The historical developments, parametrization, and numerous applications of algebraic numerical differentiation methods initiated by Fliess, Mboup, Join, and Sira-Ramírez are presented. These numerical non-asymptotic approximation approaches for higher-order derivatives of noisy signals are suited for real-time embedded systems. The different derivation approaches are reviewed according to their historical developments. Links to numerous established numerical differentiation algorithms like the Savitzky-Golay filters are discussed. Recent tuning approaches that are reviewed facilitate a parametrization for a good estimation accuracy without relying on trial-and-error approaches, which is essential for industrial applications. The use of these guidelines is demonstrated using two concrete examples: Identification of system parameters from noisy measurements and numerical inversion of the dynamics of analogue anti-aliasing filters. Moreover, an extensive literature survey with various applications of these methods by the control community in all engineering areas is presented. The problems solved include, but are not limited to, parameter estimation, state reconstruction, feedback control, fault diagnosis, anomaly detection, fault-tolerant control, and model-free control. A MATLAB and Python toolbox implementing all necessary functions for the design, analysis, and discretization of the filters is made available, for which a link is provided. Its use is demonstrated in a short appendix.
Algebraic differentiators have attracted much interest in recent years. Their simple implementation as classical finite impulse response digital filters and systematic tuning guidelines may help to solve challenging problems, including, but not limited to, nonlinear feedback control, model-free control, and fault diagnosis. This contribution introduces the open source toolbox AlgDiff for the design, analysis and discretisation of algebraic differentiators.
This thesis investigates numerical differentiation methods based on orthogonal polynomials and applies the methods to fault detection and parameter estimation problems. Time-domain and frequency-domain analyses of the differentiators are provided. Tuning guidelines and real-time implementations are proposed.
We derive a numerically stable method to compute an image representation of an unknown linear system only from data, leveraging a continuous-time version of Willems et al.’s fundamental lemma. To this end, we use derivatives approximated by algebraic differentiators. Our novel image representation avoids solving differential-algebraic equations and significantly reduces computational complexity by eliminating redundant degrees of freedom corresponding to the number of unknown quantities to be identified. Simulation results confirm the effectiveness of the proposed approach, even in the presence of severe measurement disturbances.