Differences between revisions 1 and 30 (spanning 29 versions)
Revision 1 as of 2011-02-26 21:47:44
Size: 1097
Comment:
Revision 30 as of 2011-02-27 04:23:13
Size: 5031
Comment:
Deletions are marked like this. Additions are marked like this.
Line 1: Line 1:
#format jsmath
Line 6: Line 8:
$ { \partial \over { \partial t } } { \partial^2 \over { \partial l^2 } } $
added to the spacing forces. The big problem is measuring the curvature, which is exceedingly difficult to do for millimeter scale lateral displacements over track distances of tens of kilometers.
$\large { \partial \over { \partial t } } { { \partial^2  } \over { \partial L^2 } } $ ( $ L $ is the distance along the track ) added to the spacing forces. The big problem is measuring the curvature, which is exceedingly difficult to do for millimeter scale lateral displacements over track distances of tens of kilometers.
Line 9: Line 10:
{{attachment:triangle_stability.png}} {{attachment:triangle_distance.png|triangle distance| width=300}}
Line 13: Line 14:
$ D ~ = ~ {{sqrt{2*(A^2B^2+A^2^C^2+B^2C^2)-(A^4+B^4+C^4)}}\over{2C} $ $ \large D = {\sqrt{2({A^2}{B^2}+{A^2}{C^2}+{B^2}{C^2})-({A^4}+{B^4}+{C^4})}\over{2C}} $
Line 17: Line 18:
$ {{\partial D}\over{\partial A}} ~ = ~ $ $ \large {{\partial D}\over{\partial A}} = { { A({B^2}+{C^2}-{A^2}) } \over { 2 {C^2} D } } ~ ~ ~ ~ $
$ \large {{\partial D}\over{\partial B}} = { { B({A^2}+{C^2}-{B^2}) } \over { 2 {C^2} D } } ~ ~ ~ ~ $
$ \large {{\partial D}\over{\partial C}} = { ({{A^2}+{B^2})-({C^2}+2{D^2}) } \over { 2 C D } } $

If D is small, these partials are very large. Assume that $ A \approx B $ (but ''not'' the same!) and that $ C = 2A - \epsilon $, and $ D << A $. This is the case if we are measuring the center point on a wide, very shallow and approximately symmetrical triangle.

Our equations reduce to:

$ \large D \approx \sqrt { A \epsilon } ~ ~ ~ ~ $
$ \large {{\partial D}\over{\partial A}} \approx {{\partial D}\over{\partial B}} \approx {-{{\partial D}\over{\partial C}}} \approx { A \over { 2 D } } $

What if D is about 1 cm, and A is about 100 km? Then the partials are all 5E5, which means a 20 nanometer error measuring A, B, or C results in a 1 millimeter error measurement in D. If there are independent $ \sigma = 1mm $ errors in each variable, the RMS error sum is $ \sigma = 1.7 mm $. If we are using 3 such numbers to compute a curvature, we will be taking differences of twice the center minus the ends, resulting in an error sum of $ \sigma = 3.4 mm $. So, an error of only (20 nm/100km) or 2e-14 in our distance measurement results in a 3.4 mm uncertainty in track position. If we computing the time derivatives of successive curvature measurements, attempting to compute a damping factor, the error is increased again, for a curvature velocity error of 4.9 mm/timestep .

If we are using laser interferometers through the atmosphere, especially turbulent atmosphere with moisture and haze (maybe clouds!), it is quite unlikely we can make such long baseline measurements with 1/20 wavelength accuracy. This is hard to do in a temperature controlled box inside a lab. So, even if the partial differential equations indicate a convergent solution, they imply that we can compute accurate partials from the data available. This is unlikely.

This problem is similar to Euler's column formula for a tall skinny compression member. If the column is too skinny, a very small load, a tiny fraction of the compressive strength, will buckle it. A mass stream compression member resembles an Euler column. The solution is similar - guy wires and trusses. Over a very limited correction range (a fraction of the track/rotor spacing), and with a perfectly transparent atmosphere, those "guy wires" and "trusses" can be accurate laser interferometers. However, real atmospheres with clouds and doppler-shifting air movements can mess up the interferometers. So, the laser equivalent of long guy wires are not available, except for calibration on clear calm days. Laser "truss-works" can only be expected to operate over small distances in thick clouds - they must be multi-spectral and use many redundant paths.

Besides a measurement trussworks extending laterally, to properly stiffen our "column" we will need a mechanical system that can absorb and propagate lateral forces - cable trusses and guy wires. A wider "column" with a much higher bending moment will be stiffer. But the real opportunity is adding dynamic elements that can modulate forces and shift the center of mass, such as moving actuators, moving masses, propellers and shutters in the atmosphere. With properly designed measurement and feedback, these systems can appear much stiffer than passive material elasticities can provide. With good measurement, broadly distributed mass, and fast actuators, the system can converge on an estimated neutral position and remain there, even if there are fairly large residual errors in the measurements and control, and large deviations which cannot be corrected merely with spacing control.

MORE LATER

Stability

The main issue with stability is that perturbations exceeding a few millimeters from the ideal path ( as determined by ambient forces acting on the track ) cannot be corrected by track to rotor spacing alone. Given inaccurate force measurements, and track measurements, it is very easy for such perturbations to appear, indetectably.

Stability can be achieved by applying damping forces to the curvature, some constant \large { \partial \over { \partial t } } { { \partial^2 } \over { \partial L^2 } } ( L is the distance along the track ) added to the spacing forces. The big problem is measuring the curvature, which is exceedingly difficult to do for millimeter scale lateral displacements over track distances of tens of kilometers.

triangle distance

Suppose we are trying to estimate distance D from three precision measurements for A, B, C. We can compute D from this formula:

\large D = {\sqrt{2({A^2}{B^2}+{A^2}{C^2}+{B^2}{C^2})-({A^4}+{B^4}+{C^4})}\over{2C}}

The sensitivity of D to errors in A , B , and C is just the partials:

\large {{\partial D}\over{\partial A}} = { { A({B^2}+{C^2}-{A^2}) } \over { 2 {C^2} D } } ~ ~ ~ ~ \large {{\partial D}\over{\partial B}} = { { B({A^2}+{C^2}-{B^2}) } \over { 2 {C^2} D } } ~ ~ ~ ~ \large {{\partial D}\over{\partial C}} = { ({{A^2}+{B^2})-({C^2}+2{D^2}) } \over { 2 C D } }

If D is small, these partials are very large. Assume that A \approx B (but not the same!) and that C = 2A - \epsilon , and D << A . This is the case if we are measuring the center point on a wide, very shallow and approximately symmetrical triangle.

Our equations reduce to:

\large D \approx \sqrt { A \epsilon } ~ ~ ~ ~ \large {{\partial D}\over{\partial A}} \approx {{\partial D}\over{\partial B}} \approx {-{{\partial D}\over{\partial C}}} \approx { A \over { 2 D } }

What if D is about 1 cm, and A is about 100 km? Then the partials are all 5E5, which means a 20 nanometer error measuring A, B, or C results in a 1 millimeter error measurement in D. If there are independent \sigma = 1mm errors in each variable, the RMS error sum is \sigma = 1.7 mm . If we are using 3 such numbers to compute a curvature, we will be taking differences of twice the center minus the ends, resulting in an error sum of \sigma = 3.4 mm . So, an error of only (20 nm/100km) or 2e-14 in our distance measurement results in a 3.4 mm uncertainty in track position. If we computing the time derivatives of successive curvature measurements, attempting to compute a damping factor, the error is increased again, for a curvature velocity error of 4.9 mm/timestep .

If we are using laser interferometers through the atmosphere, especially turbulent atmosphere with moisture and haze (maybe clouds!), it is quite unlikely we can make such long baseline measurements with 1/20 wavelength accuracy. This is hard to do in a temperature controlled box inside a lab. So, even if the partial differential equations indicate a convergent solution, they imply that we can compute accurate partials from the data available. This is unlikely.

This problem is similar to Euler's column formula for a tall skinny compression member. If the column is too skinny, a very small load, a tiny fraction of the compressive strength, will buckle it. A mass stream compression member resembles an Euler column. The solution is similar - guy wires and trusses. Over a very limited correction range (a fraction of the track/rotor spacing), and with a perfectly transparent atmosphere, those "guy wires" and "trusses" can be accurate laser interferometers. However, real atmospheres with clouds and doppler-shifting air movements can mess up the interferometers. So, the laser equivalent of long guy wires are not available, except for calibration on clear calm days. Laser "truss-works" can only be expected to operate over small distances in thick clouds - they must be multi-spectral and use many redundant paths.

Besides a measurement trussworks extending laterally, to properly stiffen our "column" we will need a mechanical system that can absorb and propagate lateral forces - cable trusses and guy wires. A wider "column" with a much higher bending moment will be stiffer. But the real opportunity is adding dynamic elements that can modulate forces and shift the center of mass, such as moving actuators, moving masses, propellers and shutters in the atmosphere. With properly designed measurement and feedback, these systems can appear much stiffer than passive material elasticities can provide. With good measurement, broadly distributed mass, and fast actuators, the system can converge on an estimated neutral position and remain there, even if there are fairly large residual errors in the measurements and control, and large deviations which cannot be corrected merely with spacing control.

MORE LATER

Stability (last edited 2011-02-27 07:17:58 by KeithLofstrom)