RESEARCH DIRECTIONS Srinivasa M. Salapaka Laboratory for Information and Decision Systems Massachusetts Institute of Technology Department of Mechanical Engineering Iowa State University March 25, 2003
Jan 06, 2016
RESEARCH DIRECTIONS
Srinivasa M. SalapakaLaboratory for Information and Decision Systems
Massachusetts Institute of Technology
Department of Mechanical EngineeringIowa State University
March 25, 2003
2
Outline
• Research Directions Nanopositioning Micro-Cantilever Dynamics Nanofriction Clustering Algorithms Image deblurring
ROBUST BROADBAND NANOPOSITIONING
4
MOTIVATION
• Nanopositioning High Bandwidth
High throughputs• High throughput requirements in probing material surfaces
• Binding affinity between materials, other properties• High speed requirements for studying biosystems
• Cell dynamics, probing living systems• Faster scanning requirements in various engineering applications
• Ultra high density data reading and writing Enabling feature in many studies and applications
• Studies of cell dynamics require micro/nano-second imaging capabilities
Ultrahigh precision Specifications are often in the angstrom regime In scanning probe technologies molecular and atomic forces are routinely
probed
Robustness Necessary for reliability in view of
• Uncertainty in model and environment• Diverse users –do not have the engineering expertise
5
MOTIVATION
• Nanopositioning system
High precision (probing at nanoscale) High bandwidth (high throughputs) Robustness (reliability and repeatability)
Needs ofCombinatorial
Chemistry
6
OBJECTIVE
Robust Broadband Nanopositioning System with
• 500 Hz for large scans (100 m £ 100 m)• nanometer resolution
• 1 MHz for small scans (2 m £ 2 m)• subnanometer resolution
Compatible for scanning probe applications
7
APPROACH
• Novel Device Architecture
• Novel paradigm for robustness, bandwidth and resolution
8
Proposed design
• Two stage scanning Large Scans
Motion possible by flexure based design• Sample-holders on steel platforms
• Heavy (smaller bandwidths) Actuation by stack-piezos
• Large forces, large travels (100 m)
Small Scans Cylindrical Piezoactuators
• Sample kept on actuator itself• Smaller travels (2 m)• Lighter (higher bandwidths)
Integrate the two Put the small scanner on top of large scanner
9
A Schematic of PROPOSED Nanoscope Head top
EOD, Laser
Laser to photodiode
Head
Laser from EOD
Microcantilever
Support Plate
X-Y-Z small rangenanopositioner
Large range nanopositioner
Mirror
MicrocantileverHolder
10
Large Range Scanner
11
PRESENT STATUS AND FUTURE DIRECTIONS
• Developed a precise paradigm to address: High Bandwidth High Resolution Robustness
• Modern control tools Model the plant Quantify and characterize the challenges Design feedback laws
Practically eliminated hysteresis and creep Obtained 60-70 times improvement in the bandwidth
over current popular systems Substantial improvement in the reliability and
repeatability
12
Results (cont’d.)hysteresis creep bandwidth
RepeatabilityReliability
tracking
13
Results
• Large Scanners Identified and addressed design challenges
on bandwidth, precision and robustness• Piezo actuation is predominant; hysteresis and creep
nonlinearities, design constraints• Sensors can deteriorate open loop performance
Employed modern control tools to address these challenges and achieved
Performance• controllers to achieve the desired tradeoff between
resolution and bandwidth Robustness
• By addressing model uncertainties
14
Preview based control design
• Improve tracking performance For a priori known reference trajectories
feedforward controller in addition to feedback controller
To give desired input ud such that Gud(t)=xr(t)
+ -
Feed forward Controller
Plant
Anticipatory Control design for better tracking performance
15
Preliminary Simulation Results
significant improvement in performance Substantial reduction in error
16
Multi-Input Multi-Output Control Design
Gxx
Gyx Gyy
Gxy ¼ 0
17
Multi-Input Multi-Output Control Design
• MIMO design Significant coupling effects
Gyx greater than Gyy in some frequencies Carry out control design for the MIMO model
• Glover McFarlane, Nominal and Robust H1
Multi-objective design Actuation constraints
• Specified by H1 norm
Resolution specifications• addressed by H2 norm
Control Design for plant model that includes X-Y coupling
18
Integration into the nanoscope
Integrate the probing head with the positioning system
Sample holder capable of moving in Z direction Control of tip-sample separation
MIMO control design for positioner and cantilever system (3 £ 3 model) Account for tip-sample interactions
• Nonlinear models Observer based control design
• z-displacements are measured but velocities are not measured
• Observers useful for compensation designs for nanofriction
Control Design for plant model that include positioning (X-Y) and probing (Z) aspects
19
Short Range Scanner
• For high bandwidth Low mass essential
Cylindrical piezos – scanner cum actuator Can be run open-loop
Inverse dynamic schemes • Inverse hysteresis models
Alternatively use closed loop control loop design Design/implement sensors for detection of lateral motion Employ the control design procedure as done for large
range scanner
Smaller lighter scanner implies faster scanning
20
Lateral motion sensors for AFM
• Previous experience Designed sensors for shell piezos (J scanner in an AFM)
• Designed sensors based on optical levers
Used them for feedback Loop shaping control laws
Obtained substantial improvement in the performance Resolution in order of few nm (1kHz) Bandwidth improvement of over 20 times
21
Another untried approach
• Build a new nanoscope with control design in mind Make small scanners
Lighter and therefore high resonant frequencies• Faster scanners
Bigger coupling effects• More burden on control design
Upshot Simpler device design More emphasis on control design Achieve higher bandwidths
Shift the emphasis from device design to control design and achieve faster scanning rates
CLUSTERING
23
What Is Clustering?
Clustering Separation of set of objects into groups such that objects
in one group are more ‘similar’ than those in other
• find the optimal partition {Rj} of the domain and the allocation of representative locations
X
Combinatorially complex problem Interpret and design d(x,rj) Adapt and modify Deterministic
Annealing Algorithm Simulations
24
MOTIVATION
Chemoinformatics, Combinatorial Discovery• Search by elimination through a ‘chemical space’ for a
‘backbone’ compound (drug discovery)• Enormous number of possible molecular combinations• Requires clustering algorithms to narrow the search
Essential in data mining, data compression, facility location, machine learning
25
OBJECTIVE
Develop and adapt clustering algorithms for Combinatorial Discovery
26
Present status and future directions
Partition a ‘large’ space ‘optimally’ into a given number of ‘cells’ and specify ‘representative locations under constraints
• Similar to dividing ‘chemical space’ into clusters with representative elements
Developed fast algorithms under which a new class of problems were for the first time identified precise mathematical formulations were provided Algorithms developed that are fast The developed algorithms utilized on real life systems
27
EXAMPLE SYSTEM
IMAGE RECONSTRUCTION
29
MOTIVATION
• Blurred images in scanning probe microscopy The tip-geometry convolves with the sample to
provide a blurred image
30
Objective
• Deblurred using deconvolution methods Modeled as convolution equation: y=h*x
• y is observed data, h is blurring function, x is original data Deconvolution is obtaining x given y
• Equivalent to solving a system of structured system of equations of the form Ax=b
• A is usually very large
Develop and implement deconvolution algorithms for image deblurring
31
Present Status and future directions
Developed algorithms for solving deconvolution equations
Significant reductions in the computational expense domain is not necessarily rectangular or continuous
• Common in microscopy• Scans of different areas in a sample
Implement these algorithms for deblurring applications Study other convolutions in microscopy
• Geometric convolution
32
Practical Example Systems
• Deblurring function: hn1n2=exp(-(n12+n2
2)/104)
• Substantial reduction in computational expense
MICROCANTILEVER BASED DEVICES
34
Micro-Cantilever Arrays
Multi-Cantilever arrays Parallel probing
• Higher throughputs Coupling effects
• Modeling and Analysis• Associated control design
• Distributed control structure• Individual actuation and sensing
• Fabrication and implementation issues
Parallel and faster probing to obtain higher throughputs
35
Micro-cantilever Sample Dynamics
Understanding micro-cantilever-sample dynamics Essential to probing surfaces at nanoscales Important for designing X-Y positioning systems
Studying complex dynamics Dependence on model parameters
• Complex dynamics shown analytically and observed in experiments• Important to identify avoidable conditions for imaging• Use them as test beds to study rich dynamics
Previous experience Obtained a model to describe an AFM experiment Proved and observed complex dynamics
NANOFRICTION
37
Nano-friction
• Widely studied area Fundamental understanding of interfacial phenomena
nanotribology Study these phenomena in micro/nanostructures
Magnetic storage systems, nanolithography
• System theoretic approach Not explored Obtain models to model friction at nanoscales
Explain observed phenomena Use control tools to compensate for friction
Use observer based design Friction compensation important in applications
nanolithography
System theoretic modeling, analysis and compensation for nano-friction
38
Nano-friction (cont’d.)
• Preliminary work Dynamic model for AFM
With friction model using JKR theory Simulation of model show stick-slip motion feedback laws to compensate stick-slip demonstrated in
simulation Substantial reduction of error in tracking
• z-velocities were obtained from the model in the control design
Proposed work Implement observer based design Develop models to explain more observed phenomena