacent blocks and flaking of fine cement chips along the throat fillets. As confirmed by the response of all specimens under the subsequently applied loading, these local effects must not be regarded as distress signs marking the beginning of the hinge failure process, but as characteristics of the load bearing behavior of a concrete hinge under moderate (service) loads. These observations essentially confirm those found in. 10,13,30

| Axial strains and displacements
Locally narrowing the section in the hinge throat not only generates an imperfect hinge, but also strongly reduces the axial stiffness. The resulting complex deformation states are discussed in this subsection on the basis of the measurement obtained with the DIC and FO systems, focusing on the axial strains. Figure 9 summarizes the axial strains and vertical deformations observed in Specimens CH2 and CH7 at several loading stages, using three different measuring systems. The blue lines show the data obtained with the six fiber optic sensors placed vertically along the specimen axis (y = 0), plotting the individual FO sensor readings and their mean value in light and dark blue, respectively. The green lines show the measurements obtained with the two DIC systems monitoring the front and back surfaces of the specimen (light green) and their average (dark green). The monitored surfaces are divided into several horizontal strips, and for each strip, the strains ε x in the throat than CH2, see Figure 8b (e.g., compare ε x ≈ 7.7 mm/m at N x = 9 MN in CH7 to ε x ≈ 3.8 mm/m at Nx = 3 MN in CH 2 having a three times narrower throat, hence equal stress at a third of the load). On the one hand, this can be attributed to the wider throat of CH7 resulting in lower confinement by the deviation of compressive stress trajectories 31 and consequently in a weaker and softer response of the confined concrete in the throat region. On the other hand, the higher strains in CH7 were also caused by the plastic deformations that occurred during the many rotation cycles applied to CH7, as further outlined at the end of this subsection.
The short, bold red lines in Figure 9c show the axial strains of the throat measured with the local DIC system at the south end of the throat. The strains are computed by placing three virtual strain gauges across the throat (Figure 8a). Note that these short red lines refer to the top (red) horizontal axis, whose scale is two orders of magnitude smaller than the bottom (green) axis. Extremely high strains were measured: up to 120 mm/ m = 12% for CH7, that is, roughly 40…60 times the commonly assumed short-term crushing strain of unconfined concrete, that is, 2…3 mm/m. These throat strains are significantly higher than those measured by the FO sensors. This difference could basically be attributed to several reasons: (i) the longitudinal (north and south) ends of the throat exhibited much higher strains than the central part (indeed, similarly higher strains at the longitudinal ends were also obtained in the non-linear finite element analyses performed in 32 ); (ii) the FO sensors may have smeared the actual strain distribution because of internal slip between the different layers of the sensor (also observed in 33 with the same optical sensors) or external slip between the sensor and the surrounding concrete; (iii) the extremely high strains measured by the local DIC system were merely an artifact of the flaking on the throat fillets. While this last reason appears rather improbable (the throat DIC system measured much higher strains than the FO sensor already at very low axial loads before any sign of flaking was observed), the actual reason is likely a combination of the first two points, with (ii) playing an increasing role as the magnitude of N x increased. Figure 9d shows, in the upper part per specimen, the loading history during the first part of the respective experiments, and in the lower part the axial shortening Δh measured over a 1.2 m high part of the specimen (Figure 6a) obtained with all measurement systems discussed above. While the DIC displacements are obtained directly from the systems, the FO displacements are obtained by integrating the strains with the xaxis as origin. There is a very good agreement between the specimen shortening Δh obtained with the front and back DIC measurement and that from the FO sensors. The throat DIC system yielded only slightly smaller values, indicating that at least in the longitudinal throat ends, most of the axial shortening actually occurred within the throat height. While for small axial loads the specimen shortening Δh increased fairly linearly with N x , Δh increased over-proportionally at higher loads. Moreover, Δh kept increasing also during the rotation cycles, with a gradually more pronounced increase at cycles carried out under higher the axial load, which will be further outlined in the following subsection. Most of the axial deformations occurring at high axial load were irreversible, as indicated by the shortening of Δh = 3.4 mm after reducing the load in CH7 from 9 MN to 1 MN (at t ≈ 212 min), which was significantly higher than Δh = 0.12 mm measured in the initial loading phase upon reaching N x = 1 MN (at t ≈ 13 min). The pronounced time dependency of the deformations was already observed during the short loading pause (i.e. r z and N x kept roughly constant) of CH7 at t = 153 min, where the specimen shortening Δh increased by 0.15 mm while the axial load was kept roughly constant at N x = 9 MN for 5 min. On the other hand, no increase of Δh could be observed during the loading pauses after the unloading of CH7 to N x = 8 MN (t = 192 min, paused for 5 min) nor after N x was unloaded to 1 MN (t = 240 min, paused for 15 min), indicating that creep recovery after unloading was very limited.
While due to the mentioned effects, the local strain profile across the hinge centre plane cannot be determined with sufficient certainty despite that several refined measurement systems were used, the observations yield valuable insight into the complex threedimensional strain state of the concrete in the throat region.

| Hinge resisting moment M z
Concrete hinges are imperfect hinges that oppose the rotation with resisting moments, whose magnitude should be minimized and accounted for in the design of a structure unless negligible. Imperfect hinges either oppose the rotation by friction (thereby generating an eccentricity of the axial force) or present an eccentricity by geometry. In Freyssinet hinges, both effects are combined, with the hinge resisting moment increasing as the rotation increases and moves the line of action of the axial force away from the x-axis towards the throat edge. Figures 10 and 11 show various hinge resisting moment M z -rotation r z curves for specimens with wide ( Figure 10) and narrow (Figure 11) throat. These curves are analogous to moment -curvature diagrams commonly used for flexural concrete members, yet considering rotations, that is, curvatures integrated over a specific length. Appendix C outlines how the hinge resisting moment was computed based on the available measurements. The nonlinear, progressively flattening shape of the moment -rotation curve is partly due to the progressive macro-cracking of one side of the throat. However, this cannot explain the hysteretic behavior during unloading (i.e., rotation reduction): for perfectly elastic material behavior, with limited tensile strength, a flattening M zr z curve would be obtained, but the unloading branch would coincide with the loading one. Such a nonhysteretic behavior, without irreversible deformations, can at most be expected at very low axial forces and rotations, when the concrete in the throat compression zone still behaves roughly linear-elastically, as observed, for example, in Test B1 of 34 for low axial forces and rotations, and partly in Specimen CH7 for the small rotation cycle at N x = 1 MN (gray curve in left plot of Figure 10). As the axial force and the rotation increase, the compressive stresses in the throat increase and enter the flattening part of the stress-strain relationship of confined concrete in compression, where large inelastic strains occur even under short-term loading, accompanied by gradually decreasing stress increments. This results in M zr z curves with wide hysteresis loops, as illustrated in Figures 10 and 11, and also observed in 34 at higher loads and rotations. The inelastic deformations also cause the axial shortening of the throat during rotation cycles observed in the previous subsection. Hence, the hinge resisting moment and axial shortening depend on the load path, that is, the loading history. Generally, with increasing N x and r z , the hinge resisting moment M z increased and the hysteresis became wider, with M z appearing to asymptotically approach a limit value at large rotations. This limit value is the moment resistance of the hinge, and-analogously to the ultimate bending moment of a concrete compression member at a specific axial force (M-N interaction)depends on the throat width, the confined concrete strength, and the axial force. Comparing columns (a) and (b) in Figure 11, it can be concluded that a moderate (A s / (b 1 d 1 ) < 3%) reinforcement crossing the throat only marginally increased M z in the hinges with a narrow throat. Such increase is expected to be more significant in hinges with a wider throat due to the larger depth of the reinforcement, yet likely not significantly higher at rotations relevant for practical applications with appropriate detailing of the throat reinforcement, that is, reinforcing bars passing through the throat centerline. The plots in Figure 11c show that a simultaneously acting moment M y about the strong axis clearly caused a softer hinge response and reduced the hinge moment M z resisting the rotation r z . This was also observed by 30 when discussing the test results of 7 The three times wider throat of CH7 ( Figure 10) transferred significantly higher moments M z than the narrow throats of the other specimens ( Figure 11), roughly triple at equal N x and r z . The maximum moments measured in Specimen CH7 were significantly smaller than the moment resistance of an unreinforced cross-section of the blocks of roughly 180 and 820 kNm for N x = 1 and 9 MN, respectively. Nevertheless, they may not be negligible in design, contrary to the much smaller moments in the specimens with a narrow throat.

| Ultimate axial force and rotation
Owing to the beneficial effect of confinement, average axial compressive stresses several times higher than the uniaxial concrete strength could be reached in the specimen throats without causing detrimental concrete crushing, even at considerable rotations. As already mentioned F I G U R E 1 0 Hinge resisting moment M z -rotation r z curves for specimen CH7 with a wide throat (d 1 /d = 0.3) at various axial loads N x in Section 3.1.1, flaking (fine superficial spalling) of the throat fillets occurred at axial loads and rotations that are significantly lower than the ultimate resistances are no sign of distress compromising the structural integrity of the hinge.
The specimens with narrow throat CH1, CH2, and CH4 underwent the largest rotations of this experimental campaign, which amounted to r z ≈ 60 mrad, by far exceeding concrete hinge rotations expected in practical applications. The concomitant axial force was 1.5 MN for CH4 (corresponding to a nominal average axial stress across the throat cross-sectional area of σ x = 38 MPa ≈ 0.9f c0 ) and 3 MN for CH1 and CH2 (σ x = 75 MPa ≈ 2 f c0 ). Whereas CH4 exhibited only limited flaking and inelastic deformations in the throat region at r z ≈ 60 mrad, the throat of CH1 and CH2 flaked significantly around its entire perimeter and underwent large inelastic deformations (Section 3.1.2). Nevertheless, CH1 and CH2 kept carrying the axial load without signs of imminent failure. In all three tests, the M zr z curves had significantly flattened at r z ≈ 60 mrad. The test results appear to indicate that for adequately detailed hinges, that is, hinges with narrow and short throats, failure due to excessive rotation is unlikely to occur in practice.
In Specimen CH1, the maximum rotation of r z = 60 mrad was decreased to r z = 23 mrad and subsequently kept constant while N x was gradually increased. The throat concrete showed signs of gradually increasing distress and crushing. At N x ≈ 6 MN (corresponding to σ x = 150 MN ≈ 4.5 f c0 ), the throat was no longer visible, that is, the two adjacent blocks entered in direct contact beyond the throat. At this stage, large cracks and outward bulging were observed in the shoulders of the adjacent blocks immediately above and below the throat contraction, pointing at substantial spalling of the concrete cover. The axial load was further increased up to N x ≈ 8.2 MN and subsequently the rotation was reduced to zero, whereby the two specimen halves rotated around the edge of their contact surface in a kind of rocking motion. At this point, the test was stopped, because further damage would have severely complicated the extraction of the specimen from the testing machine while bringing little additional scientific insight. Presumably, the axial force N x could have been increased significantly more, along with a progressive increase of the contact area between the two blocks. After the experiments, the specimen was reloaded to test drive the shear loading protocol of the following tests. CH1 could sustain a shear force V y = 500 kN while subjected to an axial load of merely N x = 282 kN. This is very similar to the shear capacity of CH2, whose throat, however, was not completely crushed as in CH1 (Section 3.3). These observations indicate that concrete hinges are very robust as long as they are subjected to sufficiently high axial compression.
Specimen CH7, with a wide throat, was subjected to a maximum axial load of N x = 9 MN, corresponding to an average axial stress σ x = 75 MPa ≈ 1.9 f c0 . At this axial load, the specimen was subjected to rotation cycles up to r z = 10 mrad. Besides extended flaking of the throat fillets and inelastic deformations in the throat (Section 3.1.2), the specimen showed no particular signs of distress. The axial load was partly removed, and the specimen was tested in transverse shear (Section 3.3).

| Moment M y about the strong axis
The behavior of Freyssinet concrete hinges to bending about the strong axis (i.e., the transverse y-axis) was investigated with Specimen CH5. The load history applied to CH5 is shown in Figure 5e. In the first phase of the test, the specimen was loaded with N x = 1.5 MN and subjected to rotation cycles with r z up to 15 mrad at three stages of moment M y , that is, 0, 263, and 525 kNm (corresponding to eccentricities of the axial force of 0, b 1 /6, and b 1 /3, where b 1 = 1050 mm = throat length). Afterwards, N x was increased to 3 MN, and the moment M y was increased in several stages, where at each stage r z rotation cycles were performed.
In terms of bending moments, the behavior in the longitudinal direction is similar to the transverse direction: in the unreinforced throat, the bending resistance is due to an eccentricity of the axial force. As the throat is much longer than wide, the bending resistance is also significantly higher in the longitudinal direction. Nevertheless, the throat resistance and bending stiffness are significantly lower than that of the adjacent blocks and hence, the curvatures also localize in the throat region and can be adequately described with a rotation angle r y between the top and bottom block. Figure 12a shows the moment M y -rotation r y curves at two axial force levels, N x = 1.5 and 3 MN. Similarly to the moment M z -rotation r z curves for the transverse direction (Figure 11), the M yr y curves for the longitudinal direction also exhibited an initial, stiff, practically linear phase, followed by a progressive flattening indicating a loss of stiffness of the hinge. The end of the linear phase roughly coincided with the theoretical decompression moment M y, dec = N x Áb 1 /6 of the throat for a perfectly linear elastic behavior of concrete in compression (i.e., M y, dec = 263 kNm for N x = 1.5 MN and M y, dec = 525 kNm for N x = 3 MN), when a longitudinal crack originated at the tensile (north) end of the throat. With increasing moment M y , this longitudinal crack, spanning across the full throat width d 1 , progressively propagated in the longitudinal direction. The length of the longitudinal crack, manually extracted from pictures taken by SLR cameras (Section 2.4), is plotted in Figure 12b against the moment M y . During the r z rotation cycles at constant M y , inelastic deformations occurred in the compressive zone of the throat, causing an increase in the longitudinal rotation r y . The opening of the longitudinal crack increased while its length remained roughly constant or increased only slightly. As in the transverse direction, see Section 3.1.3, the inelastic deformations on the compressive throat side caused residual longitudinal rotations upon reducing M y .
At N x = 1.5 MN and M y = N x Áb 1 /3 = 525 kNm, that is, twice the decompression moment, the longitudinal crack extended for a length of about 550 mm (corresponding fairly well to the crack length of b 1 /2 expected for perfectly linear elastic behavior in compression), with an opening of roughly 1 mm at its north end. After a rotation r z = 15 mrad was imposed on the hinge, the longitudinal crack had roughly doubled its opening while roughly maintaining its length. At this stage, flaking at the front south corner region of the throat, where the axial compressive stresses due to the biaxial bending (M y and M z ) were highest, was clearly visible. Moreover, fine splitting cracks on the south face of the adjacent members were visible. The reduction of r z and subsequently of M y was accompanied by a practically complete closure of the cracks, with only a short hairline crack visible at the north end of the throat at t = 94 min. After this removal of bending moment and rotations, the axial force was increased from N x = 1.5 MN to N x = 3 MN and the moment M y was increased again in stages, including the theoretical decompression moment and twice its value. At M y ≈ 900 kNm, the south end of the throat had practically vanished under the high axial compressive stresses, that is, the two adjacent blocks entered direct contact beyond the throat. The average axial stress across the throat section in compression (38 mm wide, roughly 700 mm long) amounted to about 113 MPa, noting that the peak stress at the north end was presumably significantly higher. At twice the decompression moment, M y = 1055 kN (t = 194 min), the longitudinal crack in the throat extended over 550 mm similar as for N x = 1.5 MN, with a maximum opening of 7 mm; the length in longitudinal direction over which the throat had vanished amounted to roughly 200 mm, and the cover concrete on the south faces of the adjacent blocks had spalled over a large area. Despite this visible damage, the subsequent r z rotation cycle could be imposed without problems; as in the previous rotation cycles, it caused a further increase of the crack opening. Finally, the moment M y was increased to 1230 kNm. At this point, the longitudinal throat crack extended over 600 mm and had opened roughly 40 mm at its north end, see Figure 12c. On the compressive side, the hinge throat had vanished over a length of roughly 250 mm. It was refrained from a further increase of M y , because further damage would have severely complicated the extraction of the specimen from the testing machine while bringing little additional scientific insight.

| Shear loading
Concrete hinges often carry shear forces parallel and perpendicular to their rotation axis. However, current design recommendations for the shear resistance of Freyssinet concrete hinges are semi-empirical, lack sufficient experimental validation and are presumably overly conservative [xx], making the shear resistance a limiting factor in design. This was the motivation to investigate the shear resistance in four of the seven tests conducted (CH2, CH3, CH6, and CH7), whereby several rotation cycles at different axial force levels were carried out beforehand in order to generate a cracked throat; these cycles were also used to investigate the behavior under axial force and rotation, see previous sections. Subsequently, Specimens CH2, CH3, and CH7 were tested in transverse shear V y , that is, shear perpendicular to the hinge rotation axis, whereas CH6 was tested in longitudinal shear V z .
The bending resistance of the end blocks was governing for the maximum applicable shear force, which was roughly 600 kN in transverse and 1200 kN in longitudinal direction. In order to exploit these capacities and the corresponding shear forces while ensuring a shear failure even in the case of the expected high resistances, and presuming a higher shear strength at higher axial force-as predicted by most design recommendations-the specimens were loaded in shear at a sufficiently high axial force N x , which was then progressively reduced at constant shear force until failure. The load combinations at shear failure are summarized in Table 1.
Specimen CH2 with a narrow unreinforced throat was loaded with a transverse shear force V y = À500 kN while being subjected to N x = 3 MN and r z = 23 mrad. No signs of distress due to the shear loading were observed. The shear load was then reversed to V y = +500 kN and the rotation r z reduced to zero, before N x was gradually decreased. The specimen failed at N x,u = 292 kN and V y,u = 501 kN, corresponding to a ratio V y /N x = 1.72. The corresponding average axial and shear stresses in the throat were σ x,u = 7 MPa and τ xy,u = 13 MPa. The shear failure plane produced by the shear force (positive at failure) extended from the front bottom to the back top corner of the throat, with an inclination of roughly 10 to the y-axis, see Figure 13a. The failure was very brittle and accompanied by an abrupt horizontal displacement of the two specimen halves and a drop in axial and shear forces.
Specimen CH6, also with a narrow unreinforced throat, was loaded with a longitudinal shear force V z = 1200 kN while subjected to N x = 2 MN. A rotation cycle up to r z = 15 mrad was applied to the specimen without causing any distress. Subsequently, after reducing the rotation to r z = 0 mrad, the axial load was gradually decreased until brittle failure occurred at N x,u = 1225 kN and V y,u = 1197 kN, at a ratio V z / N x = 0.98. The corresponding axial and shear stresses at the throat were σ x,u = 31 MPa and τ xy,u = 30 MPa. The failure plane ran horizontally across the specimen throat, and the two specimen halves displaced several centimeters in z-direction relative to each other.
Specimen CH7, with a wide unreinforced throat, was loaded with N x = 1 MN before a transverse shear force V y = 700 kN was applied. While keeping the axial and shear forces constant, a rotation cycle up to r z = 15 mrad was applied to the specimen. Afterwards, the loads were lowered to N x = 500 kN and V y = 600 kN, and a new rotation cycle up to r z = 20 mrad was imposed without causing the failure of the specimen. Subsequently, at r z = 0, the axial load was further reduced and shortly thereafter, failure occurred at N x,u = 267 kN and V y,u = 598 kN, at V z /N x = 2.24. The corresponding axial and shear stresses in the throat amounted to σ x,u = 2 MPa and τ xy,u = 5 MPa. The failure plane progressed steeply upwards until it intersected the first layer of reinforcement and continued just below it, see Figure 13b. Remarkably, despite that the throat was fully cracked along its entire length due to previous rotation cycles (Figure 13c), the shear failure still occurred across a different, steeper plane. After the test, the concrete wedge enclosed by the failure plane and the throat crack could be easily removed by hand.
All tested unreinforced Freyssinet hinges sustained very high shear forces, both in transverse and longitudinal direction, with ratios V/N x several times higher than F I G U R E 1 3 Shear failure in transverse direction of unreinforced hinges: Specimens (a) CH2 and (b) CH7 the limit values postulated by current design guidelines and rules, such as the German rules 2,25 allowing V/N x ≤ 0.125 and the British code 35 requiring V/N x ≤ 0.33 for unreinforced concrete hinges. Furthermore, the ratio V/ N x at failure decreased at higher axial compressive stresses, rather than being constant as presumed by the existing guidelines. The test results also indicate that rotations r z only have a limited effect on the transverse shear resistance: for example, Specimen CH7 underwent a large rotation, without any signs of failure, at a compressive force only marginally higher than the one that caused a shear failure with r z = 0.
The influence of reinforcement crossing the throat on the shear resistance was investigated with Specimen CH3, having a narrow throat. The specimen was loaded with V y = 500 kN, and the axial load was decreased stepwise, where at each step, a rotation cycle up to r z = 10 mrad was carried out. The last rotation cycle was applied at N x = 100 kN, without causing a shear failure. An attempt to produce a shear failure was finally made by increasing V y to 600 kN and reverting the axial force to a tensile force N x = À116 kN. At this point, the concrete in the throat was completely crushed, and the top and bottom halves of the specimen were kept together by the visibly deformed reinforcing bars crossing the throat, presumably carrying the entire loading. Even though a higher tensile force could have probably been applied to the hinge to ultimately cause a shear failure, the specimen was unloaded and the test terminated. Comparing the ultimate loads achieved by Specimens CH2 and CH3, it is evident that the shear reinforcement substantially increased the shear resistance at low axial loads with a substantial shear resistance of the reinforcing bars at zero (or even tensile) axial load. Presumably, the beneficial effect of the reinforcement is smaller at higher axial compressive load. Nevertheless a throat reinforcement is still beneficial to produce a more robust and ductile throat under shear loading even at higher axial load, as substantiated by the tests in. 10 After each of the experiments, the two specimen halves were realigned, and the specimen was reloaded to test drive the loading protocol of the following test. During these trials, high forces could be applied to the specimens that had failed in shear. This indicates that concrete hinges are very robust as long as they are subjected to sufficiently high axial compression.

| Torsion
The torsional resistance of concrete hinges was investigated with Specimen CH4. The authors are unaware of other (large-scale) torsion experiments on one-way concrete hinges. In the test CH4, after the initial rotation cycles at N x = 1.5 MN, the axial force was increased to 3 MN and subsequently, the torque M x was gradually increased until failure occurred at a torque M x = 361 kNm. Although not as brittle as the direct shear failures, the torsional failure was accompanied by a drop of torque and the axial force. The failure plane crossed the throat and separated the specimen into two parts that twisted axially relative to each other. The orientation of the failure plane at the north and south throat ends can be explained by the direction of the resulting shear stresses in the throat: at the north end of the throat, where the torque caused positive shear stresses in the y-direction, the failure surface extended from the front bottom to the back top corner of the throat, as in the shear failure shown in Figure 13a. At the south end of the throat, where the torque caused negative shear stresses in the y-direction, the failure surface had an opposite inclination (front top to the back bottom corner of the throat). It was not possible to identify the shape of the failure surface between the two longitudinal throat ends.
Assuming a triangular distribution of shear stresses across the throat, the achieved torque would require a nominal shear stress jt yx,max j = 51.8 MPa at the south and north ends of the throat. At the applied axial stress of σ x,u = 75.2 MPa, this corresponds to a ratio jt yx,max j=σ x,u = V/N x = 0.69, which appears plausible and aligns with the decreasing trend in the ratio V/N x observed in the shear tests as N x increases.
After the test was terminated, the two specimen halves were realigned, and the specimen was reloaded test drive the transverse shear loading protocol for the following test. The previously torsion-failed specimens could carry a shear load V y = 600 kN at an axial load N x = 1950 kN before shear failure occurred.

| CONCLUSIONS
Freyssinet concrete hinges boast a long history of successful application in practice, mainly in bridges. Despite the positive experiences and the numerous advantages over mechanical bearings, engineers often hesitate to use concrete hinges. This is mainly due to a lack of understanding of their mechanical behavior and the conservatism and uncertainties entailed in their design and execution. This paper presents the experimental part of a research project carried out by the authors, aiming at alleviating the existing uncertainties and facilitating the reliable design and assessment of concrete hinges.
In the frame of this project, seven large-scale Freyssinet concrete hinges were tested under general loading (all six stress resultants of a linear member) in the Large Universal Shell Element Tester of ETH Zurich. Five specimens had a very narrow (38 mm) and unreinforced throat; the influence of a wider (114 mm) throat and reinforcement crossing the through was investigated with the remaining two specimens. The specimens were subjected to a multi-stage load history targeting different failure types. A combination of DIC and distributed fiber optic measurements allowed a deeper insight into the structural behavior of the specimens.
All specimens subjected to axial compression N x and transverse rotation r z exhibited fine splitting cracks in the adjacent blocks, flaking (superficial spalling) of fine cement chips along the throat fillets, inelastic deformations, and bending cracks across the throat. The test results proved that these phenomena are no distress signs marking the beginning of the hinge failure, but merely normal consequences of the bearing behavior of a concrete hinge under moderate (service) loads.
Two specimens with narrow, unreinforced throat were subjected to rotations up to r z ≈ 60 mrad (much higher than rotations expected in practice) and simultaneously acting axial forces up to N x = 3 MN (nominal average axial stress across the throat cross-sectional area of σ x = 75 MPa ≈ 2 f c0 ) without inducing a failure. One of these specimens was loaded to failure in compression, with crushing of the throat concrete occurring at N x = 6 MN (σ x = 150 MN ≈ 4.5 f c0 ) and r z ≈ 23 mrad. Nevertheless, the axial load could be further increased significantly as the two adjacent blocks entered in contact beyond the throat.
The tests also proved that unreinforced hinges can sustain large shear forces V, in both, transverse as well as longitudinal directions. The achieved ratios V/N x at failure decreased as the magnitude of the axial stress increased, ranging between V/N x = 2.24 at σ x,u = 2 MPa and V/N x = 0.98 at σ x,u = 31 MPa. A moderate amount of reinforcement crossing the throat considerably increased the shear resistance at low axial loads and produced a very ductile shear behavior (in contrast to the brittle shear failure of unreinforced hinges). The resistance of the unreinforced hinge to moments about the strong axis and torques also proved to be significant.
Overall, the test results demonstrate that Freyssinet concrete hinges can carry very high combined loads and are very robust if sufficient axial compression is provided. A moderate amount (<3%, referred to the throat area) of reinforcement crossing the throat can be beneficial particularly to produce a ductile shear failure.
A comprehensive comparison with experimental data from the literature and current and newly-developed analytical models is presented in a follow-up paper. 23

A gt
Uniform elongation strain at peak load b Adjacent block length (in z-direction) b 1 Throat length (in z-direction) d Adjacent block width (in y-direction) d 1 Width

ACKNOWLEDGMENTS
The authors gratefully acknowledge the support by: • cemsuisse, the association of the Swiss cement producers, for partially funding this work (project number 201703). • The staff of the DSE systems company for the flawless coordination and manufacturing of the specimens.

CONFLICTS OF INTEREST
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.
• North end of the throat FLIR Grasshopper3 cameras (4096Á3000 pixels) with Rodagon lenses with a focal length = 60 mm, and 24 mm extension tube; achieved resolution = 13.7 px/mm; speckled pattern of black dots of diameter = 0.4 mm. Correlation parameters: subset size = 13 px, step size = 3 px and strain filter size = 5.
The stereo angle was approximately 25 , and the image acquisition frequency 1/4 Hz for all DIC systems. The randomly speckled patterns of black dots were applied onto the white-painted and well-lit specimen surfaces to achieve maximum contrast. The correlation was performed with the commercial software VIC-3D. 38 The average noise levels were determined according to 39 based on zero displacement tests. The average length of the virtual strain gauges used to compute the throat strains in Section 3.1.2 was 5.6 mm for CH2 and 16.5 mm for CH7, corresponding to roughly 70% and 82% of the throat height h t , respectively. For the total throat shortening plotted in Figure 9d, the average strains of the three virtual strain gauges were multiplied by the initial height h t of the throat.

B.2 | Distributed fiber optic sensing
The fiber optic sensors were connected to an optic reflectometer device (ODiSI-6104 supplied by Luna Innovations Incorporated), which allowed for quasi-continuous strain measurements along the concrete. The sensors used (BRUsens DSS 3.2 mm V9 grip) have a central glass fiber armored with a metal tube and an outer polyamide sheath. The manufacturer indicates a strain range of up to 10 mm/ m. The strain measurements were acquired with a virtual gauge length of 0.655 mm and a frequency of 0.8 Hz. The noise was reduced by first consolidating the data to a spatial resolution of 1.97 mm, then applying a low-pass filter with a cut-off frequency of 0.053 mm À1 in the space domain, and, finally, running a moving average filter with a subset size of 3.84 s. This process allowed the removal of noise and minor local effects without altering the overall strain curve's shape.
A PP E ND IX C: HINGE RESISTING MOMENT M z The hinge resisting moment M z is determined according to Figure C.1 by formulating moment equilibrium in the hinge axis on the deformed specimen halves, and averaging the result obtained from the upper and lower part (thereby eliminating the shear force since V y,bot = V y,top ): The forces, moments, and displacements at the specimen ends are measured by the LUSET. The deflection u y of the specimen center (i.e., throat) is obtained by the front and back DIC systems.