Hybrid system identification for high-performance structural control

Structural systems installed with active or semi-active control devices usually require availability of a high-fidelity model to determine appropriate control designs. Use of traditional system identification techniques has proven to be challenging, primarily due to the fact that such models must be multiple-input and multiple-output (MIMO) systems; the inputs correspond to the excitation and the control commands, while the outputs correspond to the measured responses. Even if a model is identified that can represent well the response of the structure, this model is often a non-minimal realization (i.e., the dynamics of various modes are duplicated in the model); these extra dynamics can degrade the performance of the resulting controller. This paper presents a hybrid system identification approach that can result in minimal (or near-minimal) realizations of the MIMO structure-control system models that are effective for structural control design. The first step in this approach develops a simplified model which can portray adequately the system characteristics found in the experimental data. Using information about the pole-zero relationship of the transfer function in the simplified model, a frequency-domain system identification strategy is employed subsequently to derive multiple single-input and multiple-output (SIMO) models with respect to each system input. The systems are then combined to determine the complete MIMO system model that is accurate over the frequency range of interest for the control applications. To demonstrate this hybrid approach, an example is provided to illustrate a high-fidelity identification result from the experiment of an actively isolated building system. Successful control implementation demonstrates the efficacy of this hybrid system identification approach.Structural systems installed with active or semi-active control devices usually require availability of a high-fidelity model to determine appropriate control designs. Use of traditional system identification techniques has proven to be challenging, primarily due to the fact that such models must be multiple-input and multiple-output (MIMO) systems; the inputs correspond to the excitation and the control commands, while the outputs correspond to the measured responses. Even if a model is identified that can represent well the response of the structure, this model is often a non-minimal realization (i.e., the dynamics of various modes are duplicated in the model); these extra dynamics can degrade the performance of the resulting controller. This paper presents a hybrid system identification approach that can result in minimal (or near-minimal) realizations of the MIMO structure-control system models that are effective for structural control design. The first step in this approach develops a simplified model which can portray adequately the system characteristics found in the experimental data. Using information about the pole-zero relationship of the transfer function in the simplified model, a frequency-domain system identification strategy is employed subsequently to derive multiple single-input and multiple-output (SIMO) models with respect to each system input. The systems are then combined to determine the complete MIMO system model that is accurate over the frequency range of interest for the control applications. To demonstrate this hybrid approach, an example is provided to illustrate a high-fidelity identification result from the experiment of an actively isolated building system. Successful control implementation demonstrates the efficacy of this hybrid system identification approach.