System identification based on characteristic curves: a mathematical connection between power series and Fourier analysis for first-order nonlinear systems

TL;DR

This paper extends the SORPS formalism for system identification by introducing a power series-based approach that overcomes previous limitations, enabling application to arbitrary datasets and broadening its use in nonlinear system analysis.

Contribution

It introduces a new power series-based method for system identification using characteristic curves, expanding applicability beyond sinusoidal inputs and equal time steps.

Findings

Method applies to arbitrary datasets

Overcomes FFT limitations for single-tone signals

Enables broader nonlinear system analysis

Abstract

Recently, the sinosoidal output response in power series (SORPS) formalism was presented for system identification and simulation. Based on the concept of characteristic curves (CCs), it establishes a mathematical connection between power series and Fourier series for a first-order nonlinear system [F. J. Gonzalez, Sci. Rep. 13, 1955, (2023)]. However, the system identification procedure discussed there, based on fast Fourier transform (FFT), presents the limitations of requiring a sinusoidal single tone for the dynamical variable and equally spaced time steps for the input-output dataset (DS). These limitations are here addressed by introducing a different approach: we use a power series-based model for system modeling using two hyperparameters $\hat{A}_0$ and $\hat{A}_1$ optimally defined depending on the DS. Overall, this work expands the applicability of the SORPS formalism to arbitrary DSs and represents a groundbreaking contribution for the concept of CCs. The method of CCs is complementary to the commonly used approaches of NARMAX-models and sparse regression techniques, which emphasize the estimation of the individual parameter values of the model. However, the CCs-based methods emphasize the computation of the CCs as a whole, thus presenting the advantages that the system identification is uniquely defined, and that it can be applied for any system without additional algebraic operations. Thus, the parsimonious principle defined by the NARMAX-philosophy is extended from the concept of a model with as few parameters as possible to the concept of finding the lowest model order that correctly describe the input-output DS. This opens up a wide variety of potential applications in vibration analysis, structural dynamics, viscoelastic materials, nonlinear electric circuits, voltammetry techniques, structural health monitoring, and fault diagnosis.