Introduction
This bi-monthly seminar series explores real-world applications of physics-informed machine learning (Φ-ML) methods to the engineering practice. They cover a wide range of topics, offering a cross-sectional view of the state of the art on Φ-ML research, worldwide.
Participants have the opportunity to hear from leading researchers and learn about the latest developments in this emerging field. These seminars also offer the chance to identify and spark collaboration opportunities.
About the event
The overarching goal of this work is to develop real-time data assimilation methods to create digital twins of thermoacoustic instabilities. Central to digital twins are experimental data, which are quantitatively accurate but noisy and sparse; and physics-based low-order models, which are computationally cheap but may contain model errors (biases). Real-time data assimilation allows models to self-adapt and self-correct any time that reference data become available. With model biases, however, traditional real-time data assimilation methods are ill-posed because the estimators are assumed unbiased. Model biases are difficult to infer because they are “unknown unknowns”, i.e., we may not know their functional form a priori.
We design a real-time digital twin framework that integrates machine learning-based bias estimation through echo state networks (which are generalized auto-regressive functions), physics-based low-order models, and experimental data. We propose a bias-aware Bayesian data assimilation framework to perform combined state, parameter, and bias estimation. The mathematical solution to the optimization problem is the Regularized bias-aware Ensemble Kalman Filter (r-EnKF). The r-EnKF enables the real-time inference of modelling parameters, states, and biases, which makes qualitative low-order models more quantitatively accurate.
The proposed framework is deployed to create a digital twin of azimuthal thermoacoustics of a hydrogen-based laboratory combustor using raw pressure data from microphones. We find that the real-time digital twin (i) autonomously predicts azimuthal dynamics, in contrast to traditional data assimilation; (ii) uncovers the physical acoustic pressure from the raw data, i.e., it acts as a physics-based filter; (iii) generalizes existing models by providing a time-varying parameter solution; and (iv) infers the model bias on the fly for all operating conditions under investigation.
This work opens new opportunities for low-order modelling and real-time digital twinning of nonlinear and multi-physics problems.
Watch on demand:
Recording will be uploaded after the seminar.
