A series of disasters owing to abnormal or accidental loading conditions (i.e., impact, fire, blast, collision, gas explosion, landslides, floods, and so on) that occurred around the world have highlighted the poor performance of certain structural typologies and, in general, the extreme vulnerability of strategic and critical infrastructure to abnormal loads. Therefore, over the last decade, great effort has been made to investigate the progressive collapse resistance of structures [1-9]. Due to these unforeseen hazards not explicitly considered in the design, the structure can be at risk of initial local damage that can result in a spread of failure to surrounding members and, eventually, lead to the collapse of or a disproportionately large part of the structure, known as progressive collapse. In place of calculations demonstrating the effects of abnormal loads on buildings, the existing guidelines make use of the alternate path approach to determine the susceptibility to progressive collapse. This approach presumes that one critical member (typically vertical load-carrying columns or bearing walls) is removed because incapable of supporting the load and the remaining structure must be able to span across the removed member. After a column is destroyed by abnormal loads, different mechanisms are mobilized to transfer the gravity loads on superstructures and avoid collapse: Vierendel action, catenary action, columns acting as suspension, compressive arching in beams, membrane action in slabs. The “catenary action” refers to the ability of beams to resist the vertical loads by developing a string-like mechanism under large displacements and rotations [10-14]. After column removal, a new equilibrium is found by redistributing the internal forces of the remaining structure. Initially, the beams adjacent to the removed column tend to resist the vertical load by the generation of a bending moment. However, with increasing vertical displacement (i.e., in the large deformation stage) the gravity loads are mainly resisted by the vertical components of the axial forces that develop in the beams (i.e., catenary forces). This so-called “catenary action” generates a supplemental resistance that prevents progressive collapse of the structure. Practically, the load-carrying mechanism changes from plastic hinge action to tensile catenary action. However, its activation depends on the geometrical nonlinear behavior of the structure. At first, the structure should be ductile enough to allow large inelastic deformations and make possible the transition from the flexural resistance to the tensile catenary resistance. Then, the constraints of the beam end provided by the side columns should give a sufficient lateral anchorage to these catenary actions. Finally, the beam-to-column connections should be able to transfer the increased axial force in the beams due to the catenary action.

After the removal of a critical vertical load-bearing element (typically a column), an alternative load path should be provided. In this scenario, the connections of steel moment-resisting frames (MRFs) play a critical role since an alternative load path can be mobilized only if the beam-column connections have enough rotational capacity to sustain the abnormal extra-loads following the column removal. Usually, the beam-to-column joints of moment-resisting frames are designed to withstand gravity and seismic loads. However, the internal actions on these connections resulting from the sudden column removal differ significantly from those under earthquake loading. First, the connections are loaded cyclically under earthquake ground motions, while the column removal results in monotonic loading. Then, the beam-to-column joints are subjected primarily to bending under seismic loading, while they are subjected to bending and axial force under column removal scenarios. After a specific value of rotation, a considerable tensile axial force occurs in the beam and connections due to the development of catenary action, thus making its behavior very different from the seismic situation. This tensile axial force significantly reduces the moment capacity of the connection. Consequently, even moment frames appropriately designed for seismic loads may not resist progressive collapse since a significant axial force can develop in the connections. Thus, the behavior of the beam-to-column joints undergoing large deformation should be properly studied since moment, shear, and tension, in conjunction with high ductility demand, simultaneously occur. The connections should be ductile enough to sustain significant deformation while maintaining sufficient integrity to develop catenary forces. Only in this case, the catenary action can give an alternate load path after the flexural strength is reached and the flexural failure begins. The catenary action can provide considerable resistance depending on different factors: adequate vertical displacement to produce a significant vertical component in the axial force, high lateral restraint provided by the adjacent structures, and subsequent high tensile axial forces in beams and connections. Therefore, the performance of beams and beam-to-column connections plays an important role in the activation and development of catenary action under large deformations [15, 16]. The nonlinear behavior of connections in steel frames depends on many parameters, including the performance of steel material, slips, local inelastic deformations, and fractures in bolts, weldings, flanges, and so on. In the past, a great effort was made to evaluate the effect of the catenary action in increasing the progressive collapse resistance of steel frames [17, 18]. However, many studies in the literature have been dedicated to the performance of beam-to-column connections under pure bending [2-4, 6-9, 12-14], thus neglecting the negative effect of the tensile axial force in connections due to the catenary action. However, under critical column removal scenarios, the steel connections are subjected to bending moment combined with the tensile force due to the large vertical displacement. Thus, in recent years many experimental studies have been developed to study the dual effect of axial tension in beam-to-column connections and its impact on the progressive collapse resistance including the catenary effect [19-38]. Nevertheless, most existing analytical models for beam assemblies under a progressive collapse scenario essentially assume that the plastic strength remains available despite very large deformation [39]. Consequently, the strength degradation due to local failure is neglected and, thus, the effect of catenary action on the resistance can be largely overestimated.

The experimental studies available in the literature have shown that different beam-to-column joints have different performances under a column removal scenario, as far as both the progressive collapse resistance and the catenary mechanism. Therefore, it seems necessary to develop a database of full-scale test results on double-span assemblies. This paper tries to overcome this gap by presenting a critical review of experimental tests of different types of beam-to-column connections under bending and axial tension due to column removal. To this aim, fully rigid, semi-rigid, and flexible connections are investigated. Their load-displacement response both in flexural and catenary action stages are described and commented on below. The resistance and rotational capacities of beam-column joints are investigated, and their effect on the activation and extend of catenary action is evaluated. The failure modes under sudden column loss as determined experimentally are discussed in detail. The test results are finally used to evaluate the rotation capacity of several steel subassemblies. The values obtained are finally compared to the acceptance criteria specified by the main progressive collapse guidelines for several beam-to-column connection categories.

The Northridge earthquake in 1994 and the Kobe earthquake in 1995 revealed the brittle failure of beam-to-column connections, thus generating alarm about the reliability of the seismic codes where the ductility of beam-to-column connections is essential for the steel moment frames to be considered highly ductile. In the same way, the collapse of the World Trade Center produced concern about the design against abnormal loads, thus stimulating the development of standards and design guidelines to resist progressive collapse [40-43]. The beam–to–column connections play a critical role in progressive collapse resistance of steel frames since the rotation capacity of joints usually controls the activation of the catenary action. The corresponding acceptance criteria were defined using both tests results and numerical simulations available in the literature. In the 2005 version of UFC 4-023-03 [40] the rotational capacity values for connections were based on the 2003 version of GSA [42] and reasonably agreed with those in ASCE 41 [44]. However, the ASCE 41 acceptance criteria were based on experimental tests of beam-to-column connections under cyclic loads, where the rotational capacity is limited by degradation and loss of strength due to low cycle fatigue. On the contrary, in the experimental tests under a central-column-removal scenario, the beam-to-column connections are subjected to monotonic loads and significant axial tension forces. Therefore, significant performance differences were found, due to the effects of loading (monotonic vs. cyclic) and the ultimate state (moment-axial tension interaction vs. moment only). As a consequence, the modeling and acceptance criteria in the 2016 version of UFC [41] and GSA [43] were defined based on a comparison between the deformation limits contained in various documents and guidelines, such as ASCE 41 [44], Eurocode 3 [45], 2005 UFC 4-023-03 [40] and GSA Test Program [46]. Nevertheless, it seems necessary to develop an extended database of test results since different beam-to-column connections show significant behavioral differences, especially in the catenary mechanism.

After column removal, a “double-span” scenario arises, and the performance of the affected moment connections plays a central role in the progressive collapse resistance. Therefore, the experimental tests are performed on double-span assemblies comprising two adjacent beams connected by a failure column at the center (Fig. 1). In this situation, the moment connections are the weakest components, thus affecting the anti-collapse resisting capacity of the beam–to–column assemblies. Among them, the simple connections, which are nominally pinned connections, are very popular due to their simplicity and, thus, they’re widely used both in non-seismically designed steel frame structures and in the gravity frames of seismically designed steel buildings. However, previous research papers highlighted that the simple connections are very vulnerable to progressive collapse, even if they’re designed to meet the tie-force requirements of building codes [18, 19, 47-49]. If large rotations are not explicitly accounted for in the design stage, the simple connections can fail before the development of tensile catenary action because of the large ductility demand [10, 50]. On the other side, the progressive collapse performance of fully rigid and semi-rigid beam-to-column connections is very sensitive to the tensile axial force in the connected beams since the catenary action is the fundamental mechanism to resist vertical loads in the large deformation stage. This is the reason why many types of beam-to-column connections have been experimentally investigated, including simple, rigid, and semi-rigid connections. In the next section, since beam-to-column connections play a critical role in progressive collapse resistance of steel structures due to the catenary mechanism, the performance of different beam-to-column connections against progressive collapse is investigated using available full-scale tests results on double-span assemblies.

Fig. (1). Double-span assembly

In this section are summarized the experimental tests on simple (flexible) steel beam-column connections available in the literature. Table 1 shows the details about the experimental tests herein considered, including reference, connection type, specimen, span-to-depth ratio, and failure mode. The beam-to-column connection types herein considered are 1-14) bolted-angle connections tested by Yang & Tan [20]; 15-18) web cleat, top and seat angle, top and seat with web angle (TWSA), and fin plate connections tested by Yang & Tan [21]; 19) web cleat connection tested by Liu et al. [22]; 20) double web angle tested by Zhong et al. [23]; 21) shear-connection tested by Alrubaidi et al. [24]. Yang & Tan [20] experimentally investigated bolted-angle connections under tension, considering fourteen specimens by varying some important parameters such as material properties, bolt size, angle thickness, and bolt hole positions. Depending on the strength ratio between angles and bolts, five different types of failure modes were observed during the tests (i.e., angle fracture at bolt holes, angle fracture at bolt holes with yielded bolts, angle fracture close to heel, angle fracture close to heel with yielded bolts and bolt fracture with yielded angles). The test results show that the load-displacement responses of bolted-angle connections were dominated by the response at the large deformation stage rather than at the small deformation stage . Therefore, the ultimate strength of connections is much greater than the yield strength. Yang & Tan [21] carried out experimental tests on simple connections including web cleat, top and seat angle, top and seat with web angle (TSWA) (8 mm thickness angles), and fin plate connections. They investigated the failure modes of these connections and their abilities to deform in the catenary mode. The test results show that the tensile capacity of the beam-to-column connection controls the failure mode and the mobilization of catenary action. For web cleat and fin plate connections, the behavior is dominated by catenary action since the applied load increases significantly only at the large deformation stage. The web cleat connection has the best performance in the development of catenary action if compared to the other connections. For top and seat angle connections, the behavior is governed by the flexural action with a little load contribution from the catenary action. Finally, the rotation capacities of beam-to-column connections based on the experimental results were found much higher than the values recommended by UFC [41] and GSA [43].

Liu et al. [22] developed experimental tests and numerical analyses to investigate the dynamic behavior of web cleat beam-column connections subjected to a sudden column removal scenario. Test results showed that the maximum displacement of the web cleat connections under sudden column removal would be significantly increased compared with the one under static loading conditions. The dynamic response of the connections (defined as the relationship between the initial support force of the middle column and the maximum dynamic displacement) was also investigated. The analysis results showed that the maximum dynamic load capacity of this web cleat connection was about 2.8 times lower than its static load capacity. Zhong et al. [23] developed experimental tests to study the mechanical behavior and load-deformation response of a double web angle connection (DWA). The failure of the specimen occurs for the fracture of the double web angles at the bolt hole. The behavior is dominated by the catenary action since the resistance of the specimen is mainly provided by this effect. Alrubaidi et al. [24] investigated the progressive collapse risk of three one-third scale single-story, two-bay steel frames under a column removal scenario, also considering a shear-connection (S-C) specimen. The shear connection showed a very high progressive collapse risk since the beams are unable to redistribute the load carried by the removed column to the adjacent members.

This section summarizes the experimental tests on rigid steel beam-column connections available in the literature. Table 2 shows the details of the experimental tests herein considered, including reference, connection type, specimen, span-to-depth ratio, and failure mode. The beam-to-column connection types herein considered are: 1-2) Beam-to-tubular column moment connections tested by Li et al. [25] (N.1 welded flange, welded web (CO-W); N.2 welded flange, bolted web (CO-WB)); 3-4) welded unreinforced flange, bolted web (WUF-B) and reduced beam section (RBS) connections tested by Lew et al. [26]; 5-6) welded unreinforced flange-bolted web connections (SI-WB) tested by Li et al. [27]; 7-9) welded flange-weld web connection (I-W), welded flange-bolted web connection (I-WB) and welded flange-bolted web connection with shear diaphragm (ST-WB) tested by Wang et al. [28]; 10) welded unreinforced flange-welded web connection (WUF) tested by Zhong et al. [23]; 11-12) welded cover plate flange connection (CWP) and haunch end plate bolted connection (EPH) tested by Dinu et al. [29]; 13-15) welded unreinforced flange-bolted web connections (WUF) tested by Meng et al. [30]; 16) bolted flange plate connection (BFP) tested by Wang et al. [31]; 17) SidePlate moment connection tested by Faridmehr et al. [32]. Li et al. [25] presented two full-scale laboratory tests on steel beam-to-tubular column moment connections with outer-diaphragm, a welded flange-bolted web connection (Specimen CO-WB), and a welded flange-welded web connection (Specimen CO-W). The specimens exhibited the same typical flexural behavior at the early stage, while exhibited two different modes of response after the fracture of the bottom flanges. For the specimen CO-W, the crack propagated deep into the web plate, thus causing a catastrophic reduction of the effective beam section. For the specimen CO-WB, the fracture at the bottom flange was interrupted from propagating into the web plate by the presence of the bolts. Thus the catenary action was developed following the flexural failure. Lew et al. [26] tested two full-scale steel beam-column connections, namely a welded unreinforced flange bolted web (WUF-B) connection and a reduced beam section (RBS) connection. The beam-column connections show an initial elastic response dominated by flexure. Then, the connections yield with increasing the vertical displacement, and tensile axial forces develop in the beams, indicating the catenary action until the connections fail under the combined bending and axial stresses. The rotational capacities were approximately twice as large as those based on seismic test data. Li et al. [27] investigated the catenary behavior of two full-scale beam-to-column assemblies, with typical H-beam and square-column moment connections with a welded unreinforced flange-bolted web connection (i.e. WUF-B connection). The results showed that both specimens were able to develop an effective catenary action via the bolted web following the primary flexural phase. The failure modes depend on the bolt arrangement. When the bolts were arranged in two rows, the shear tab cracked at the section across the bolt hole, and this makes the connection less robust. Wang et al. [28] investigated three connection types including flange-weld web connection with internal diaphragms (specimen I-W), welded flange-bolted web connection with internal diaphragms (specimen I-WB), and welded flange-bolted web connection with short through diaphragms (specimen ST-WB). The test results showed satisfactory ductility supply and load resistance of the three specimens. The lowest ductility against the initial fracture was exhibited by the specimen ST-WB due to nonsymmetrical stress distribution over the width of the beam flange. Remarkable catenary action was developed by I-WB and ST-WB specimens, while specimen I-W showed a marginal catenary action due to quick propagation of the crack over the entire beam section. Zhong et al. [23] conducted experimental tests on a welded unreinforced flange-bolted web connection (WUF specimen). The WUF connection experienced a flexural mechanism phase, a flexure–catenary mixed mechanism phase, and finally a catenary mechanism phase. The failure of the specimen occurs for the fracture of the tension flanges of the beam. Dinu et al. [29] investigated two rigid connection types designed to meet the seismic design requirements for special moment-resisting frame connections, namely, the welded cover plate flange connection (CWP) and the haunch end plate bolted connection (EPH). The specimens showed a large catenary response and both failed in the connection, owing to the combined effects of the flexural and tensile actions. The CWP and EPH connections showed ultimate rotations (i.e., 0.193 rad for the CWP specimen and 0.130 rad for the EPH specimen) greater than the acceptance criteria of UFC [41] with a significant contribution of catenary action to the ultimate load resistance. Meng et al. [30] tested three types of specimens with different span ratios (1:0.6, 1:1.0, and 1:1.4), realized using welded unreinforced flange-bolted web connections. The load-displacement curves show multiple peak loads since the local failure occurred several times in the connection zone, due to the initial fracture of the tension flange connection to the failure column, followed by the fracture of the tension flange of the beam's connection to the corresponding side column. Wang et al. [31] investigated the progressive collapse behavior of various bolted flange plate (BFP) connections. In the early stage of the test, the resistance of the frame was mainly offered by the flexural action. When the first peak load was achieved, the flexural resistance decreases as an effect of compressive arch action. Then, the tensile axial force of the beam gradually increased while the bending moment decreased, and the load-resistance mechanism shifts from the flexural mechanism to the catenary mechanism. The failure in the catenary stage was controlled by the tension fracture of the net section of flange plates or shear tab. Faridmehr et al. [32] tested the anti-collapse behavior of double-span assemblies with a fully rigid connection known as Side Plate moment connection (welded flange-welded web) using a model scaled down to 1/6th of the real size. The experimental results show that the SidePlate moment connection reached significantly high load and rotational capacities developing full catenary action. The failure mode is controlled by the tensile resistance of the beam-column joint after undergoing large rotations.

This section summarizes the experimental tests on semi-rigid steel beam-column connections. Table 3 shows the details of the experimental tests, including reference, connection type, specimen, span-to-depth ratio, and failure mode. The beam-to-column connection types herein considered are: 1-2) welded unreinforced flange-fillet welded web connection (WUF-FW) and bolted unstiffened extended end-plate with pre-tensioned high-strength bolts connections (4 E-BUEEP-P) tested by Alrubaidi et al. [24]; 3-5) complete joint penetration (CJP) and reduced-web-section welded (RWS) connections tested by Lin et al. [33]; 6-8) flush end-plate, extended end-plate and top and seat-web angle connections tested by Yang & Tan [37]; 9) flush end-plate (EPF) connection tested by Liu et al. [34]; 10) top and seat-web angle (TSDWA) connection tested by Zhong et al. [23]; 11-12) reduced beam section welded connection (RBS) and unstiffened extended end plate bolted connection (EP) tested by Dinu et al. [29]; 13-18) flush end-plate connections tested by Gao et al. [35]. Alrubaidi et al. [24] investigated experimentally one-third scale single-story two-bay steel frame with two different steel intermediate moment frame (IMF) connections conforming to ANSI/AISC 358–16, namely a welded unreinforced flange-fillet welded web (WUF-FW specimen) and a bolted unstiffened extended end-plate with pretensioned high-strength bolts (4 E-BUEEP-P specimen). For the WUF-FW specimen, the fracture of fillet welding between the bottom flange of the beam and middle column occurs, thus preventing the mobilization of the full catenary action. For the 4 E-BUEEP-P specimen, the performance of the connection was initially controlled by flexural action, and then, with increased vertical displacement, it is dominated by the full development of the catenary action. Lin et al. [33] developed an experimental study on steel beam-column connections with complete joint penetrations (CJP) and reduced-web-section (RWS) connections with different circular web openings geometric parameters. The experimental results show that the vertical resistance was provided by flexural action in the early stages and then became catenary action as the vertical displacement increased. Two different failure modes were observed: continuous failure by the weld line fracture in CJP connection and interrupted failure in RWS connections due to the fracture of a perforated section. Yang et al. [37] presented the experimental tests of different beam-to-column connections, including flush endplate, extended end plate, and TSWA (12 mm angle) connections. The test results show that the tensile capacity of the beam-column joint, after undergoing large rotations, usually controls both the catenary action and the failure mode. The flexural action dominates the behavior of extended end plate connections since a little load contribution comes from the catenary action. On the contrary, the flush endplate and TSWA connections develop catenary action before failure. The rotation capacities of beam-column joints based on the experimental results were found much higher than the design values of UFC [41]. Liu et al. [34] conducted dynamic tests to investigate the behavior of flush endplate steel beam-column cruciform connections by using a quick-release mechanism to simulate the instantaneous removal of a column. Test results showed that the maximum dynamic displacement was significantly increased compared to the corresponding quasi-static displacement. Zhong et al. [23] investigated the progressive collapse behavior of a top-seat angle with a double web-angle connection (TSDWA). The test results show that the TSDWA connection experienced a flexural mechanism phase, a flexure–catenary mixed mechanism phase, and finally a catenary mechanism phase. The failure of the specimen occurs for the fracture of the tension angles connecting the column and beam at the bolt hole (TSDWA specimen). Dinu et al. [29] investigated the progressive collapse behavior of a reduced beam section welded connection (RBS), and an unstiffened extended end plate bolted connection (EP). The experimental results showed that the RBS specimen showed ultimate rotations greater than the acceptance criteria of UFC [41] with a significant contribution of the catenary action to the ultimate load resistance. On the contrary, the specimen EP shows the lowest ductility and ultimate load resistance, and no catenary action occurs due to the fracture of bolts in tension. Gao et al. [35] conducted a series of flush endplate semi-rigid composite joints under pure bending, pure tension, and a combination of bending moment and tension, using the specimens named SJS (sagging moment), SJSTand SJST2 (sagging moment + tensile force), SJH (Hogging moment), SJHT (Hogging moment + tensile force) and SJT (tensile force). The results from experimental tests show under pure bending moment, the semi-rigid composite joints possess sufficient rotation capacity for forming the catenary action. Under the combination of bending moment and tension, the moment capacity of the composite joint decreases linearly along with the increase of tensile force. Under tensile force, the joints tend to fail at the “catenary phase”.