Optimal Model Order Reduction Technique for Dynamic Motor Vehicle Suspension Systems

High-fidelity full-vehicle suspension models with hundreds to thousands of states are computationally prohibitive for real-time control and virtual prototyping. Existing model reduction techniques distribute approximation effort uniformly across the entire frequency spectrum, producing unnecessarily large errors in the (6–10)-Hz band where human sensitivity to vertical acceleration is highest. This paper introduces the frequency-limited two-sided iterative algorithm, a novel H2-optimal projection framework that concentrates reduction accuracy within a user-defined comfort-relevant interval ω1,ω2 by embedding frequency-weighting filters analytically into the two-sided iteration, thereby avoiding the expensive Gramian computations required by frequency-limited balanced truncation. The framework is validated on a benchmark 1000-state full-vehicle lumped-parameter model that incorporates a flexible chassis with vibrational modes, wheels with asymmetric left-side bump excitation, and detailed driver/passenger seat dynamics. A 23-state FLTSIA reduced model achieves markedly lower errors than IRKA, TSIA, FLBT, and FLIRKA within the prescribed [5, 15] rad/s band, reduces CPU simulation time, and shrinks memory usage. These results demonstrate that FLTSIA enables reliable real-time active suspension control and streamlined virtual prototyping for next-generation vehicle platforms.