New spatial curve method classifies transient regimes in dynamic systems
A data-driven approach using arc length outperforms existing classifiers in multivariate systems.
A team of researchers led by Cristian Puerto-Santana and Carlos Ocampo-Martinez (Universitat Politècnica de Catalunya) has introduced a new data-driven methodology for classifying transient and stationary regimes in dynamic systems. The approach, detailed in a preprint on arXiv (2606.17347 [eess.SY]), addresses a critical gap in existing sensor-based regime classification: many current methods require multiple parameter settings and fail in multivariate systems containing periodic signals.
The key innovation is representing the system's behavior as a spatial curve constructed from its sample mathematical moments. By connecting concepts from stability theory, differential geometry of curves, and stationary stochastic processes, the team designed two regime classifiers based on the arc length and curvatures of this curve. Both classifiers can detect transient regimes under multivariate asymptotically stable, marginally stable, and cyclostationary conditions. In performance comparisons, the arc-length classifier notably outperformed conventional techniques on simulated linear, nonlinear, and discontinuous multivariate systems, offering a parameter-light, computationally efficient alternative for real-time monitoring of complex dynamic systems.
- Uses a spatial curve derived from sample moments to represent multivariate system dynamics.
- Two classifiers based on arc length and curvature outperform existing methods in simulated tests.
- Handles linear, nonlinear, and discontinuous systems with periodic signals without complex parameter tuning.
Why It Matters
Enables robust, automated transient detection in industrial control systems and robotics without manual tuning.