Detection and Analysis of Combustion Instability From Hi-Speed Flame Images Using Dynamic Mode Decomposition
12 Oct 2016-Vol. 1
About: The article was published on 2016-10-12. It has received 15 citations till now. The article focuses on the topics: Diffusion flame & Combustion.
Citations
More filters
01 Jan 2020
TL;DR: This work extracts sequential image frames from high-speed images of a premixed, bluff-body stabilized flame which exhibits varying levels of combustion instability, and applies an efficient detection framework (based on 2-D convolutional neural networks) to detect the growth of an unstable mode, which can lead to effective control schemes.
Abstract: Combustion instabilities are prevalent in a variety of systems including gas turbine engines. In this regard, the introduction of active control opens the potential for new paradigms in combustor design and optimization. However, the limited ability to detect the onset of instabilities can lead to difficulty in implementing active control approaches. Machine learning—specifically deep learning tools—may be employed to detect instabilities from various measurement and sensor data related to the combustion process. Deep learning models have recently shown remarkable potential for extraction of meaningful features from data without the need to hand-craft. As one of the early studies of deep learning for combustion instability detection, we extract sequential image frames from high-speed images of a premixed, bluff-body stabilized flame which exhibits varying levels of combustion instability. Using an efficient detection framework (based on 2-D convolutional neural networks) to detect the growth of an unstable mode can lead to effective control schemes. In addition, we apply a second deep learning framework to capture the temporal correlations in the data with corresponding learned spatial features.
15 citations
01 Jun 2021
TL;DR: A novel deep learning architecture called 3D convolutional selective autoencoder (3D-CSAE) is proposed to detect the evolution of self-excited oscillations using spatiotemporal data and clearly shows performance improvement in detecting the precursors and the onset of instability.
Abstract: While analytical solutions of critical (phase) transitions in dynamical systems are abundant for simple nonlinear systems, such analysis remains intractable for real-life dynamical systems. A key example is thermoacoustic instability in combustion, where prediction or early detection of the onset of instability is a hard technical challenge, which needs to be addressed to build safer and more energy-efficient gas turbine engines powering aerospace and energy industries. The instabilities arising in combustion chambers of engines are mathematically too complex to model. To address this issue in a data-driven manner instead, we propose a novel deep learning architecture called 3D convolutional selective autoencoder (3D-CSAE) to detect the evolution of self-excited oscillations using spatiotemporal data, i.e., hi-speed videos taken from a swirl-stabilized combustor (laboratory surrogate of gas turbine engine combustor). 3D-CSAE consists of filters to learn, in a hierarchical fashion, the complex visual and dynamic features related to combustion instability from the training videos (i.e., two spatial dimensions for the image frames and the third dimension for time). We train the 3D-CSAE on frames of videos obtained from a limited set of operating conditions. We select the 3D-CSAE hyper-parameters that are effective for characterizing hierarchical and multiscale instability structure evolution by utilizing the dynamic information available in the video. The proposed model clearly shows performance improvement in detecting the precursors and the onset of instability. The machine learning-driven results are verified with physics-based off-line measures. Advanced active control mechanisms can directly leverage the proposed online detection capability of 3D-CSAE to mitigate the adverse effects of combustion instabilities on the engine operating under various stringent requirements and conditions.
14 citations
12 Nov 2018
TL;DR: With the continuous variation of the control parameter, this work can successfully detect the critical transition to a state of high combustion instability demonstrating the robustness of the proposed detection framework, which is independent of the combustion inducing protocol.
Abstract: Detecting the transition to an impending instability is important to initiate effective control in a combustion system. As one of the early applications of characterizing thermoacoustic instability using Deep Neural Networks, we train our proposed deep convolutional neural network (CNN) model on sequential image frames extracted from hi-speed flame videos by inducing instability in the system following a particular protocol — varying the acoustic length. We leverage the sound pressure data to define a non-dimensional instability measure used for applying an inexpensive but noisy labeling technique to train our supervised 2D CNN model. We attempt to detect the onset of instability in a transient dataset where instability is induced by a different protocol. With the continuous variation of the control parameter, we can successfully detect the critical transition to a state of high combustion instability demonstrating the robustness of our proposed detection framework, which is independent of the combustion inducing protocol.
9 citations
01 Sep 2021
TL;DR: The critical flame location is found to be in the flame root region and is effective in the early detection of incipient LBO in a realistic gas turbine engine combustor under engine-relevant conditions.
Abstract: A data-driven approach using machine learning is presented for the identification of the critical flame location for the early detection of an incipient lean blowout (LBO) in a realistic gas turbine engine combustor under engine-relevant conditions. This method is demonstrated by utilizing the temperature ( T ) and the hydroxyl radical mass fraction ( Y O H ) data from high fidelity large eddy simulations (LES) of Jet-A combustion. The fuel flow rate is progressively reduced in numerical simulations with a fixed airflow rate to mimic experimental studies of LBO in the gas turbine combustor. These simulations are the first of their kind for a fully resolved realistic combustor geometry with adaptive mesh refinement and have accurately captured the dynamics of the LBO process and global lean blowout equivalence ratio. Time-series of T and Y O H are extracted in the primary zone of the combustor, from stable flame condition to LBO condition, to train the machine learning model. A Support Vector Machine (SVM) model with radial basis function is successfully developed to identify the critical flame location for early detection of incipient LBO condition in a practical combustor for the first time. The performance of the SVM model is quantified using the F-score, and the critical flame location corresponds to the maximum value of the F-score. The critical flame location is found to be in the flame root region and is effective in the early detection of incipient LBO. The conventional statistical measures are compared with the results of the trained machine learning model to assess the feasibility of the latter for online flame health monitoring. The machine learning model successfully prognosticated the LBO approximately 20 ms before the event, and this study has shown significant promise for the use of the SVM model in engine prognostics and health management.
9 citations
TL;DR: In this paper, the combustion oscillation characteristics of a supersonic ethylene jet flame in a hot coflow were investigated utilizing a 5 kHz high-speed hydroxyl planar laser-induced fluorescence (OH-PLIF) technique and an advanced postprocessing method, namely, dynamic mode decomposition (DMD).
7 citations
References
More filters
TL;DR: In this article, a method is introduced that is able to extract dynamic information from flow fields that are either generated by a (direct) numerical simulation or visualized/measured in a physical experiment.
Abstract: The description of coherent features of fluid flow is essential to our understanding of fluid-dynamical and transport processes. A method is introduced that is able to extract dynamic information from flow fields that are either generated by a (direct) numerical simulation or visualized/measured in a physical experiment. The extracted dynamic modes, which can be interpreted as a generalization of global stability modes, can be used to describe the underlying physical mechanisms captured in the data sequence or to project large-scale problems onto a dynamical system of significantly fewer degrees of freedom. The concentration on subdomains of the flow field where relevant dynamics is expected allows the dissection of a complex flow into regions of localized instability phenomena and further illustrates the flexibility of the method, as does the description of the dynamics within a spatial framework. Demonstrations of the method are presented consisting of a plane channel flow, flow over a two-dimensional cavity, wake flow behind a flexible membrane and a jet passing between two cylinders.
4,150 citations
01 Jan 2002
TL;DR: A broad survey of combustion research can be found in this article, where a number of closed loop feedback concepts are used to improve the combustion process as demonstrated by applications to automotive engines.
Abstract: Combustion dynamics constitutes one of the most challenging areas in combustion research. Many facets of this subject have been investigated over the past few decades for their fundamental and practical implications. Substantial progress has been accomplished in understanding analysis, modeling, and simulation. Detailed laboratory experiments and numerical computations have provided a wealth of information on elementary dynamical processes such as the response of flames to variable strain, vortex rollup, coupling between flames and acoustic modulations, and perturbed flame collisions with boundaries. Much recent work has concerned the mechanisms driving instabilities in premixed combustion and the coupling between pressure waves and combustion with application to the problem of instability in modern low NO x heavyduty gas turbine combustors. Progress in numerical modeling has allowed simulations of dynamical flames interacting with pressure waves. On this basis, it has been possible to devise predictive methods for instabilities. Important efforts have also been directed at the development of the related subject of combustion control. Research has focused on methods, sensors, actuators, control algorithms, and systems integration. In recent years, scaling from laboratory experiments to practical devices has been achieved with some successebut limitations have also been revealed. Active control of combustion has also evolved in various directions. A number of experiments on laboratory-scale combustors have shown that the amplitude of combustion instabilities could be reduced by applying control principles. Full-scale terrestrial application to gas turbine systems have allowed an increase of the stability margin of these machines. Feedback principles are also being explored to control the point of operation of combustors and engines. Operating point control has special importance in the gas turbine field since it can be used to avoid operation in unstable regions near the lean blowoff limits. More generally, closed loop feedback concepts are useful if one wishes to improve the combustion process as demonstrated by applications to automotive engines. Many future developments of combustion will use such concepts for tuning, optimization, and emissions reduction. This article proposes a broad survey of these fast-moving areas of research.
726 citations
TL;DR: The dynamic mode decomposition is a data-decomposition technique that allows the extraction of dynamically relevant flow features from time-resolved experimental data and image-based flow visualizations and is demonstrated on data from a numerical simulation of a flame based on a variable-density jet and on experimentalData from a laminar axisymmetric water jet.
Abstract: The dynamic mode decomposition (DMD) is a data-decomposition technique that allows the extraction of dynamically relevant flow features from time-resolved experimental (or numerical) data. It is based on a sequence of snapshots from measurements that are subsequently processed by an iterative Krylov technique. The eigenvalues and eigenvectors of a low-dimensional representation of an approximate inter-snapshot map then produce flow information that describes the dynamic processes contained in the data sequence. This decomposition technique applies equally to particle-image velocimetry data and image-based flow visualizations and is demonstrated on data from a numerical simulation of a flame based on a variable-density jet and on experimental data from a laminar axisymmetric water jet. In both cases, the dominant frequencies are detected and the associated spatial structures are identified.
292 citations
TL;DR: In this article, the first Hopf bifurcation of the flow past a circular cylinder is analyzed using the Koopman operator and the dynamic mode decomposition (DMD) algorithm.
Abstract: The Koopman operator provides a powerful way of analysing nonlinear flow dynamics using linear techniques. The operator defines how observables evolve in time along a nonlinear flow trajectory. In this paper, we perform a Koopman analysis of the first Hopf bifurcation of the flow past a circular cylinder. First, we decompose the flow into a sequence of Koopman modes, where each mode evolves in time with one single frequency/growth rate and amplitude/phase, corresponding to the complex eigenvalues and eigenfunctions of the Koopman operator, respectively. The analytical construction of these modes shows how the amplitudes and phases of nonlinear global modes oscillating with the vortex shedding frequency or its harmonics evolve as the flow develops and later sustains self-excited oscillations. Second, we compute the dynamic modes using the dynamic mode decomposition (DMD) algorithm, which fits a linear combination of exponential terms to a sequence of snapshots spaced equally in time. It is shown that under certain conditions the DMD algorithm approximates Koopman modes, and hence provides a viable method to decompose the flow into saturated and transient oscillatory modes. Finally, the relevance of the analysis to frequency selection, global modes and shift modes is discussed.
284 citations
08 Jul 1985
TL;DR: In this paper, the determination of an internal feedback mechanism which leads to combustion instability inside a small scale laboratory combustor is presented, and the experimental findings show that a large vortical structure is formed at an acoustic resonant mode of the system.
Abstract: The determination of an internal feedback mechanism which leads to combustion instability inside a small scale
laboratory combustor is presented in this paper. During combustion instability, the experimental findings show that a large vortical structure is formed at an acoustic resonant mode of the system. The subsequent unsteady burning, within the vortex as it is convected downstream,
feeds energy into the acoustic field and sustains the large resonant oscillations. These vortices are formed when the
acoustic velocity fluctuation at the flameholder is a large fraction of the mean flow velocity. The propagation of
these vortices is not a strong function of the mean flow speed and appears to be dependent upon the frequency of the
instability. Continued existence of large vortical structures which characterize unstable operation depends upon the fuel-air ratio, system acoustics, and fuel type.
147 citations