14.02.2013 Views

Thesis - Leigh Moody.pdf - Bad Request - Cranfield University

Thesis - Leigh Moody.pdf - Bad Request - Cranfield University

Thesis - Leigh Moody.pdf - Bad Request - Cranfield University

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Chapter 7 / Missile Trajectory Optimisation<br />

_ _<br />

7.12 Discussion<br />

The guidance module in the missile<br />

simulator provides the building blocks for<br />

PN and CLOS guidance laws. If none are<br />

selected the program invokes on-line<br />

trajectory optimisation. The optimisation<br />

process is latched to the host’s processor<br />

clock to study the effect of limiting the<br />

available processing time. For example if<br />

the next control to be taken by the autopilot<br />

last for 0.5 s re-optimisation of the missile<br />

controls is allowed for precisely this amount<br />

of time. For this to be meaningful the<br />

algorithms must be written efficiently.<br />

Critical parameters must be stored at the<br />

beginning of each cycle in case the reoptimisation<br />

is prematurely terminated.<br />

Because the algorithms are to be executed in<br />

real-time, global optimisation algorithms<br />

currently in favour such as simulated<br />

annealing and shooting methods are<br />

inappropriate. Given the recent increase in<br />

computing capacity some of the simpler<br />

early optimisation techniques have been<br />

selected.<br />

Gradient techniques have proven to be<br />

robust for real-time optimisation given a<br />

reasonable set of initial controls,<br />

<strong>Moody</strong> [M.12] . Propagating the dynamic<br />

equations using ATPN with the missile<br />

initially pointing at the target providing the<br />

initial trajectory. The optimiser state space<br />

selected to define the missile dynamics was<br />

Cartesian position, speed and direction with<br />

respect to the earth referenced Alignment<br />

frame. The terms in the cost function and<br />

adjoined dynamic constraints can be<br />

weighted depending on target characteristics<br />

providing a more flexible system. Gradient<br />

projection in function-space along the<br />

direction of steepest-descent with an Armijo<br />

univariate search minimising the cost<br />

function is the primary algorithm. If cost<br />

reduction stalls a conjugate directions are<br />

explored using an exact Fibonnacci search. The line search step-length is<br />

doubled, or halved, depending on cost function values to accelerate the<br />

search for a local minimum.<br />

7-20<br />

SELECT INITIAL SET OF<br />

ADMISSIBLE CONTROLS<br />

AND IMPACT TIME<br />

STATE EQUATION<br />

FORWARD<br />

INTEGRATION FROM<br />

CURRENT TO IMPACT<br />

TIME<br />

COST<br />

EVALUATION<br />

IMPACT POINT<br />

ADJOINTS FROM<br />

TRANSVERSALITY<br />

CONDITIONS<br />

BACKWARD<br />

INTEGRATION OF<br />

EULER-LAGRANGE<br />

EQUATIONS<br />

HAMILTONIAN<br />

VARIATION WRT<br />

CONTROLS, STATES<br />

AND ADJOINTS<br />

( GRADIENTS )<br />

GRDIENT SCALING AND<br />

SEARCH DIRECTION<br />

UNIVARIATE SEARCH<br />

AND CONTROL UPDATE<br />

WITHIN ADMISSABLE<br />

REGION<br />

COST<br />

EVALUATION<br />

PRE AND POST<br />

LAUNCH EXIT<br />

CONDITIONS<br />

Figure 7-2<br />

Optimisation Flowchart

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!