This paper examines and compares the commonly used machine learning algorithms in their performance in interpolation and extrapolation of flame describing function (FDFs), based on experimental and simulation data. Algorithm performance is evaluated by interpolating and extrapolating FDFs and then the impact of errors on the limit cycle amplitudes are evaluated using the extended FDF (xFDF) framework. The best algorithms in interpolation and extrapolation were found to be the widely used cubic spline interpolation, as well as the Gaussian processes (GPs) regressor. The data itself were found to be an important factor in defining the predictive performance of a model; therefore, a method of optimally selecting data points at test time using Gaussian processes was demonstrated. The aim of this is to allow a minimal amount of data points to be collected while still providing enough information to model the FDF accurately. The extrapolation performance was shown to decay very quickly with distance from the domain and so emphasis should be put on selecting measurement points in order to expand the covered domain. Gaussian processes also give an indication of confidence on its predictions and are used to carry out uncertainty quantification, in order to understand model sensitivities. This was demonstrated through application to the xFDF framework.
In this paper we investigate how supervised machine learning can be leveraged to improve online prediction of thermoacoustic combustion instabilities using dynamic pressure readings. We discuss the current state of the art tools for the online detection of combustion instabilities, namely the Hurst exponent and Auto-Regressive model, the precursors they detect and how they go about doing so. We show that the generality of these tools comes at the cost of predictive power, which can be recovered using supervised machine learning. To demonstrate this we apply two different supervised machine learning approaches (using Hidden Markov Models and Automatic Machine Learning) to the classification of the state of a combustor given the dynamic pressure readings. We then observe the changes in predictive power when different information is added or removed from the signal. We find that the HMM based approach reduces to the AR model when the signal is normalised. We also find that the performance of a model trained on a signal transformed using the Detrended Fluctuation Analysis (DFA) can be met by a model trained on the Hurst exponent and the DFA transformation at a single (short) scale.
Modern, low emission combustion systems with improved fuel-air mixing are more prone to combustion instabilities and therefore use advanced control methods to balance minimum NOx emissions and and the presence of thermoacoustic combustion instabilities. The exact operating conditions at which the system encounters an instability is uncertain because of sources of stochasticity, such as turbulent combustion, and the influence of hidden variables, such as un-measured wall temperatures or differences in machine geometry within manufacturing tolerances. Practical systems tend to be more elaborate than laboratory systems and tend to have less instrumentation, meaning that they suffer more from uncertainty induced by hidden variables. In many commercial systems, the only direct measurement of the combustor comes from a dynamic pressure sensor. In this study we train a Bayesian Neural Network (BNN) to predict the probability of onset of thermoacoustic instability at various times in the future, using only dynamic pressure measurements and the current operating condition. We show that, on a practical system, the error in the onset time predicted by the BNNs is 45% lower than the error when using the operating condition alone and more informative than the warning provided by commonly used precursor detection methods. This is demonstrated on two systems: (i) a premixed hydrogen/methane annular combustor, where the hidden variables are wall temperatures that depend on the rate of change of operating condition, and (ii) full scale prototype combustion system, where the hidden variables arise from differences between the systems.
This paper examines and compares commonly used Machine Learning algorithms in their performance in interpolation and extrapolation of FDFs, based on experimental and simulation data. Algorithm performance is evaluated by interpolating and extrapolating FDFs and then the impact of errors on the limit cycle amplitudes are evaluated using the xFDF framework. The best algorithms in interpolation and extrapolation were found to be the widely used cubic spline interpolation, as well as the Gaussian Processes regressor. The data itself was found to be an important factor in defining the predictive performance of a model, therefore a method of optimally selecting data points at test time using Gaussian Processes was demonstrated. The aim of this is to allow a minimal amount of data points to be collected while still providing enough information to model the FDF accurately. The extrapolation performance was shown to decay very quickly with distance from the domain and so emphasis should be put on selecting measurement points in order to expand the covered domain. Gaussian Processes also give an indication of confidence on its predictions and is used to carry out uncertainty quantification, in order to understand model sensitivities. This was demonstrated through application to the xFDF framework.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.