Large-scale linear dynamic systems generate high-dimensional models. Engineers therefore apply model-order reduction techniques. These methods create smaller models that retain essential behaviour.
Several techniques dominate current practice. Modal truncation keeps the most important vibration modes. Balanced truncation ranks states by controllability and observability energy. Krylov subspace methods match moments of the transfer function. Proper orthogonal decomposition extracts dominant spatial patterns from simulation data. Each approach reduces system size while aiming to preserve key dynamics.
Accuracy forms the first evaluation criterion. Researchers compare reduced-model outputs with full-order responses. They measure errors in time-domain trajectories and frequency-response functions. Relative error norms quantify the difference. In addition, they examine how well the reduced model captures resonance peaks and transient behaviour. Higher reduction ratios usually increase approximation error. Careful selection of retained states therefore improves accuracy.
Stability requires equal attention. A reduced model must remain stable if the original system is stable. Some methods guarantee stability preservation under certain conditions. Balanced truncation offers strong theoretical guarantees for linear systems. Other techniques may introduce unstable poles. Analysts therefore check eigenvalue locations after reduction. They also verify energy dissipation properties. Consequently, stability tests form a standard part of the evaluation process.
Real-time performance determines practical usefulness. Reduced models must run fast enough for control, monitoring, or interactive simulation. Computational cost drops significantly with lower system order. Memory requirements also decrease. Engineers measure execution time on target hardware. They further assess whether the reduced model meets real-time constraints under sampling rates of interest. Moreover, they evaluate parallelisation potential and numerical robustness.
Trade-offs appear between the three criteria. Aggressive reduction improves speed but may harm accuracy or stability. Moderate reduction often provides a better balance. Researchers therefore perform parametric studies. They vary reduction order and compare resulting accuracy, stability margins, and timing results. Different application domains demand different priorities. Control applications emphasise stability and speed. High-fidelity simulation places greater weight on accuracy.
Current studies combine theoretical analysis with numerical benchmarks. Standard test systems allow fair comparison across methods. Advanced error estimators guide adaptive reduction. Overall, careful evaluation of accuracy, stability, and real-time performance enables reliable use of reduced-order models in engineering practice.