Engineering Paper - Orion Crew Module Water Landing Simulation at NASA

Crew Module Water Landing Simulation
Methods Development for NASA

Mahesh Patel
Engineering Manager, Altair ProductDesign Inc
38 Executive Park, Suite 200, Irvine, CA, 92614 4709, USA
mahesh@altairpd.com

Drew Burkhalter
Senior Project Engineer, Altair ProductDesign Inc
38 Executive Park, Suite 200, Irvine, CA, 92614 4709, USA
drew@altairpd.com

www.altairproductdesign.com
copyright Altair Engineering, Inc. 2011


Abstract

The NASA Orion Crew Module water landing is a very complex dynamic event and accurately
creating a computer simulation model of this event is both computationally taxing and input
sensitive.

Outside of the many physical variables associated with the event there are many simulation
input variables such as mesh density, boundary conditions, and contact interfaces which have
an enormous effect on the accuracy of the simulation results. Consequently, identifying the
desired outputs from the simulation results is a critical aspect to the complexity of the model
and determining input variable sensitivities to drive the desired solution. Although there are
closed form solutions available to predict quasi-dynamic water landing events, it is imperative
to have accurate physical test data in order to correlate and anchor the finite element (FE)
simulation models such that the models can potentially be used as a predictive tool with a high
level of confidence for certain conditions encountered during water landings and can be
effectively be used for structural design iterations.

The NASA Engineering and Safety Center (NESC) aims to establish a clear understanding of
the specific modeling methods needed to perform dynamic simulations of the Orion Crew
Module water landings. Specifically, the work seeks to determine the critical simulation
variables, methods and physical testing needed to create an accurate computer simulation
FEA model. With an accurate FEA model, accelerations, loads and trajectories can be used to
evaluate and develop astronaut safety systems during water landing as well as predict the
structural stability of the Crew Module structure itself. Explicit Arbitrary Lagrangian Eulerian
(ALE) analysis was performed using the Radioss Block 100 dynamic solver of angled vertical
entry water landing of the Orion . Several models were created to hone-in on the critical input
variables and their effects on the desired outputs (Crew Module trajectory, accelerations, and
pressure loads on the heat shield)

Nomenclature
NASA = National Aeronautics and Space Administration
NESC = NASA Engineering and Safety Center
FEA = Finite Element Analysis
ALE = Arbitrary Lagrangian Eulerian
IMU = Inertial Measurement Unit
GAP = contact activation disttance
Lc = characteristic element length of the fluid
Stfac = contact stiffness factor
Ρ = highest fluid density in the model
V = impact velocity of the Lagrangian mesh
Sel = surface area of Lagrangian impact element
CG = center of gravity

Copyright Altair Engineering, Inc., 2011 2


1.0 Introduction

The structural design of the Orion Crew Module considered various loading conditions
experienced by the Crew Module such as liftoff loads, launch abort loads, re-entry loads, and
water impact landing loads. It was determined that one of the largest loading conditions to the
structure of the Crew Module was Earth water landing. In order to maintain structural integrity
and increase safety of the astronaut crew it is desired to have a more clear understanding of
the dynamic loads generated during water impact.

Using Finite Element Analysis (FEA) to accurately predict water impact loads at splashdown
on the Orion Crew Module would ultimately increase astronaut safety, optimize the Crew
Module main and sub-structure, and provide dynamic data of the entire event. This would
allow for the study of various landing conditions, velocities and impact angles. FEA technology
allows engineers to investigate the entire event in detail unavailable if strictly physical test
methods are applied.

The main limitation of applying FEA simulation to dynamic impact events is that physical test
data is required to anchor initial FEA models. Once the initial FEA models are correlated to
physical tests, they can be confidently and effectively used as a predictive tool during
structural design and analysis.. The goal of this effort was to establish simulation methods and
best practices when modeling this type of dynamic event.

2.0 Physical Test Set-Up

A full scale boiler plate Crew Module was fabricated by NASA to perform the physical testing
and was especially tailored for water landing impacts. This Crew Module was primarily built
from steel with reinforcements so that it could be analytically treated as a rigid body. This steel
version of the Crew Module was instrumented with several data collecting devices such as
accelerometers, strain gauges, inertial measurement unit (IMU), and pressure sensors.
Photogrammetric targets were also placed on the outside surfaces to accurately measure the
Crew Module trajectories (Figure 1). High speed video cameras were placed at strategic
locations. The tests were performed at a still fresh water deep lake.



Figure 1: Crew Module Physical Testing

Sixteen physical drops were performed at slightly different impact angles and impact velocities
(Table 1). The collected data from the sixteen drops showed very high repeatability for the
trajectory and acceleration data. High quality acceleration and photogrammetry data were
obtained from physical testing.

Heat shield pressure sensor data and strain gauge data from the tests were acquired as
auxiliary data to prepare for a second set of physical tests. Strain data obtained from the tests
were not used during the initial correlation process of this phase of the project. High speed
video data was used during model correlation and provided high resolution slow motion
insights into the impact events. Processing and quality assurance of all the test data collected
during the drops proved to be a data management challenge.

Out of the sixteen physical drops that were tested, two (Drop 2 and Drop 8) were identified as
having the most stable data and it was decided to correlate the FE models to only these two
reference drops. The raw data from these two drops were filtered, and supplied to the
simulation team to aid with the correlation of the FEA models. Test data sample rates, type of
filtering used on the raw test data and filtering cut off frequencies were also supplied to the
simulation team.



Table 1 - Crew Module Drops Tested

3.0 Simulation Model Set-Up

Crew Module

Computer-Aided Design (CAD) files of the NASA boilerplate Crew Module were provided and
meshed uniformly using HyperMesh 10.0 SA1 with a four inch by four inch average mesh
size. The model was set-up so the mesh of the outer surface of the Crew Module was
connected to the rest of the internal Crew Module mesh structure through the use of tied
contacts. This method of modeling allowed for the heat shield surface mesh to be changed
easily so that a mesh sensitivity study could be performed more efficiently.

Although all components of the Crew Module structural components were meshed in detail,
the Crew Module structure, excluding the heat shield, was modeled as a rigid body. This level
of detail allowed the meshed Crew Module to have the correct moments of inertia and to be
within 1% of the measured mass of the fabricated Crew Module. It also provided the flexibility
to switch and run a deformable (non-rigid) model if necessary. Local coordinate systems and
accelerometers were positioned in the model using the Crew Module coordinate system
replicating the locations of the physical test accelerometers (Figure 2). An internal Radioss
SAE 1000 filter was setup for each accelerometer output.



Figure 2: Crew Module Accelerometers (Top View)

Air/Water

A 30 foot by 30 foot plan view section of the water and air was modeled with an Eulerian mesh
(Figure 3). A nonreflective boundary condition was assigned to the outer perimeter of the
water and air to better simulate an infinite area.

Twenty-five feet of water depth was used to match the drop test condition and the base nodes
were constrained with a reflective boundary condition. In order to fully envelope the Crew
Module at time zero in air, thirteen feet of air height was modeled and the top layer of nodes
were constrained with rigid boundary conditions.

Figure 3: Air and Water Eulerian Mesh



Since the air pressure change over 13 ft is negligible the air was given an initial atmospheric
pressure at sea level. Layers of initial pressures were assigned to the water mesh based on
water depth. Although the solver would have eventually converged on the correct water
pressures due to the imposed initial gravity field, the model stabilizes much faster if initial
pressures are assigned to the water mesh based on depth (Figure 4).

Figure 4: Water Pressure Distribution at t = 0

Initial Conditions

In the simulation, the Crew Module was initially eight inches from the water interface so the
ALE coupling contact interface between the Crew Module and water could be established
effectively. This initial distance also captures the air compression caused by the accelerating
Crew Module at the water to air boundary.

The contact definition between the Crew Module and surrounding fluid was a penalty-
stiffness-based-coupled Eulerian-Lagrangian (CEL) method. Instead of using the default
Radioss parameters for contact stiffness and activation distance, these parameters were
calculated using established equations (Figure 5) available within Radioss3.



Figure 5: Radioss Interface Parameter Equations

4.0 Baseline Analysis

Prior to the physical drop tests, initial ALE models were created and analyzed to see if the
FEA tool could successfully predict reasonable “blind” results. Mesh and modeling parameters
were selected based on achieving reasonable model run times and mesh convergence was
not performed on these models.

When test data did become available, comparison of trajectory and rotation data from the
physical tests matched fairly well with baseline simulation results. The acceleration data,
however, did not correlate very well as the “blind” FEA acceleration results and appeared to
oscillate around the test acceleration data (Figure 6).



Figure 6: Test Data vs. Blind FEA Results: Crew Module Trajectory (top);
Crew Module Acceleration (bottom)

5.0 Model Correlation

After the test data was available to the FEA simulation team, correlation was performed by
varying FEA modeling input parameters in order to more closely match physical test
accelerations.

Acceleration data was the main focus of this correlation work since trajectory and rotation data
seemed to change little due to model input parameter changes and the physical test pressure
data collected was deemed unreliable. Sensitivity analysis on parameters such as interface
stiffness, mesh density, fluid pressure distribution, and boundary conditions were performed
with relation to acceleration data.



Although optimizing all these parameters was crucial to correlating the model, mesh density
was by far the most influential parameter. During this initial correlation phase, the Crew
Module mesh was kept at the same mesh dimensions (average of 4 inch by 4 inch) and only
the fluid mesh elements were varied in dimension.

During the correlation process it appeared the mesh density was much more sensitive in the
direction normal to the impact surface (vertical). In order to better match the test data it was
necessary to have elements with relatively small dimensions in the direction of the impact
while the other two dimensions were less sensitive. Keeping the dimensions that were less
sensitive relatively large allowed for faster runtimes while not losing much accuracy. From the
correlation studies, it was determined that two inch by two inch by one inch fluid elements at
the impact interface of the air and water (Figure 7) while keeping the Crew Module mesh size
four inch by four inch were producing the best results.

Figure 7: Best Element Size at Water/Air Impact Interface

Additionally to speed up runtimes the fluid mesh was gradually biased in all three directions
from the water/air interface impact point. Several iterations were performed to optimize mesh
biasing. Final mesh biasing showed little variation on the acceleration results, but reduced
model size and improved computer runtime significantly.

To further increase confidence in the simulation model, overall peak simulation acceleration
results were validated using a closed form solution by Von Karman4 on a separate zero angle
Crew Module simulation analysis (Figure 8).



Figure 8: Radioss Solution vs. Von Karman Closed From Solution

Simulations were performed comparing accelerations on a completely rigid crew model with
accelerations on a non-rigid (steel) crew module. It was determined that the accelerations
varied very little between the rigid and non-rigid crew modules. This is largely due to the fact
that the actual construction and design of the steel Crew Module built for performing physical
tests is naturally a very stiff structure resembling properties of a rigid structure.

Post correlation simulation model results are show in Figure 9 (Drop2) and Figure 10 (Drop8).
As can be seen from both drop correlation results, the simulation accelerations and
trajectories match closely to the physical tests. Initial acceleration ramps for both drops also
match very closely at predicted peak acceleration times. There are minor differences in peak
accelerations compared to test peak accelerations for both drops. Overall time history
signatures of the acceleration also correlate very well. There is a slight dwell or fluctuation
during the latter part of the event. Data filtering was performed using a Butterworth filter (95Hz
cut-off frequency and a time conversion of 0.7).



Figure 9: Drop 2 Acceleration, Displacement and Rotation Correlation



Figure 11: Drop 8 Acceleration, Displacement, and Rotation Correlation

6.0 Mesh Study Matrix

In order to better understand the combined effects of both the Lagrangian (Crew Module) and
ALE (Air/Water) mesh dimension sensitivities on the Crew Module acceleration, a mesh matrix
sensitivity study was performed using the correlated simulation model.



A matrix consisting of twenty separate models was created with different combinations of
mesh density for the Lagrangian Crew Module as well as the ALE fluid mesh (Table 2). For all
of the models, the overall size of the fluid modeled was reduced in order to have enough
computer resources to complete the matrix in a timely manner.

Table 2: Mesh Study Model Matrix

Simulation analysis run times varied from 20 minutes with four parallel cpus for the coarse
mesh models to 18 hours with 24 parallel cpus for the finer mesh models. The overall
dimensions of the fluid volume modeled were kept constant and only the mesh size varied.

The mesh of the fluid volume was varied in length, width and depth while the mesh of the
Crew Module was only varied in length and width because 2D shell elements were used to
represent the Crew Module. The first column in Table 2 describes the length, width and depth
of the fluid elements at the air/water interface where depth is the direction normal to the water
surface.

Columns two through five roughly describe the average length and width dimension of the
Crew Module mesh that comes into contact with the water. Element size on the Crew Module
heat shield was kept constant as much as possible. Notice that the Crew Module mesh is
never smaller than the fluid mesh. This is considered standard practice for ALE simulations.

The Radioss interface stiffness for each of twenty models was pre-calculated and applied
based on mesh size using the Radioss formulas as shown in Figure 5. Due to the number of
analysis runs and data generated, only the initial conditions for Drop 2 were analyzed during
this mesh matrix sensitivity study.

For each of the twenty models, acceleration data for the three accelerometers were plotted
and compared to the physical test data accelerations as done with the correlated models.
Additionally, the acceleration data was overlaid between each of the runs in the matrix to
determine patterns and trends of mesh changes.

In general, the results showed that a finer mesh in the direction normal to the impact produces
results that most closely correlate to the physical test data. Figure 11 shows a plot of the
simulation accelerations verses the physical test data. This plot only includes the models that



most closely correlated to the test data. Notice that all of them have a one inch fluid mesh
dimension in the direction of the impact.

The models with a fluid mesh of four inch cubes did not produce very good results (Figure 12).
The acceleration curves appear to oscillate around the test curve. This matched what the
original “blind” results showed early in the project before the test data was supplied. If a mesh
sensitivity study would have been performed, Figure 11 shows the solution converges around
the test data.

Figure 11: Best acceleration models verses test data



Figure 12: 4”x4”x4” fluid mesh with variable Crew Module
mesh accelerations verses test data

Figure 13 shows very good correlation between all of the models when applying the smallest
mesh dimension to the fluid mesh (one inch) in all three directions and only varying the Crew
Module mesh size to be same or larger that the fluid mesh in dimension.

Figure 14: 1”x1”x1” fluid mesh with variable Crew Module
mesh accelerations verses test data



This next set of plots (Figures 14, 15, 16 and 17) show that even with a one inch fluid mesh
size normal to the impact direction the Crew Module mesh size makes a significant difference
to the acceleration results. This set of plots illustrates a trend of improved correlation to the
test accelerations when the Crew Module mesh size is equivalent to or larger than the
length/width dimension of the fluid mesh.

In each plot the group that is plotted against the test data has the same Radioss interface
definition (stfac and gap distances are same between the three models). The only difference
within each set is the length/width dimension of the fluid. The results appear to be less
sensitive to fluid mesh size as the Crew Module mesh size gets larger.

Figure 14: Compare models with same gap and stfac,
same depth and Crew Module mesh of one inch



one inch depth and two inch Crew Module mesh

one inch depth and four inch Crew Module mesh



one inch depth and seven inch Crew Module mesh

If acceleration is the primary focus from the simulation it appears that several different mesh
combinations would produce good acceleration correlation. If Crew Module heat shield
pressure correlation is also required, then one must look deeper into the mesh matrix results
to determine which combination produces the best correlated pressure results.

The testing found a maximum pressure of 1.0 normalized units of pressure, and as can be
seen from Table 3 the maximum pressures vary significantly between the models. More
analysis is needed to get acceptable acceleration and pressure data from a single model.

Table 3: Normalized Pressure Comparison



7.0 Conclusions and Future Work

The initial “blind” fluid mesh size selected prior to the availability of physical test data produces
accelerations which oscillate and over estimate peak accelerations. This emphasizes the
importance of having physical test data to anchor the prediction accelerations. Once a
correlated model is derived, it should be possible to perform other landing angle conditions
and have greater confidence in the acceleration results.

Correlation of simulation models to physical test data determined that the best mesh size and
mesh ratios were two inch by two inch by one inch for the fluid volume and four inch by four
inch for the Crew Module heat shield surface. The same mesh size and mesh ratio produced
good correlation for both drops tests with different impact angles.

The results from the mesh sensitivity study matrix revealed that mesh size ratios of Crew
Module to fluid are very important in obtaining good correlation. It is critical to have the
smallest fluid mesh dimension normal to the impact direction. The length and width
dimensions of the fluid elements should be smaller or equivalent to the Crew Module element
length and width dimensions to produce the best results. The model with the two inch by two
inch by one inch fluid mesh size and four inch by four inch Crew Module mesh size had good
acceleration correlation while keeping a reasonable analysis run time.

This study did not take in to account lateral Crew Module velocities which are typically present
during water landings and the mesh size and mesh ratio which include lateral velocities need
to be investigated as future work. It is predicted that the length dimension of the fluid mesh in
the direction of travel will also be an important variable to consider.

The simulation methods developed in this study can be applied to similar water landing events
with similar Crew Module size, inertia, mass and shape properties. The methods developed in
this study are only applicable to the Radioss block 100 dynamic FEA solver and a similar
study may be required when using other dynamic solvers.

The next phase of this work will be to extend the correlation to include pressures on the heat
shield and strains on Crew Module structure. A second set of physical testing is planned to
support this future correlation work, and methods development.

Acknowledgments

This work was sponsored by the National Aeronautics and Space Administration under
contract to Alliant Techsystems. Special thanks to the entire Crew Module Water Landing
Simulation Methods Development team under the direction of NASA.

References
1
von Karman, T., The Impact of Seaplane Floats during Landing, Technical Note 321 NACA,
Washington, D.C., 1929.



2
Altair® HyperWorks®10 Radioss User Guide Manual, Altair Engineering, Troy, Michigan,
October 2009.


Engineering Paper - Orion Crew Module Water Landing Simulation at NASA

More Related Content

Viewers also liked (7)

Similar to Engineering Paper - Orion Crew Module Water Landing Simulation at NASA (20)

Recently uploaded (20)

Engineering Paper - Orion Crew Module Water Landing Simulation at NASA