Engineering Structures 24 (2002) 227–237 www.elsevier.com/locate/engstruct Moment–rotation model of semi-rigid connections with angles Sang-Sup Lee, Tae-Sup Moon * Division of Architectural Engineering, Hanyang University, Seoul 133-791, South Korea Received 15 December 2000; received in revised form 26 June 2001; accepted 26 June 2001 Abstract A 2-parameter log model to describe the non-linear M–q relationship of semi-rigid connections is proposed. The proposed model accurately describes the moment–rotation behavior of nearly all connections by controlling shape parameters a and n. To overcome the disadvantage that the shape parameters have very little physical meaning, the semi-analytic approach that optimum values of a and n, agreeing rather well with experimental curves, are derived in terms of initial stiffness (ki) and plastic stiffness (kp) from statistical regression analysis is applied to this paper. The analytic formulations of ki and kp are developed for double web angle connection and top and seat angle connection respectively. 2002 Elsevier Science Ltd. All rights reserved. Keywords: Moment–rotation model; Semi-rigid connection; Angle connection 1. Introduction For conventional analysis and design of a steel-framed structure, the actual behavior of beam-to-column connections is simplified to the two idealized extremes of either rigid-joint or pinned-joint behavior. However most connections used in steel frames actually exhibit semirigid deformation behavior that can contribute substantially to overall force distribution in the members [1]. As such steel frame connections should be treated as being a ‘semi-rigid frame’ for the purpose of proper analysis and design. Accurate modeling of their moment–rotation (M–q) relationship is necessary if the effects of connections are to be considered in structural analysis [2]. Because it would be prohibitively expensive to obtain the moment–rotation relationships for all practical connection types by full-scale tests, an attractive solution is to derive a single standardized model for each connection type. Many models have been proposed and are generally composed of parameters depending on the strength, stiffness and ductility for a given connection and shape parameters treated as a curve-fitting parameter. As these parameters are mainly expressed in * Corresponding author. Tel.: +82-02-2290-0312; fax: 82-022296-4145. E-mail address: bluemoon@email.hanyang.ac.kr (T.-S. Moon). terms of the geometric variable, the expressions prove to be accurate only for the limited range of data used in regression analysis. The problem can be solved by using the many test data, or by theoretical parametric study. The purpose of this paper is to propose a practical model regarding semi-rigid connections with angles that predicts the moment–rotation characteristics by determining two shape parameters, and to develop the analytic formulations for evaluating the two shape parameters for double web angle connection and top and seat angle connection as shown in Fig. 1. 2. Modeling of connection 2.1. Connection model [3] 2.1.1. Linear model The first attempt at using mathematical expression to define the M–q curve dates back to Rathbun (1936). A single tangential line to the initial slope of the M–q curve was defined as the semi-rigid connection factor Z. Z⫽q/M (1) Tarpy and Cardinal (1981), Melchers and Kaur (1982), Lui and Chen (1983) also proposed the bilinear models in which the initial slope of the moment–rotation line is replaced by a shallower line at a certain transition 0141-0296/02/$ - see front matter 2002 Elsevier Science Ltd. All rights reserved. PII: S 0 1 4 1 - 0 2 9 6 ( 0 1 ) 0 0 0 6 6 - 9 228 S.-S. Lee, T.-S. Moon / Engineering Structures 24 (2002) 227–237 Fig. 1. Semi-rigid connections with angle. (a) Double web angle connection; (b) top and seat angle connection. moment. Razzaq (1983) used the sophisticated piecewise linear models represented by a series of straight line segments. Even though these linear models are easy to use, the inaccuracies and sudden jump in stiffness at the transition points make their practical use difficult. 2.1.2. Polynomial model A polynomial function model was proposed by Sommer (1969) used has also been used to express rotation in terms of moment. q⫽C1(KM)1⫹C2(KM)3⫹C3(KM)5 (2) where K is the standardization factor dependent upon the connection type and geometry and C1, C2 and C3 are curve-fitting constants. This model represents the M–q behavior reasonably well. The main drawback is that the nature of a polynomial is to peak and trough within a certain range. The connection stiffness may become negative at some values of M. This negative stiffness may cause numerical difficulties in the analysis of frame structures if the tangent stiffness formulation is used. 2.1.3. Power model Richard and Abbott (1975) proposed primarily to represent the stress–strain relationship, which expresses moment in terms of rotation as follows: M⫽ (ki−kp)q n (ki−kp)q 1+ M0 冉| |冊 ⫹kpq (1/n) (3) where: ki kp M0 n is is is is the the the the initial slope of connection strain hardening slope of connection reference moment shape parameter One advantage of the Richard–Abbott function over the Ramberg–Osgood function is that whereas the former permits positive, zero, and negative values of kp, the latter permits only a zero value. Ang and Morris (1984) followed the same procedure but used the Ramberg–Osgood (1943) function to express the standardized moment–rotation behavior in the following form: 再 冋 册 冎 q (KM) KM ⫽ 1⫹ q0 (KM)0 (KM)0 (n−1) (4) where q0, (KM)0, n are constants that define shape of function, K is the standardization constant dependent on the connection type and geometry. Depending upon the value of n, the shape of the curve changes, allowing one to model a sharp ‘knee’-type M–q curve or one with a long, gradual decreasing slope. The Ramberg–Osgood function has the advantage that its derivative, hence the slope of the M–q curve, does not fluctuate in value contrary to the inherent oscillatory nature of polynomials. 2.1.4. Exponential model Lui and Chen (1986) proposed the multi-parameter exponential model. 冘冋 m 冉 冊册 |q| Cj 1⫺exp ⫺ 2ja j⫽1 M⫽ ⫹M0⫹Rkf|q| (5) where M0 is the starting value of the connection moment to which the curve is fitted, Rkf is the strain-hardening stiffness of the connection, a is a scaling factor, and Cj is a curve-fitting constant obtained from a linear regression analysis. This model gives as good a curve-fitting to test data as that of the cubic B-spline model. However, if there is a sharp change in slope in the M–q curve, this model cannot represent it adequately. Kishi and Chen (1986) refined the Lui and Chen exponential model to accommodate any sharp change in slope in the M–q curve as follows 冘冋 m 冉 冊册 冘 |q| Cj 1⫺exp ⫺ 2ja j⫽1 M⫽M0⫹ n ⫹ Dk(q⫺qk)H[q (6) k⫽1 ⫺qk] where M0 and a are defined as in Eq. (5) and Dk is a constant parameter for the linear portion of the curve, qk S.-S. Lee, T.-S. Moon / Engineering Structures 24 (2002) 227–237 is the starting rotation of the linear component of the curve, H[q] is Heaviside’s step function (unity for qⱖ 0, zero for q⬍0) and Cj and Dk and curve-fitting constants obtained from a linear regression analysis. Although this model can deal with connection loading and unloading for the full range of relative rotation in a second-order structural analysis with secant connection stiffness, it is not manageable. 2.2. Modeling of M–q curve Using the two shape parameters a and n, the model to represent the moment–rotation relationship of semirigid connections is proposed as follows M⫽aln(n·103·q⫹1)n Fig. 2. moment and the shape parameter n governs an the entire shape of a curve. Hence, this log model is found suitable and adjustable for the representation of the various connection behaviors connection. As can be observed in Fig. 3, the proposed model gives a good curve-fitting with test data selected in the database (SCDB; The Steel Connection Data Bank program) by Kishi and Chen (1986) [3]. And this model can be easily applied to the second-order frame analysis with semi-rigid connection because the tangent stiffness (kt) can be determined directly from Eq. (8) without iteration. dM n·103 (8) kt⫽ ⫽n·a[ln(n·103·q⫹1)]n−1· dq n·103+1 (7) where a and n are shape parameter which are determined by using the method of least squares for differences between the predicted moments and the experimental test data. This log model has the properties that the larger the index n, the more linear the curve, and the smaller, the flatter as shown in Fig. 2(a)–(c). For example, the curve is nearly akin to the linear behavior in that n equals 3.0 and shows the general M–q behavior in the case of n=1.0. Otherwise, if n is 0.1, the curve is characterized by bilinear behavior. For constant values of a, the plotting of Eq. (7) for different values of n is shown in Fig. 2(d). Consequently, it can be argued that the shape parameter a has an effect upon initial stiffness and ultimate 229 Fig. 3. Comparison of curve-fitting curves and test curves. Characteristic of shape parameters. (a) n=3; (b) n=1; (c) n=0.01; (d) sensitivity of n 230 S.-S. Lee, T.-S. Moon / Engineering Structures 24 (2002) 227–237 2.3. Study of model parameters ⫽0.91) As described above, it can be seen that the proposed model represents a close fit with the experimental tests by controlling a and n only. While other models are expressed in terms of stiffness, strength, and standardization factor etc, the proposed model has no such parameters. This fact could be a demerit that the parameters a and n in this log model have very little physical meaning. On the other hand, it can be offset by simplicity and accuracy that this model requires only two parameters and can provide a good fit with various connection M– q curves. Some parameters of most models have a form of function derived theoretically, and the others are determined by using the statistical analysis for experimental test data. The accuracy of those models is under the influence of the number of data used, and the availability has the limited type of connection M–q curve. To overcome these disadvantages, the approach that the optimum values of a and n, agreeing rather well with experimental curves, are derived in terms of initial stiffness (ki) and plastic stiffness (kp) from statistical regression analysis is applied to this paper. a⫽f(ki,kp) (9a) n⫽f(ki,kp) (9b) The ultimate moment is excluded from Eqs. (9a) and (9b), because most M–q curves of top and seat angle connections and double web angle connections exhibit elastic-plastic hardening characteristics and do not flatten out near the final state of loading. To express a and n as a function of ki and kp, a total of 75 experimental data were used for the derivation of Eqs. (10a, 10b, 11a) and (11b). A set of 50 double web angle connections and a set of 25 top and seat angle connections are employed. These experimental data were selected from database collected by Kishi and Chen. A simple statistical regression analysis on shape parameters is carried out along with ki and kp. The expressions of a and n, representing double web angle connections, are statistically analyzed and regressed to a plane equation are shown in Fig. 4. a⫽1.499E⫺03·ki⫹1.449E⫺03·kp⫹0.704 (R2 (10a) ⫽0.83) n⫽⫺3.594E⫺05·ki⫺3.496E⫺05·kp⫹1.170 (R2 (10b) ⫽0.80) Similarly, the expressions of shape parameters of top and seat angle connections are statistically analyzed and plotted in Fig. 5. The a and n can be expressed as a⫽9.689E⫺0.4·ki⫹9.562E⫺04·kp⫹3.850 (R2 (11a) n⫽4.500E⫺06·ki⫹4.400E⫺06·kp⫹0.601 (R2 (11b) ⫽0.92) where R2 is correlation factor and the unit of ki (kp) is kN·m. 3. Behavior of angle segment If the initial stiffness and the plastic stiffness could be determined analytically, the prediction of the M–q relationship from the proposed model can be made independent of experiment. In order to determine ki and kp, the load–deformation (P–⌬) the behavior of angle segment in tension needs to be examined first of all. The behavior of angle segment will be analyzed for the two states as follows 1. Elastic state 2. Plastic state in which the collapse mechanism is developed The basic assumptions for analyzing are as follows. 1. Deformation of connection is small. 2. Deformation of beam and column is negligible compared with the deformation of connection. 3. The slip deformation is negligible. 3.1. Pe–⌬e relationship in elastic state The assembly angles of connection can be simplified as shown in Fig. 6(a). It is assumed that the fasteners hold the angle legs fixed at points A and C. Point C moves downward only because of symmetry, and the rotation at B is the same in both legs as a result of continuity. If the angle is temporarily free to translate and rotate at C in Fig. 6(c), the moment (Me) and the horizontal load (He) contribute to the deformation of point C: (1) vertical displacement (⌬v); (2) horizontal displacement (⌬h); (3) rotation (q). As the situation of point C is actually a fixed restraint, there is no horizontal displacement or rotation. Therefore, the boundary conditions are ⌺⌬h⫽⌬h(Pe)⫹⌬h(He)⫹⌬h(Me)⫽0 (12a) ⌺qr⫽qr(Pe)⫹qr(He)⫹qr(Me)⫽0 (12b) ⌺⌬v⫽⌬v(Pe)⫹⌬v(He)⫹⌬v(Me)⫽⌬e (12c) As the angles shown in Fig. 6(b) are pulled out in an initial elastic state, it is assumed that the fixed support at point A is the first fastener-nut edge in the leg adjacent to the column face and the concentrated load Pe acts at S.-S. Lee, T.-S. Moon / Engineering Structures 24 (2002) 227–237 Fig. 6. Fig. 4. Statistical regression analysis on double web angle connections. Fig. 5. Statistical regression analysis on top and seat angle connections. Angle segment in elastic state. (a) Double angle segment; (b) modeling; (c) free body diagram. the center of fastener holes in the leg adjacent to the beam web face because of the higher restraint caused by clamping action [4]. Based on these assumptions and using various elastic theories, boundary conditions with respect to material and geometrical properties can be rewritten 冉 231 冊 冉 冊 He g32 Me g22 Pe g1g22⫹ ⫺ g1g2⫹ ⫽0 ⫺ (g21g2)⫹ EI EI 3 EI 2 (12aa) 冉冊 冉冊 冉 冊 冉冊 Pe g21 He Me ⫺ (g1g2⫹g22)⫹ (g1⫹g2)⫽0 EI 2 EI EI (12bb) Pe g21 He g21g2 Me g21 ⫺ ⫹ ⫽⌬e EI 3 EI 2 EI 2 (12cc) where: g1 distance from the leg’s center line to the first fastener-nut edge in leg adjacent to the column 232 S.-S. Lee, T.-S. Moon / Engineering Structures 24 (2002) 227–237 Fig. 7. g2 EI Angle segment in elastic state. (a) Mechanism of double angles; (b) location of plastic hinges; (c) deformed configuration. distance from the leg’s center line to the center of fastener holes in leg adjacent to the beam bending stiffness of angle segment per unit length. Solving the simultaneous equation Eqs. (12aa, 12bb) makes it possible to determine the relative magnitude of He and Me. Substituting He and Me solved for Pe into Eqs. (12cc), the load–deformation (Pe–⌬e) of angle segment in elastic state can be obtained as 冉 冊 12EI g1+g2 ·⌬ Pe⫽ 3 g1 g1+4g2 e (13) 3.2. Pp–dp relationship in plastic state The comprehensive experiments showed that there was a significant difference in behavior of angles with different mechanisms. In this state, the geometrical changes of angles are large compared to the restrained lengths of the angle legs. Therefore, it is necessary that the deformed shape of angles be known or at least closely approximated so that the locations of hinges can be determined. Figs. 8 and 9 shows the deformed configuration of angle consistent with vertical deformation ⌬p. It is assumed that the angle segment behaves in the following manner. 1. As the plastic hinge at point B is formed, the A–B part rotates as a rigid body around the base at A. 2. As the plastic hinge at point C is formed, the C–D part rotates as a rigid body around the base at D. 3. The heel of angle (fillet) does not deform. As observed in Fig. 3, the behavior of connections has a transition between the elastic state and plastic state [5]. The load–deformation relationship in transition state is not treated here because kp is necessary to be determined in plastic state in which the collapse mechanism is developed. As the angles are pulled out still more, the plastic hinges will be formed at point A primarily, and then point B and point C simultaneously. Usually, point A is the edge of the first fastener-nut in the leg adjacent to the column face and the points B and C are in the vicinity of the two intersections of the fillet and the angle legs as shown in Fig. 7. However, point A has a migration toward the angle’s toe that is caused by a snug tight fastener and relatively thicker angle (Fig. 8) [6]. Fig. 8. Migration of plastic hinge. (a) Mode 1; (b) Mode 2. Fig. 9. Detail of angle configuration in plastic state. S.-S. Lee, T.-S. Moon / Engineering Structures 24 (2002) 227–237 By equating the moment at these points to the plastic moment, the equilibrium equation is 冦 冧 g3(冑g23+⌬2p−g3) g3·⌬p ⫹Mp⫹Hp· 2 2⫽ ⫺Pp· g3⫺ 2 2 冑g3+⌬p 冑g3+⌬p 冦冢 冣 g3 冑g +⌬ 2 3 2 p ·g2− 冑 冑g +⌬ g3( g23+⌬2p−g3) 2 3 2 p 冧 (14) 4.1.1. Initial stiffness (ki) Typical double web angle connection is shown in Fig. 1(a). In the initial elastic state, it is assumed that there is no slip and that the leg of angle adjacent to the column behaves linearly elastic, while the leg of angle adjacent to the beam behaves as a rigid body [7]. Since the higher restraint between the angle and the column face arises from the clamping action of bolts, it is assumed that the full length of the angle can resist the bending moment. The center of rotation is close to the mid-length of the connection [8]. Using the Pe–⌬e relationship of angle segment in Eq. (13) and the load distribution as shown in Fig. 10, the relationship between the moment M and the rotation q is obtained as +Mp (at point C) 冦 冢 冣冧 g3 +Hp· g1−g3+⌬p+ 1− 冑g +⌬ 2 3 2 p ·g2 =−Mp M⫽2⫻ (15) where: g1 distance from the leg’s center line to the plastic hinge (point A) in leg adjacent to the column distance from the leg’s center line to the first fastener-nut edge in leg adjacent to the beam distance from the toe of fillet to the plastic hinge (point A) in leg adjacent to the column plastic moment (=t2a·sy/4 (ta: thickness of angle and sy: yield strength of angle) g2 g3 Mp For the purpose of simplification, the higher-order terms (⌬3p…) are neglected and the expression in Eq. (16) is used. 冑g +⌬ ⬇g · 2 3 2 p 3 冋 冉 冊册 1 ⌬p 1⫹ 2 g3 4. Determination of initial stiffness and plastic stiffness 4.1. Double web angle connection (all bolted) ⫺Mp (at point B) −Pp· g1− 1− 233 2 (16) 冉 冊冉 冊 冉 冊 1 4 0.5la ·q Pe·0.5la · ·0.5la ⫻ 2 3 ⌬e (18) where la: the length of angle Consequently, initial stiffness ki is given by 冉 冊 2EI g1+g2 3 ·l ki⫽ 3 g1 g1+4g2 a (19) 4.1.2. Plastic stiffness (kp) According to experimental results reported by Lewitt et al. [8], it can be seen that the plastification of angle takes place perfectly at ⌬p=5.08 mm (0.2 in) and the center of rotation has been assumed to be at 0.8 of the length of angle. This actually varied with the value of ⌬p, the center of rotation is conservatively determined to be at 0.8la in this paper. As the moment increases, the load distribution in the length of angle has a tendency to be concentrated on bolts [9]. Therefore it is assumed that the total tension force in connection angle is the sum of the product of the applied load(F1,…, Fn) and diameter of the nut(Dnut). Assuming the deformation of angle is proportional to the distance from the center of rotation Eliminating Hp from Eqs. (14) and (15) and then using the Pp–⌬p relationship it can be shown that Pp ⫽ 再 再 1 2Mp (g1−g3)(g23+⌬2p)+ g2⌬2p 2 冉 冊 冉 冎 1 1 g3(g1−g3) g23+ ⌬2p +g3⌬p g2⌬p−g1g3+g23 2 2 (17) 冊冎 Fig. 10. Force distribution of double web angle in elastic state. 234 S.-S. Lee, T.-S. Moon / Engineering Structures 24 (2002) 227–237 Fig. 11. Force distribution of double web angle in plastic state. as shown in Fig. 11, the relationship between the moment M and the rotation q is obtained as 冉冘 冉 冊 M⫽2⫻ ⫻ 冊冉 li ⌺l ·Pp·Dmit ⫻ ⫹0.1la 0.8l ⌺li a i⫽1 n 2 i 冊 (20) 0.8la ·q ⌬p where: li Dnut Distance from the center of rotation to the each center of fastener Diameter of the nut Finally, plastic stiffness kp is given by kp⫽ 4Mp兵(g1−g3)(g23+⌬2p)+0.5g2⌬2p其·Dmit ·(⌺l2i {g3(g1−g3)(g23+0.5⌬2p)+g3⌬p(0.5g2⌬p−g1g3+g23)} (21) 1 ⫹0.1la·⌺li)· ⌬p where ⌬p=5.08 mm because the dimension properties and material strength are available. The parameters used in each test and the comparison of analytical and experimental initial stiffness is listed in Table 1. As shown in Table 2, the prediction of plastic stiffness is relatively accurate, and hence a good relationship of M–q can be expected. Finally, substituting ki and kp calculated from Eqs. (19) and (21) into Eqs. (10a) and (10b), the shape parameters a and n can be easily obtained. Fig. 12 illustrates that the proposed model gives a good curve-fitting with test curve. 4.2. Top and seat angle connection (all riveted) 4.2.1. Initial stiffness (ki) Typical top and seat angle connection is shown in Fig. 1(b). Based on previous assumptions that were made for analyzing behavior of angles, the initial stiffness can be obtained from simple elastic theory. The top angle provides resistance to the bending moment, in which the fixed support is assumed to be in the vicinity of the rivethead edge in the leg adjacent to the column. Because the clamping force of the rivet is lower than that of the bolt, it is assumed that the effective length (diameter of rivet-head×number of rivet) can resist the bending moment. The center of rotation for the connection is located at the leg adjacent to the compression-beam flange at the end of the beam as shown in Fig. 13 [10]. Similarly, using these assumptions and the Pe–⌬e relationship of angle segment in Eq. (13), the relationship between the moment M and the rotation q is obtained as M⫽(Pe·le)⫻(db⫹g1)⫻ 4.1.3. Comparison of analytical and experimental results For the purpose of comparison of the analytical ki and kp with the ki and kp measured from a test curve, the six test data from SCDB are employed [3]. One bolted test conducted by Lewitt, one bolted test conducted by Bose and five bolted tests reported by Thompson are selected 冉冊 db ·q ⌬e (22) where: le db Effective length of top angle. Depth of beam Table 1 Comparison of analytic and curve-fitting initial stiffness (double web angle connection) Test ID g1 (mm) g2 (mm) g3 (mm) ta (mm) Dnut (mm) la (mm) E (N/mm2) ki(anal.) (kN·m/rad) ki(fitting) (kN·m/rad) ki(anal.)/ki (fitting) FK4AB-M B-1 A1-2 ALT B1-2 ALT D1-2 ALT E1-3 ALT A2-3 ALT 38.32 52.95 44.93 42.55 25.88 39.37 44.93 84.14 82.50 37.31 34.93 37.31 34.93 37.31 20.83 31.87 29.85 25.08 10.80 21.91 29.85 9.53 15.00 7.94 12.70 7.94 12.70 7.94 31.75 31.75 31.75 31.75 31.75 31.75 31.75 292.10 400.00 292.10 292.10 292.10 292.10 292.10 2.06E+05 2.06E+05 2.06E+08 2.06E+05 2.06E+05 2.06E+05 2.06E+08 4293.03 17669.13 1996.21 9672.58 8909.51 11913.81 1998.26 5503.21 17349.86 2002.74 9768.32 8571.32 11563.58 2014.17 0.78 1.02 0.99 0.99 1.04 1.03 0.99 S.-S. Lee, T.-S. Moon / Engineering Structures 24 (2002) 227–237 235 Table 2 Comparison of analytic and curve-fitting plastic stiffness (double web angle connection) Test ID g1 (mm) g2 (mm) g3 (mm) Number db (mm) of bolt sy (N/mm2) ⌬p (mm) kp(anal.) (kN·m/rad) kp(fitting) (kN·m/rad) kp(anal.)/ kp(fitting) FK4AB-M B-1 A1-2 ALT B1-2 ALT D1-2 ALT E1-3 ALT A2-3 ALT 57.37 37.07 76.68 74.30 57.63 55.25 76.68 36.51 36.62 37.31 34.93 37.31 34.93 37.31 39.88 16.00 61.60 56.83 42.55 37.78 61.60 4 5 4 4 4 4 4 286.3644 344.2257 247.8415 247.8415 247.8415 247.8415 344.2257 327.05 6496.30 120.97 338.58 182.42 535.40 168.13 300.63 5843.02 129.29 322.51 174.56 482.83 172.98 1.09 1.11 0.94 1.05 1.05 1.11 0.97 456.95 456.95 533.15 533.15 533.15 533.15 533.15 5.08 5.08 5.08 5.08 5.08 5.08 5.08 Fig. 13. Force distribution of top and seat angle in elastic state. Consequently, initial stiffness ki is given by 冉 冊 12EI g1+g2 ·l ·(d ⫹g ) ki⫽ 3 g1 g1+4g2 e b 1 Fig. 12. Comparison of predicted results with experimental results (DWA connection). (23) 4.2.2. Plastic stiffness (kp) As the moment increases, the plastic hinges are formed at the critical location of top angle. In cases the angle is thick or the g3 is rather short compared to the part of fillet, the plastic hinge of point A in Fig. 8 may be developed at the rivet-head edge near the toe of the angle. Therefore, it is necessary that the values of g1 and Table 3 Comparison of analytic and curve-fitting initial stiffness (top and seat angle connection) Test ID g1 (mm) g2 (mm) g3 (mm) ta (mm) Hrivet (mm) Number le (mm) of rivet db (mm) ki(anal.) (kN·m/rad) ki(fitting) (kN·m/rad) ki(anal.)/ ki(fitting) No.2 No.5 No.9 No.10 No.20 No.22 No.24 41.24 42.83 41.24 39.65 41.24 41.24 38.91 58.76 60.35 58.76 57.18 58.76 58.76 60.31 20.60 23.78 20.60 17.43 20.60 20.60 18.28 15.88 12.70 15.88 19.05 15.88 15.88 15.88 28.65 28.65 28.65 28.65 28.65 28.65 33.30 2 2 2 2 2 2 2 310.39 458.72 458.72 458.72 355.09 406.65 458.72 25297.85 26572.30 53669.10 102450.24 34279.93 44229.51 76343.07 1.05 0.97 1.04 1.05 1.00 1.00 0.99 57.30 57.30 57.30 57.30 57.30 57.30 66.60 26594.90 25704.32 55885.75 107958.84 34293.39 44382.42 75297.11 236 S.-S. Lee, T.-S. Moon / Engineering Structures 24 (2002) 227–237 meters a and n can be easily obtained. The proposed model gives a good curve-fitting with test curve as shown in Fig. 15. 5. Summary and conclusions In this paper, a practical model is proposed to represent the moment–rotation relationship of semi-rigid connection and the proposed model is simple to use and accurately describes the moment–rotation behavior of Fig. 14. Force distribution of top and seat angle in plastic state. g3 are adequately modified. In plastic state in Fig. 14, the center of rotation keeps location in initial elastic state. Again based on the Pp–⌬p relationship of angle segment in Eq. (17), the relationship between the moment M and the rotation q is derived as M⫽(Pp·le)⫻(db⫹g1)⫻ 冉冊 db ·q ⌬e (24) Finally, plastic stiffness kp is given by kp⫽ 2Mp{(g1−g3)(g23+⌬2p)+0.5g2⌬2p} db ·l (d ⫹g )· {g3(g1−g3)(g23+0.5⌬2p)+g3⌬p(0.5g2⌬p−g1g3+g23)} e b 1 ⌬p (25) where ⌬p=5.08 mm 4.2.3. Comparison of analytical and experimental results To compare the analytical ki and kp with the ki and kp measured from a test curve, 12 riveted tests conducted by Hechtman are selected in the present comparison because the dimension properties and material strength can be available from SCDB [3]. Similarly, the parameters used in each test and the comparison of analytical and curve-fitting ki are listed in Table 3. The analytical plastic stiffness agrees in general with the experimental values as shown in Table 4. Eventually, substituting ki and kp calculated from Eqs. (23) and (25) into Eqs. (11a) and (11b), the shape para- Fig. 15. Comparison of predicted results with experimental results (TSA connection). Table 4 Comparison of analytic and curve-fitting plastic stiffness (top and seat angle connection) Test ID g1 (mm) g2 (mm) g3 (mm) ta (mm) le (mm) db (mm) sy ⌬p (mm) kp(anal.) (kN·m/rad) (N/mm2) kp(fitting) (kN·m/rad) kp(anal.)/ kp(fitting) No.2 No.5 No.9 No.10 No.20 No.22 No.24 65.11 66.70 65.11 63.53 65.11 65.11 66.66 39.66 41.25 39.66 38.08 39.66 39.66 38.11 44.48 47.65 44.48 41.30 44.48 44.48 46.03 15.88 12.70 15.88 19.05 15.88 15.88 15.88 57.30 57.30 57.30 57.30 57.30 57.30 66.60 310.39 458.72 458.72 458.72 355.09 406.65 458.72 258.22 258.22 258.22 258.22 258.22 258.22 258.22 1075.08 1417.89 2269.88 3076.13 1462.23 1816.87 2443.18 1.01 0.94 0.99 1.14 0.95 0.99 1.03 5.08 5.08 5.08 5.08 5.08 5.08 5.08 1087.37 1331.25 2241.86 3502.24 1392.07 1789.83 2514.84 S.-S. Lee, T.-S. Moon / Engineering Structures 24 (2002) 227–237 nearly all connections by controlling shape parameter a and n. It introduces a semi-analytic procedure that the optimum values of a and n, agreeing rather well with experimental curves, are described as a function of initial stiffness (ki) and plastic stiffness (kp) from statistical regression analysis. Herein, the initial stiffness (ki) and plastic stiffness (kp) of double web angle connection and top and seat angle connection are derived analytically and the analytical results agree in general with the experimental values. It is difficult to apply Eqs. (10a, 10b, 11a) and (11b) to all angles used in practice because not all the geometry and material properties of angles are considered in this study. However, the proposed model and semi-analytical procedure will be a very effective tool for designers to execute the frame analysis with semi-rigid connections. Acknowledgements The financial assistance of STRESS (advanced STructure RESearch Station) of KOSEF(KOrea Science and Engineering Foundation), Hanyang University is appreciated. 237 References [1] Attiogbe E, Morris G. Moment–rotation functions for steel connections. J Struct Engng ASCE 1991;117(6):1703–18. [2] Kishi N, Chen WF. Moment–rotation relations of semi-rigid connections with angles. J Struct Engng ASCE 1990;116(7):1813– 34. [3] Chen WF, Toma S. Advanced analysis of steel frames. Boca Raton (FL): CRC Press, 1994. [4] Yee YE, Melchers RE. Moment–rotation curves for bolted connections. J Struct Engng ASCE 1986;112(3):615–35. [5] Colson A, Louveau JM. Connections incidence on the inelastic behavior of steel structures. Euromech Colloquium 1983;174(Oct). [6] Fleischman RB, Chasten CP, Lu LW, Driscoll GC. Top and seat angle connection and end-plate connections: snug vs. fully pretensioned bolts. Eng J AISC 1991;First quarter:18–28. [7] Wu FH, Chen WF. A design model for semi-rigid connections. Eng Struct 1990;12;12:88–97. [8] Lewitt CW, Chesson E Jr, Munse W. Restraint characteristics of flexible riveted and bolted beam to column connections. Bulletin No.500, Urbana: Engineering Experiment Station, University of Illinois, January, 1969. [9] Sherbourne AN, Bahaari MR. Finite element prediction of end plate bolted connection behavior. II: analytic formulation, J Struct Eng ASCE 1997;123(2):269–81. [10] Chen WF, Lui EM. Stability design of steel frames. Boca Raton (FL): CRC Press, 1991.