Chapter 5 Robust Performance Tailoring with Tuning - SSL - MIT

Data: initial iterate, y0, termination tolerances, barrier parameters, ɛ, µ0
Result: optimal design variables
begin
Initialize: y = y0, k =1,µ = µ0s;
while termination conditions not met do
evaluate g (y), ∂g(y)
∂y ;
evaluate B (y) (Equation 4.6);
evaluate f (y) [Hardware test];
evaluate J = f (y)+µB (y);
calculate finite difference gradients [Hardware test];
calculate ∇B (y) and∇J (y);
calculate descent direction;
if steepest descent then
d = −∇J (y) (Equation A.3);
else if conjugate gradient then
d = −∇J (y)+βdk−1 (Equation A.15);
end
calculate step-size;
if decreasing then
αk = α0 √k (Equation A.10);
else if line minimization then
αk =argminα∈[0,s] f (yk + αdk) (Equation A.11);
end
evaluate new iterate, yk+1 = yk + αkdk;
increment barrier sequence, µk+1 = µ0ɛ k ;
increment iterate, k = k +1;
end
end
Figure 4-8: Barrier method implementation.
126