LSTM-Based Classification of Time Series Data for Predicting Thermoacoustic Instability Regimes
Session Number
Project ID: CMPS 06
Advisor(s)
Dr. Debolina Dasgupta, Dr. Chandrachur Bhattacharya, Argonne National Laboratory
Discipline
Computer Science
Start Date
17-4-2024 8:15 AM
End Date
17-4-2024 8:30 AM
Abstract
In this research project, predictive capabilities of Long Short-Term Memory (LSTM) models for identifying thermoacoustic instability within combustion systems are explored. LSTMs, a subset of Recurrent Neural Networks, are particularly effective in overcoming the vanishing gradient problem, thus enhancing their ability to learn from and classify time series data. As a part of this project, a machine learning model will be developed that employs LSTMs for time series based- classification. This will be used to distinguish between time series data of different types, or “classes”. The initial approach employs these networks to classify hand-crafted datasets that emulate the conditions indicative of stable behavior and instability. This approach will focus on 3-4 types of time series data, using the LSTM model to distinctly classify them. This year's work lays the groundwork for applying the methodology to actual experimental and numerical data in the following year. As a part of this project, a machine learning model employing LSTMs for time series- classification will be developed to distinguish between time series from different classes. The expected outcome is a robust model capable of accurately classifying the stability regime of the Rijke tube, thereby contributing to the design and optimization of safer, more efficient combustion systems.
LSTM-Based Classification of Time Series Data for Predicting Thermoacoustic Instability Regimes
In this research project, predictive capabilities of Long Short-Term Memory (LSTM) models for identifying thermoacoustic instability within combustion systems are explored. LSTMs, a subset of Recurrent Neural Networks, are particularly effective in overcoming the vanishing gradient problem, thus enhancing their ability to learn from and classify time series data. As a part of this project, a machine learning model will be developed that employs LSTMs for time series based- classification. This will be used to distinguish between time series data of different types, or “classes”. The initial approach employs these networks to classify hand-crafted datasets that emulate the conditions indicative of stable behavior and instability. This approach will focus on 3-4 types of time series data, using the LSTM model to distinctly classify them. This year's work lays the groundwork for applying the methodology to actual experimental and numerical data in the following year. As a part of this project, a machine learning model employing LSTMs for time series- classification will be developed to distinguish between time series from different classes. The expected outcome is a robust model capable of accurately classifying the stability regime of the Rijke tube, thereby contributing to the design and optimization of safer, more efficient combustion systems.