13.07.2015 Views

Assessment and Future Directions of Nonlinear Model Predictive ...

Assessment and Future Directions of Nonlinear Model Predictive ...

Assessment and Future Directions of Nonlinear Model Predictive ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

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

174 H.G. Bock et al.to yield not only an improvement <strong>of</strong> feasibility, but also <strong>of</strong> optimality for theoriginal NLP (8).The remaining online computations are slightly more expensive than for levelsB <strong>and</strong> C, as we need to recover the multipliers λ QPk,µQP k<strong>of</strong> the uncondensedQP (14) for the transition step 3, as follows:First, the inequality multipliers µ QPkare directly obtained as the multipliersµ cQPk<strong>of</strong> the condensed QP (16): µ QPk:= µ cQPk. Second, the equality multipliersλ QPkcan be computed as λ QPk:= (BS T ) −T S(A∆w k + a k − C T µ QPk)whereS isa projection matrix that maps w to its subvector s.The matrix BS T contains only those columns <strong>of</strong> B that correspond to thevariables s, cf Eq. (10), <strong>and</strong> is thus invertible. Abbreviating a := S(A∆w k +a k − C T µ QPk), a =(ax 0 ,az 0 ,...,ax N ), we can compute λQPk=(λ x 0 ,λz 0 ,...,λx N )recursively backwards: Starting with λ x N := ax N , we compute, for i = N − 1,N−2,...,0:λ z i =(Z z i ) −T ( a z i +(X z i ) T λ x i+1), λxi = a x i +(X x i ) T λ x i+1 − (Z x i ) T λ z i .where we employ the submatrix notation <strong>of</strong> Eq. (10) for the matrix B, respectivelyBS T . The pro<strong>of</strong> <strong>of</strong> nominal stability <strong>of</strong> NMPC based on this variant followsthe lines <strong>of</strong> the pro<strong>of</strong> for the st<strong>and</strong>ard scheme mentioned in section 3.3.5 A Real-Time NMPC ExampleTo demonstrate the real-time applicability <strong>of</strong> the NMPC schemes discussedabove, a simulation experiment has been set up. In this experiment, a chain<strong>of</strong> massive balls connected by springs is perturbed at one end, <strong>and</strong> the controltask is to bring the system back to steady state.An ODE <strong>Model</strong> for a Chain <strong>of</strong> Spring Connected MassesConsider the following nonlinear system <strong>of</strong> coupled ODEsẍ i + β ẋ i − 1 m (F i+ 1 − F 2 i− 1 ) − g =0, i =1, 2,...,N−1 (18a)2()F i+ 1 S L1 −(x2 i+1 − x i ), i =0, 1,...,N−1. (18b)‖x i+1 − x i ‖x 0 (t) ≡ 0, ẋ N (t) =u(t), (18c)for the ball positions x 0 (t),...,x N (t) ∈ R 3 with boundary conditions (18c) <strong>and</strong> aprescribed control function u(t) ∈ R 3 . Equations (18a)–(18c) describe the motion<strong>of</strong> a chain that consists <strong>of</strong> eleven balls (i.e. N = 10), numerated from 0 to 10,that are connected by springs. At one end, the first ball is fixed in the origin, thevelocity <strong>of</strong> the other end (the ”free” end) is prescribed by the function u. Themotion <strong>of</strong> the chain is affected by both laminar friction <strong>and</strong> gravity (gravitationalacceleration g =9.81 m/s 2 ) as an external force. The model parameters are: massm =0.03 kg, spring constant S =1N/m, rest length <strong>of</strong> a spring L =0.033 m,

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

Saved successfully!

Ooh no, something went wrong!