Dr. Sami Kilic Bogazici University. Istanbul, Turkey

Transcription

1 Dr. Sami Kilic Bogazici University Dept. of Civil Engineering Istanbul, Turkey

2 Outline Observations about tank damage after the 17 August 1999 Turkish earthquake (Marmara event). Finite element modeling techniques for cylindrical containment tanks with fixed roofs. Seismic simulation of an actual tank design subjected to the recorded ground motion of the 17 August 1999 event. Visualization of tank wall deformations and sloshing waves induced in the tank.

3 Tank Damage during the 17 August 1999 Marmara Earthquake Mw=7.4 earthquake occurred near a heavily concentrated industrial zone serving Istanbul and the greater area killing 17,000 people and leaving half a million people homeless. Floating-roof tanks containing liquid naphta caught fire. Tank damage was also observed for fixed-roof tanks. Tanks were designed per American Petroleum Institute Standard API-650. Advanced finite element modeling techniques allows us to investigate the sloshing problem in more detail compared to the simplified analysis methods presented in the design codes.

4 Floating Roof Naphtha a Tanks (undamaged..) aged

5 Floating Roof Naphtha a Tanks (damaged!) aged

6 Tank Farm Damage age on 17 August 1999

7 Cooling of the Combustible Contents for the Risk of a Secondary Fire

8 Roof Sloshing Damage for Large Diameter Petroleum Tanks - I

9 Roof Damage age due to Sloshing Effects -II

10 Roof Shell Buckling due to Sloshing Effects -I

11 Roof Shell Tearing due to Sloshing Effects -II

12 Garden Variety of Tank Damages Observed Sloshing Roof Damage Tear/rupture of the Roof Shell Elephant Foot Buckling of shell (from literature) Leakage & Fire

13 Seismic Se s c Issues ssues Related e ated to Tank a Performance e o a ce Cylindrical tanks are used to store a variety of liquids Satisfactory performance of tanks during an earthquake is crucial for modern facilities. Inadequately designed tanks have suffered extensive damage during past earthquakes. 99 Turkey, 95 Kobe, 94 L Los A Angeles, l 89 San Francisco,

14 API 650 Design Standard d Impact of the sloshing of the liquid on the tank roof and the tearing of the shell are not directly considered. Nonlinear effects near the foundation such as pressure accumulation causing elephant foot bucking are not taken into account. Furthermore, uplifting of the foundation for un-anchored tanks is not considered. Baffle design to reduce the sloshing effect is not present in the code. Base uplift plan view of the tank bottom

15 Seismic Design Approach of the API 650 Standard d The liquid at the bottom of the tank is considered as the impulsive mass (high frequency response liquid goes through rigid body motion) and the liquid close to the surface is considered as the convective mass (low frequency response). A site-specific response spectrum is used to determine the corresponding spectral accelerations per American Society of Civil Engineers ASCE 7 Code. Combined with the impulsive and convective masses, the corresponding seismic forces are determined. Non-linear dynamic analysis methods incorporating fluid-structure and soilstructure interaction effects are permitted to be used.

16 Impulsive and Convective e Masses Top part of the liquid moves in long-period sloshing motion which is called the convective response. Rest moves rigidly with the tank wall which is called the impulsive response. API 650 theory dates back to Wozniak and Mitchell (1978)!!

17 Illustration of Liquid Masses

18 Illustration of Various Cylindrical Mode Shapes Using Mathematical Software Tools 1st Mode 2nd Mode 3rd Mode 4th Mode

19 Modal Analysis Using Finite Element Tools MSC MD.Nastran SOL 103 (Eigensolver)

20 Seismic Tank Sloshing -Analysis s Approach MD Nastran SOL700 with Fluid Structure Interaction (FSI) used in analysis Velocity time history applied to the tank base Rigid tank model feasibility study to run 50 sec earthquake event Flexible tank model partial results In this study, the tank is considered to be anchored to the ground. Un-anchored tank model needs additional contact t modeling with the ground.

21 Properties of the Tank Design Under Investigation Tank diameter = 21 m. Tank wall height = 15.8 m. Tank spherical dome drop height = 2.67 m. Side wall thickness varies between 20 mm to 10 mm from bottom to top p( (except for the top pp perimeter seam around the roof; t=35 mm). Tank shell material specifications: ASTM A537 Fy=345 MPa (side wall) and ASTM A283 Fy=205 MPa (roof). An actual built tank that is in service; design blueprints courtesy of Tekfen Engineering & Manufacturing Ltd., Istanbul, Turkey. Cooperating engineers: Mr. Ediz Sezginel and Mr. Koray Unalan.

22 Lagrangian a g a model cont d Elasto-plastic MATD024 material model used for shell elements: E = 200 GPa Poisson s ratio = 0.29 Density = Kg/m^3 Yield stress = 345 MPa (side wall) and 205 MPa (roof) ETAN = 800 MPa Failure strain = 0.22

23 Lagrangian a g a model Approx. 3,000 nodes and 3,000 shell elements for the Lagrangian tank enclosure. Belytschko-Wong-Chiang shell formulation. Wall thickness variation is shown by different colors (10-20 mm).

24 Fluid udstuctue Structure Interaction teacto in MD.Nastran a SOL700 MD SOL700 features two solving techniques: Lagrangian methods for structures based on LS-Dyna Elements of constant mass Mass move with grid points Eulerian methods for Fluids based on MSC Dytran Elements of constant volume Material move across elements

25 General Coupling and Fluid Structure Interaction ti Fluid interacts with structure through coupling surfaces Coupling surfaces can have an arbitrary shape and may comprise faces of Lagrangian elements As these Lagrangian elements deform, so will the coupling surfaces The coupling surface acts as boundary to the Euler mesh The Euler mesh applies load to the structure as the materials flow Coupling surface intersects Euler mesh to establish control volume

26 Eulerian ue model Two Euler domains are needed, one for the fluid inside the tank (FV6) and the second domain for the air outside (FV7). Elements in Euler domains are solid HEX (approx. 100,000 elements per domain). As shell elements in coupling surface fail, Euler domains interact t with each other as defined by COUPINT entry this allows Euler material from one domain to move to the next domain. Coupling surface failure is activated by DYPARAM, FASTCOUP,, FAIL Coupling surface Euler domain

27 Eulerian ue model cont d Euler domains initialized with water and air inside the tank(fv6) and only air outside (FV7). Fluid inside of tank is also initialized with hydrostatic pressure (HYDSTAT). FV6 and FV7 are just big enough so that t the tank stays inside id of the domain during the earthquake. Convective periods; T1=4.85 sec., T2=2.82 sec., T3=1.18 sec. Euler inside tank initialized with water

28 Velocity time history applied at tank base X direction (East-West) t) X vel (m/sec) Recorded YPT station ground motion on 17 August 1999 located within 2 kilometers of the Tupras Refinery site. -1 Xvel Note: The hump in the long period range of the acceleration response spectrum range is not covered by the design code.

29 Velocity time history applied at tank base Y direction (North-South) th) Y vel (m/sec) Yvel

30 Velocity time history applied at tank base Z direction (Up-Down) 0.4 Z vel (m/sec) Zvel

31 Rigid gdtank Results esuts Max pressure at tank roof = 2.58 MPa

32 Rigid gdtank Results esuts Max pressure at tank bottom = 1.73 MPa

33 Rigid gdtank Results esuts Envelope plot of fluid free surface

34 Flexible be Tank Results esuts Not able to run full 50 sec transient After tank roof fails, water inside tank spills out, clumps start controlling time step and time step becomes too small (in the order of 1.E-10), resulting in job termination Job terminated t at time 10.6 sec

35 Flexible be Tank Results esuts Maximum pressure at tank = 4.46 MPa

36 Flexible be Tank Results esuts Tank deformation at time 10.6 sec

37 Flexible be Tank Results esuts Fluid at time 10.6 sec water starts spilling out to outside Euler domain but job terminates due to small time step

38 Flexible be Tank Results esuts Side by side view

39 Flexible be Tank Results esuts Side by side view

40 Scalability ab Study (0.5 sec transient t run) 1 CPU 5 CPU s 16 CPU s 25 CPU s CPU time (sec) 9,961 2,586 1,962 2,529 Wall time (sec) 11,746 5,363 2,758 2,965 14,000 12,000 10,000 8,000 6,000 CPU time (sec) Wall time (sec) 4,000 2,

41 CONCLUSIONS C O S Running a Rigid Tank case gave us good qualitative results and was a check that the model/bc s were properly defined. Good scaling when Shared Memory Parallel l (SMP) processing used in the runs (up to 16 CPU s). [We can harness the power of parallel computers]. The rigid tank model shows that the peak pressure applied by the fluid on the tank correlates quite well with damages seen in the tanks after the earthquake. Flexible tank model needs a little more fine tuning so it can continue the run after the tank wall fails and spilling of the liquid content occurs. Additional contact modeling with the ground is needed to investigate free- standing un-anchored tanks. Possible with MD Nastran SOL 700. Tank wall baffle designs to reduce the sloshing effects can also be investigated. Buoyancy effects are needed to incorporate floating roof tanks. Possible with MD Nastran SOL 700. Nonlinear dynamic finite element simulation can be incorporated into the design scene.

42 Acknowledgements e Mr. Koray Unalan and Mr. Ediz Sezginel from Tekfen Engineering and Construction of Istanbul, Turkey for providing the 21m diameter fixed roof sample tank design. Bogazici i University it Research Fund contract t no. 07HT102 for providing the core funding.