Title: Modeling, Simulation, and Control of Digital Clay Mod
1Modeling, Simulation, and Control of Digital
ClayModélisation, Simulation, et Control de
lArgile Numérique
Students Stephen ASKINS Samuel DESSOLIN
Project Directors Monsieur André BARRACO (ENSAM,
Paris) Doctor Wayne BOOK (GATech, Atlanta)
2Modeling, Simulation, and Control of Digital Clay
- NSF funded haptics research project at GATech
Digital Clay - Haptics Interfacing with a computer through
touch by transmitting or sensing forces. - Carried out as PFE at ENSAM/Paris as a result of
collaboration through GTL Program.
3General Introduction to Digital Clay
- Novel distributed input/output device with
continuously variable surface. - To operate like clay
- Surface shaped by user è computer model shape
mode. - Computer model è surface assumes shape display
mode. - Computer model è surface assumes feel impedance
mode. - Fluidically actuated.
- Applications design, art, education.
4Digital Clay Architectures
- Currently no fixed design for digital clay, in
process of making these decisions. - Possible architectures
- Bed of Nails.
- Bellows.
- Distributed architecture.
5PFE
- Part of Digital Clay project
- Beginning of a 5-year project.
- Members of Controls Group, lead by Dr. Book.
- Far from other teams involved ? few
communication. - Modeling, Simulation, and Control
- No concrete values or even the kind of structure
used - ? be as general as possible.
- ? use logical numbers, see their influence.
6General Approach
- Assumptions Fluidic cells, changing size, 2-
or 3-D network, membrane
7Cell Modeling
- Using MATLAB/Simulink.
- Goals
- Translate phenomena and physical laws into block
diagrams. - Build the base of a general model of Digital
Clay. - Steps
- Conventional model.
- Dual Mass model.
- Results and comparisons.
8Cell Assumptions
- Kinetic mass variable mobile submitted to
forces. - Hydraulic flow due to pressure difference,
conservation of mass, incompressible fluid. - Consider height limitation.
- Fexternal divided in 2 effects
- Squeezing due Fmin.
- Pushing due to (Fmax-Fmin).
- Pcell in equilibrium with Fsqueezing and Fc.
9Conventional Model
10Dual Mass Model
Only small difference in acceleration when
unanchored.
11Conclusions
- Conventional Model (Fave) and Dual Mass Model
perform equivalently on tests so far. - Different satisfying tests Display mode, Shape
mode. - Continue using both models for verification.
-
12Membrane Modeling
- Using ANSYS Finite Element Code.
- Membrane advantages
- Better visual and tactile effects.
- Enclose device space.
- Goals
- Feed MATLAB network model with analytical
formula. - Build Analytical Models by identification.
- Steps
- Linear Analysis and Linear Analytical Model.
- Nonlinear Analysis and Nonlinear Analytical Model.
13Linear Analysis and Linear Analytical Model
14Nonlinear Analysis
- Nonlinearity due
- Geometry of load.
- Material type.
15Nonlinear Analysis
16Nonlinear Analytic Model
Only 1 type of bar
17Conclusions
- Linear Model
- First step.
- Useful for MATLAB network model.
18General Modeling
19Network Modeling
- Use piston-ram cell.
- Bed of nailsarchitecture
- Single layer.
- Grid pattern.
- Conventional model sufficient.
- Form 36-cell device.
- Cover with membrane.
20Creating Network
- All as single Simulink Model file.
- Using subsystems for simplifying.
- Work intensive
- Both creating and editing these block diagrams
can be difficult and error prone. - Possibility of automating this process using
MATLAB commands. - Possibility for at most 100 cell networks in pure
simulink format. - Attempt at discretizing using MATLAB code failed.
21Network Testing (Uncontrolled)
Results for Single Frozen Cyl
Results for Single Forced Cyl
- Tested with various inputs to verify expected
behavior and find bugs in block diagram. - These results show speed of response for
unpressurized cell.
22Control Implementation
- Proportional
- Standard first implented.
- On/Off
- Simplest allows 2 state valves.
- Neighbor Influenced
- Proportional but with gain dependent on expected
resistance from membrane. - May be useful given importance of membrane to
cell response. - PID (proportional integrator derivative)
- Standard control improvement.
- No need or advantage seen
- not shown
General Control
Controls Implemented
23Examples of Control Tests
- Shown with Proportional Control
- Kc 100
- Time scale 5s
- Only limiting factor for forming shapes is
saturation of pressure against membrane
24Control Results Proportional
- Results shown for K 25 100 500 and 10000 on
scale of 3s. - Even as gain is increased so that system
approaches speed capacity negligible overshoot
seen. - System, as modeled, very stable probably no need
for complex control for forming shapes.
Effect of Reducing Gain
25Control Results On/Off
- Because of stability at high gains, On/Off
(equiv. to K ) deemed possible. - Even w/ dead zone on order of 1x10-5 m we see
only small over shoot and fast settling - RTmax .104s.
- Advantage valves need not be continuous
- ie. solenoid style valves.
26Control Results Neighbor Influenced
- Based on equation
- Adjust gain based on stretch on membrane
- Only adds to Kn for Herrors such that membrane
resists cell movement. - Some advantage seen, but currently no great
advantage over raising gain in overall
performance.
27Conclusions
- Cell Modeling
- Two models developed which should be useful for
simulating most digital clay architectures and
will hopefully require little modification for
adapting to real actuators. - Membrane Modeling
- Limiting factor in having useful ANSYS results is
displacements achieved. - New analytical model would be useful.
28Conclusions
- Network Modeling
- Demonstrated capability of all Simulink devices
on order of 50 to 100 cells, though tedious. - Showed difficulties in employing MATLAB m-files.
- Control
- We note that the Bed of Nails Architecture
appears inherently easy to control. - Cannot offer conclusions on more complex
architectures.
29Next Steps
- To be continued in GATech as MS thesis.
- Cell Modeling
- Extension of movements in horizontal directions
? 3-D. - Membrane Modeling
- Get larger vertical displacements in Nonlin Ana
if possible. - Try a different approach of curve fitting.
- Network Modeling
- Simulate with nonlinear springs (hyperelastic
bars). - Build a multiple levels structure.
- Generate models with MATLAB commands.