COMMISSION OF THE EUROPEAN COMMUNITIES FP7- INFRASTRUCTURES-2008-1 SP4-Capacities SERIES SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES Geographically distributed continuous hybrid simulation tests using shaking tables Ferran Obón Santacana Uwe E. Dorka Universität Kassel (UNIKA) SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES Overview 1. Dorka’s substructure algorithm 2. Adapting the algorithm 3. Test set-up and numerical models. 4. Evaluation of the control of the actuators 5. Tests using OpenFresco. 6. Tests using NSEP protocol. 7. Comparison of protocols. 8. Numerical tests with Celestina 9. Large numerical models 10. Conclusions SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 1. Dorka’s substructure algorithm Basics of substructure testing: •Substructure testing is a numerical time integration process including restoring forces measured as response from one or several substructures. •No iterative solver can be used in this process. •A method is required with unconditional numerical stability and high accuracy. SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 1. Dorka’s substructure algorithm Any time integration scheme can be derived from time discretisation: Dynamic equilibrium: With M d2 dt 2 u(t ) C dtd u(t ) Ku (t ) f l (t ) f r (t ) f c (t ) fr: vector of numerical nonlinear restoring forces fc : vector of restoring forces measured at the interface between numerical model and tested substructure Shape functions to discretise the displacements over 3 steps in time: N i 1 1 / 2 N i 1 1 N i 1 1 / 2 with: t / t SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 1. Dorka’s substructure algorithm and using a weighted residual approach: weighting function W: d M u i 1 d 2 N i 1 u i d 2 N i u i 1 d 2 N i 1 dt dt dt 1 C u i 1 dtd N i 1 u i dtd N i u i 1 dtd N i 1 1W K u i 1 N u i N u i 1 N i 1 n i 1 i 1 i i 1 f* N i 1 f* N i f* N i 1 2 2 f* f l f c f r 1 W 12 d 1 1 W 1 2 1 Wd ; 1 1 1 2 1 d Wd ; 1 SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 1. Dorka’s substructure algorithm leads to a 3-point recurrence scheme: u i 1 M tC t 2 K 1 2 M 1 2 tC 12 2 t 2 K u i 2 i 1 1 M 1 tC 2 t K u 2 i 1 2 i 1 t f* 2 2 t f* 1 2 i 1 t f * 2 with: u0 i 1 M tC t K 2 1 u i 1 u0i 1 G f ri 1 f ci 1 2 M 1 2 tC 12 2 t 2 K u i 2 i 1 1 M 1 tC 2 t K u 2 i 1 2 i 1 t f l 2 2 t f* 1 2 i 1 t f * 2 Every time integration algorithm is a linear control equation within each time step that has an initial value u0 and a constant gain G: G t 2 M tC t 2 K 1 SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 1. Dorka’s Substructure Algorithm Performance of various time integration schemes: Period elongation Numerical damping Any of those can be used in Dorka algorithm but the unconditionally stable Newmark scheme without numerical damping yields the best results! SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 1. Dorka’s substructure algorithm • • Sub-stepping avoids iteration within the time step and allows the actuators to run continuously. If any equilibrium error occurs at the end of the time step it can be compensated. u i 1 u0i 1 G f ri 1 f c i 1 new load and error compensation: f*i f l i f ei f ri f ci initialization with new u0 i 1 j =1 apply displacement at each sub-step: j i i 1 j u u0 (1 ) u0 ( ) G f r f c k k calculate fr measure f s i=i+1 yes: j = j + 1 j<k error force compensation no time derivatives of shape functions give: i 1 d i 1 d2 ; dt 2 u dt u error force: fu i 1 (M fl d2 dt 2 i 1 f ri 1 f ci 1 u i 1 C d dt u i 1 Ku i 1 ) SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 2. Adapting the algorithm SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 2. Adapting the algorithm • u i 1 u0i 1 G f ri 1 f c i 1 • • • • Only a portion of the G matrix has to be exchanged between facilities. The communication is performed at the step level. If there are multiple facilities with several sub-structures the communication must be done at the sub-step level if there is a strong influence between the coupling nodes. The network speed is unreliable. Special treatment of the actuators has been considered: • • The actuators will continue with the ramp function at the end of the sub-step. The actuators will stop in case data is not received for several steps. SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 3. Test set-up and numerical modelling • Experiments performed: • • • Protocols: • • Model used as a benchmark of a steel frame. Evaluation of the control of the actuators. Purely numerical with empty cylinders. Sub-structure test with a single Degree of Freedom (DOF) and the UHYDEfbr device. OpenFresco. Platform for Networked Structural Experiment (PNSE). Model used for testing. SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 3. Test set-up and numerical modelling Fixed TMD. The friction device was active and with the springs they provided the necessary force to stabilise the test. There were no velocity dependant forces. • • • Coupling Force. Pressure drop (0.25 bar to 0.1 bar) 2000 1500 Shaking table with UHYDE-fbr and fixed TMD. • The UHYDE-fbr operates through solid friction. The hysteresis loops can be varied by changing the air pressure. Coupling Force [N] • UHYDE-fbr [9] 1000 500 0 -500 -1000 -1500 -2000 -2500 -0.03 -0.02 -0.01 0 Displacement [m] 0.01 0.02 0.03 SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 4. Evaluation of the control of the actuators Accuracy of the actuators 20 Control value Measured 15 Continuous tests performed with HYSTEC using OpenFresco and an empty cylinder. Scale factor of 200. 10 0 -5 -10 -15 -20 Accuracy of the actuators 2 -25 -30 0 2000 4000 6000 8000 10000 1 Number of substeps Due to reduced P the actuators present error in the positioning. Phase lag is important Displacement [mm] Displacement [mm] 5 12000 14000 Control value Measured 16000 0 -1 -2 -3 1 1.1 1.2 1.3 Number of substeps 1.4 1.5 1.6 4 x 10 SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 5. Tests using OpenFresco • • • • • Adwin Pro II Another experimental control facility was designed by Andreas Schellenberg as a request from UNIKA. It does not assume any controller and bridges the communication between OpenFresco and the control hardware using predefined messages over TCP/IP. The speed limit with University of Oxford was 200 ms/step (scale factor of 20, RTT = 0.036 s). The speed limit with U. California, Berkeley, was 1 s/step (scale factor of 100, RTT = 0.2 s) Although the communication stopped at one point, there were no effects on the test. SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 5. Tests using OpenFresco Displacement of the cylinder. Constant Pressure 0.25 bar. Displacement of the cylinder. Pressure drop (0.25 bar to 0.1 bar) 0.03 0.03 Measured displacement Control displacement (Numerical value) 0.02 0.02 0.01 0.01 Displacement [m] Displacement [m] Measured displacement Control displacement (Numerical value) 0 0 -0.01 -0.01 -0.02 -0.02 -0.03 0 1000 2000 3000 4000 Number of substeps 5000 6000 -0.03 7000 0 1000 1500 1500 1000 1000 500 500 Coupling Force [N] Coupling force [N] 2000 0 -500 -1500 -1500 -2000 -2000 0 Displacement [m] 0.01 6000 7000 -500 -1000 -0.01 5000 0 -1000 -0.02 3000 4000 Number of substeps Coupling Force. Pressure drop (0.25 bar to 0.1 bar) Coupling force. Constant pressure 0.25 bar. 2000 -2500 -0.03 2000 0.02 0.03 -2500 -0.03 -0.02 -0.01 0 Displacement [m] 0.01 0.02 0.03 SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 6. Tests using NSEP protocol • The same test was performed for comparison. • The new version of the NSEP protocol (NSEPv2) was used. • It required single-precision for exchanging data. • The time step was always 2s without considering the proximity of the facility. • This could be due to limitations in the NSEP server running under Windows 7. SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 6. Tests using NSEP protocol Displacement of the cylinder. Pressure drop (0.25 bar to 0.1 bar) Displacement offf the cylinder. Constant pressure 0.25 bar. 0.03 0.03 Measured displacement Control displacement (Numerical value) 0.02 0.02 0.01 0.01 Displacement [m] Displacement [m] Measured displacement Control displacement (Numerical value) 0 0 -0.01 -0.01 -0.02 -0.02 -0.03 -0.03 0 1000 2000 3000 4000 Number of substeps 5000 6000 7000 0 1000 2000 1500 1500 1000 1000 500 500 0 -500 -1500 -1500 -0.01 0 Displacement [m] 0.01 6000 7000 -500 -1000 -0.02 5000 0 -1000 -2000 -0.03 3000 4000 Number of substeps Couplint Force. Pressure drop (0.25 bar to 0.1 bar) 2000 Coupling Force [N] Coupling Force [N] Coupling Force. Constant pressure 0.25 bar. 2000 0.02 0.03 -2000 -0.03 -0.02 -0.01 0 Displacement [m] 0.01 0.02 0.03 SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 7. Comparison of the protocols: OpenFresco vs. NSEPv2 Comparison of protocols 0.03 OpenFresco NSEPv2 0.02 Displacement [m] 0.01 0 -0.01 -0.02 -0.03 0 1000 2000 3000 4000 Number of substeps 5000 6000 7000 SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 8. Numerical tests with Celestina SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 9. Large numerical models • • • • Large numerical models where tested. The UHYDE-fbr was used as a non-linear substructure. 256 cores where used in the Linux Cluster. Intel MKL libraries 4 cores where used in the regular desktop (Core i5). OpenBLAS libraries / With optimizations. Size Complex models (METU) 1000 DOFs Time per iteration Regular Desktop Linux Cluster 0.7 ms - 10000 DOFs 200 ms /43 ms 87 ms 50000 DOFs Not possible 450 ms SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES 10. Conclusions Dorka’s sub-structure algorithm allows performing geographically continuous distributed testing with different protools: • The tests could be performed with a time scale factor of 100 with U. California, Berkeley (USA), 20 with U. Oxford (UK) and 200 with HYSTEC (S. Korea) using OpenFresco A time scale factor of 200 using PNSE with NCREE (Taiwan) and U. Oxford (UK). Celestina (Purely numerical) • • • Since the actuators run continuously updating must be performed between no-more than three sub-steps. As a rule of thumb, the speed of the test (step time) can be set to: • • Step Time • 0.8 Special attention has to be paid when considering multiple sub-structures that are not located at the same laboratory. • • • nSubsteps * Average Round Trip Time The communication has to be performed at the sub-step level when there is a strong influence between the sub-structures. Multi-protocol, multi-site testing. This is the next step. The algorithm allows performing sub-structure tests with large numerical models in combination of supercomputers/regular desktop computers (10.000 DOF in 0.087 s/step or 43 ms/step with optimizations). SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES Acknowledgements Special thanks go to: 1. University of California, Berkeley: Stephen A. Mahin, Andreas Schellenberg and Selim Gunay. 2. HYSTEC: Chul-Young Kim and Dae-Sung Jung. 3. NCREE: Keh-Chyuan Tsai and Kung-Juin Wang. 4. University of Oxford: Tony Blakeborough and Ignacio Lamata. SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR EUROPEAN SYNERGIES Thank you for your attention!