September 11 Pentagon Attack Simulations
Using LS-Dyna

Phase I, Completed September 11, 2002


Mete A. Sozen, Sami A. Kilic  and  Christoph M. Hoffmann


The Larger Team and Responsibilities


Project Conception Mete Sozen, Civil Engineering
Simulation Setups Sami Kilic, Civil Engineering

Infrastructure Support

James Bottum, Information Technology at Purdue
Ahmed Sameh, Computer Science, Computing Research Institute

Project Direction Christoph Hoffmann, Computer Science, Computing Research Institute
Mesh Generation Christoph Hoffmann, Computer Science, Computing Research Institute
Engineering Models Sami Kilic, Civil Engineering

                                 Scientific Supervision

Mete Sozen, Civil Engineering

Supercomputer Runs William Whitson, Information Technology at Purdue


Voicu Popescu, Computer Science

Story Board

Scott Meador, Computer Graphics Technology

Graduate Students Amit Chourasia, Computer Graphics Technology
Hendry Lim, Computer Science



If any good can come from the events of September 11, it would be to understand in detail what damage occurred, how it occurred, and why it occurred. Then, we should be able to create superior structures that will protect life.  By simulating the chilling sequence of events, in this detail, we are able to fashion tools that help decision makers in the future to explore potential disasters before they happen.

Problem Statement

Simulate as faithfully as possible the effects of crashing an air frame loaded with fuel (simulating a Boeing 757) into a reinforced concrete frame similar to the one supporting the Pentagon building.  In particular, model the columns to have properties reproducing the behavior of spirally reinforced columns including the difference in material response of the concrete within and outside the spiral reinforcement.

Purpose of the Effort

Use the physically correct simulation results as input to animations and visualizations to produce a vivid reenactment of the impact of the aircraft on the Pentagon building and provide the larger team with the necessary data to construct these using 3D Studio Max, AutoCAD, and research tools.

Problem Formulations

Several problem formulations were investigated with the results sketched by the animations below.  A basic hypothesis, informally confirmed with engineers knowledgeable in this subject, is that the bulk of the impact damage is due to the body of fuel in the wing and center tanks.  Most of the aircraft structure is light-weight low-mass, and relatively low strength, with the exception of the wheel undercarriage.  The experiments are

  1. A single body of fluid hits a single column.  The purpose of this simulation is to understand the response of a reinforced concrete column subjected to high-speed impact of the fuel in the aircraft tanks.  In particular, the relationship between the impact velocity of the fluid and the acceleration of the column has been studied, as have different models of concrete columns.
  2. The fuel tank hits three rows of columns.  The wing enclosure breaks open and the fluid spills.  Wings are modeled without ribs, leading to a balloon effect.

  3. The fuel fluid, shaped by the wing tank, hits the columns and destroys those in the first two rows.  This simulation does not include a wing enclosure and so can be used to better understand the effect of the wing strength on the fuel dispersal.

  4. The right-wing fuel tank hits the first three rows of columns.  The fuel is modeled as fluid, and the problem is a mixed arbitrary Euler-Lagrange mesh formulation.  The fuel tank disintegrates, and the fuel disperses into the structure.  This time, the wing has ribs and the break-up is realistic.

  5. Full model run, coarse model formulation, run on the IBM Regatta.  Model includes all columns in the simulated frame and the complete aircraft.  The model size is approximately 300,000 nodes.  The run took about 24 hours for 40 frames covering 0.2 sec real time.

  6. Full model run, detailed model formulation, run on the IBM Regatta.  Model has 1,000,000 nodes. With 50 frames computed in close to 68 hours, the simulation covers 0.25 sec real time.  Several observations stand out:

Simulation Parameters

The simulation uses adaptive time stepping which averages to approximately 0.000001sec time steps.  We generate snapshots approximately 0.005sec.  The airplane is assumed to arrive with an estimated initial velocity of 800 ft/sec.  Penetration to column row 4 takes approximately 0.1sec.  

The Animations

  Animation Image Preview
(click to see animation)
Slower Animation
and  Stills
Problem Size
Compute Time
1. Single Column

Pentium 4

  100 K 10 hours
2. Wing w/o ribs

IBM Nighthawk2
16 processors

Low-speed Replay Fluid

3. IBM Nighthawk2
16 processors
Low-speed Replay 500 K 106 hours
4. IBM Regatta
8 RISC processors
Low-speed Replay


1.2 M nodes approx.
20 hours
5. IBM Regatta
coarse model
Low-speed Replay 300 K nodes 24 hours

0.2 sec
real time

6. IBM Regatta
detailed model
Low-speed Replay

Frame 33

1 M nodes 68 hours

0.25 sec
real time



This work has been supported in part by Purdue's Computing Research Institute and by NSF grants EIA 02-16131, DMS/CCR 01-38098, CCR 99-02025, CMS 02-04537, and by  ARO Contract 39136-MA.  Equipment donations from Intel and IBM are also gratefully acknowledged.


Some of the sites no longer have the original story on line.  In that case, only the organization and date of the web page is listed

Purdue University News Release, Sep. 10, 2002, Sep 10 (no longer available) and Sep 28, 2002

Science Daily, Sep. 11, 2002
Indianapolis Star, Sep. 11, 2002
Supercomputing Online, Sep. 11, 2002

UPI, Sep. 12, 2002

BBC News, Sep. 13, 2002; cached pdf file, Sep. 13, 2002
HPC Wire, Sep 13, 2002; story access for members only, Sep 13, 2002

The Civil Engineer, Sep. 23, 2002

LSTC FEA Information, Issue 9, Sep. 2002; cached pdf file, not dated

New York Times, Nov.5, 2002; cached pdf file