1. 1 지진하중을 받는 구조물의 MR 댐퍼의 동특성을 고려한 반능동 신경망제어 Heon-Jae Lee 1), Hyung-Jo Jung 2), Ju-Won Oh 3), In-Won Lee 4) 1) Graduate Student, Dept. of Civil and Environmental Engineering, KAIST 2) Assistant Professor, Dept. of Civil Engineering, Sejong University 3) Professor, Department of Civil Engineering, Hannam University 4) Professor, Dept. of Civil and Environmental Engineering, KAIST 2003 년도 대한토목학회 학술발표회

2. 2 Introduction  Semi-active control strategy has attractive feature such as - bounded input and bounded output stability - small energy requirement - not only the reliability of passive control but also the adaptability of fully active control  Neural network also has attractive feature such as - ability of producing continuous control signal - adaptability for non-linear problems - no need the mathematical model for solving any engineering problems

3. 3 Proposed Method STRUCTURE Neural Network MR Damper Clipped Algorithm Clipped Neuro-Control Block diagram of the proposed method

4. 4 Clipped Algorithm desired force (by neural network) : generated force (by MR damper) :

5. 5 Numerical Example velocity of 3 rd floor ground acceleration control force displacement of 3 rd floor velocity of 1 st floor displacement of 1 st floor Example Structure ;Dyke el al. (1996) Neural Netwokr used in this study El Centro (0.348g) Kobe (0.334g) California (0.156g) Training Verification

6. 6 0246 -0.5 0 0.5 1 0246 -0.5 0 0.5 1 Time(sec) Displacement (cm) Displacement of the 3 rd floor (El Centro) Training Results Clipped optimal Proposed method : uncontrolled : controlled

7. 7 Acceleration of the 1 st floor (El Centro) Training Results Clipped optimal Proposed method 0246 -1000 -500 0 500 1000 Time(sec) Acceleration (cm/sec 2 ) Acceleration (cm/sec 2 ) 0246 -1000 -500 0 500 1000 : uncontrolled : controlled

8. 8 Control Results Control Strategy Uncontrolled Clipped optimal Proposed method 0.5410.115 (1.000)0.117 (1.017) 0.8250.186 (1.000)0.171 (0.919) 0.9630.236 (1.000)0.240 (1.017) 0.5410.115 (1.000)0.117 (1.017) 0.3200.090 (1.000)0.111 (1.233) 0.2010.101 (1.000)0.082 (0.812) 861733 (1.000)429 (0.585) 1040746 (1.000)503 (0.674) 1401705 (1.000)571 (0.810) -954 (1.000)852 (0.893) Maximum responses under El Centro earthquake

9. 9 Control Results Control Strategy Uncontrolled Clipped optimal Proposed method 0.3410.054 (1.000)0.043 (0.796) 0.4800.094 (1.000)0.071 (0.755) 0.5760.114 (1.000)0.100 (0.877) 0.3410.054 (1.000)0.043 (0.796) 0.1790.041 (1.000)0.048 (1.171) 0.1070.041 (1.000)0.039 (0.951) 570389 (1.000)181 (0.465) 653285 (1.000)250 (0.877) 744285 (1.000)273 (0.958) -411 (1.000)319 (0.776) Maximum responses under California earthquake

10. 10 Control Results Control Strategy Uncontrolled Clipped optimal Proposed method 1.4860.700 (1.000)0.480 (0.686) 2.3731.181 (1.000)0.819 (0.693) 2.8771.444 (1.000)1.013 (0.702) 1.4860.700 (1.000)0.480 (0.686) 0.8950.487 (1.000)0.373 (0.766) 0.5280.328 (1.000)0.208 (0.634) 23371936 (1.000)940 (0.486) 27931929 (1.000)1438 (0.745) 36762280 (1.000)1445 (0.634) -1324 (1.000)1513 (1.143) Maximum responses under Kobe earthquake

11. 11 00.20.40.60.8 0.3 0.4 0.5 0.6 Peak Ground Acceleration (g) Maximum drift of 3 rd floor Clipped optimal Proposed method Kobe earthquake California earthquake El Centro earthquake Control Results Maximum drift of the 3 rd floor

12. 12 Peak Ground Acceleration (g) Clipped optimal Proposed method Kobe earthquake California earthquake El Centro earthquake Control Results Maximum acceleration of the 1 st floor Maximum acceleration of 1 st floor 00.20.40.60.8 0 0.4 0.8 1.2 1.6

13. 13 Conclusions A semi-active neuro-control technique using MR damper for seismically excited structure is proposed. The performance of the proposed method is better than that of the clipped optimal control method. ( max. drift of 3 rd floor : 5 ~ 34 % reduction, max. acceleration of 1 st floor : 37 ~ 69 % reduction ) The proposed semiactive neuro-control technique using MR dampers could be effectively used for control of seismically excited structures !