A numerical approach to determine optimal gains for active optimal vibration control of continuous structures by the finite element method is presented. The approach uses an analogy between the optimality equations for control in the time domain and the governing equations for a set of static beams in the spatial domain. The finite element model of control is generated and analyzed in a fictitious spatial domain. The results are then transferred to the time domain to describe the optimal dynamic response of the system. The analogy (referred to as the beam analogy) allows for an efficient application of the finite element method to solve two-point-boundaryvalue problems for the finite time control cases. Here it is applied to the infinite time control cases. An algorithm for a direct calculation of optimal gains for closed loop control of time-invariant problems without using Riccati's equation is discussed in detail.
IntroductionOptimal gains for active vibration attenuation are typically determined by minimizing a quadratic performance index. The corresponding Riccati equations for the problem are in the form of coupled non-linear matrix equations, difficult to solve for larger matrices [1,2]. Although several iterative algorithms have been proposed to get optimal gains from these equations, 'the solutions of the Riccati equation are the most timeconsuming part of any optimal control problem' ([1], p. 109). Besides being very intensive numerically these algorithms 'do not universally guarantee stable and accurate computation' ([2], p. 248).Potter's method was used in [3] to obtain the optimal gains for control of piezoelectric plates. In such a method the gains are determined by first solving the complex eigenvalue problem of an asymmetric matrix (referred to as Hamiltonian's matrix) formed of the matrices defining the states and the performance index (weighting matrices).A different approach, in which the gains are iteratively improved, was proposed in [4]. However, the corresponding algorithm is essentially intuitive without a proof of convergence. The approach was somewhat refined in [5].Optimal gains can also be calculated by using the general optimization techniques. In such an approach first the controls and then the states are expressed in terms of gains and substituted into the performance index. Next, treating the gains as the optimization variables, the performance is gradually improved until minimum is reached [6]. A genetic algorithm was used for that purpose in [7], while gradient-based algorithms were applied in [8-10]. The above approach is numerically very intensive. In order to facilitate its convergence, various auxiliary optimality conditions were derived and used as extra constraints while minimizing the performance. The constrained optimization process was converted into unconstrained one by using either the Lagrange multipliers [8,10], or a penalty method [9].The concept of independent modal-space control methodology was used in [11] in which the gains for each independent mode are calculat...