Spatial balanced model reduction for flexible structures (Q1295089)

The authors begin with the model of a vibrating beam using the Euler-Bernoulli model \[ \partial^2/\partial\xi^2[EI(\xi) \partial^2y/\partial\xi^2]+ \rho A\partial^2y/\partial t^2= W(t,\xi)\tag{1} \] with deflection \(y(t,\xi)\) and bending moment \([EI(\xi) \partial^2y/ \partial\xi^2]\) set to zero at both ends of the beam, that is, assuming a free support at both ends. Assuming natural modes to be functions \(\phi_i(\xi)\) which are orthonormal with respect to the weight \((\rho A)\), we can write the solution in the form \[ y(t,\xi)= \sum_i q_i(t)\phi_i(\xi).\tag{3} \] Now, we can regard these modes \(\phi_i(\xi)\) as solutions of ordinary differential equations resulting from substituting equality (3) into (1). In an energy conserving system the corresponding transfer function is \[ [y'(s,\xi)/W(s)]= [\rho A]^{-1} \sum\phi_i(\xi_\delta) \phi_i(\xi)/(s^2+ \omega^2_i), \] where \(\xi_\delta\) is the location of a unit point load (Dirac delta) and \(\xi\) is the location where we measure the deflection. If dissipation is considered, the authors add a term \(2\gamma_i s\omega_i\) to the denominator, where \(\gamma_i\) is the damping factor. The reviewer comments that more realistic assumptions concerning energy dissipation involve fractional order derivatives as outlined for example in the papers of Bagley et al., resulting in fractional order derivatives of the Laplace transform variable \(s\) in the transfer function. The system studied by the authors has obviously infinitely many modes. In practical solutions this is approximated by a finite-dimensional expansion. The beam control model is reduced to the form \[ dx/dt= Ax(t)+ Bu(t),\;y(t,\xi)= C(\xi) x(t),\;y_m= C(\xi_m)x(t),\tag{2} \] where \(y_m\) is the observed value of \(y\). The cost functional is the integral of the function \(y^2+ Ru^2\) over all positive time and space occupied by the beam. The principal theorem states that in a stable system (2) the linear map between inputs and outputs has the \(L_2\) norm of the transfer function equivalent to the \(H_\infty\) norm induced between spaces of inputs and outputs. The authors present the concept of balanced truncation, containing the output of the system of a finite-dimensional space. Such a truncation is shown to preserve the basic properties of the original infinite-dimensional model, and specifically it also assures that the upper bound of the \(H_\infty\) norm of the error does exist.