Data-driven detection of drifting system parameters

TL;DR

This paper presents a machine learning approach to detect and analyze changes in system parameters over time using observational data, demonstrated on a turbulent fluid flow experiment.

Contribution

It introduces a method combining data-driven modeling and first principles analysis to identify parameter drifts in complex systems from observational data.

Findings

Successfully identified governing equations for turbulent flow

Detected variations in physical parameters over time

Demonstrated effectiveness in complex, real-world data

Abstract

Data taken from observations of the natural world or laboratory measurements often depend on parameters which can vary in unexpected ways. In this paper we demonstrate how machine learning can be leveraged to detect changes in global parameters from variations in an identified model using only observational data. This capability, when paired with first principles analysis, can effectively distinguish the effects of these changing parameters from the intrinsic complexity of the system. Here we illustrate this by identifying a set of governing equations for an experiment generating a weakly turbulent fluid flow, then analyzing variation in the coefficients of these equations to unravel the drift in its physical parameters.