2024 AIChE Annual Meeting
(4cw) Optimization for Sustainable Energy Systems and Data-Driven Predictive Analytics for Smart Manufacturing
Author
I am a fifth-year Ph.D. candidate advised by Dr. George Bollas at the University of Connecticut. My research, spanning both my master's, post-master’s, and Ph.D. studies, has focused on solving complex problems in energy systems and precision manufacturing for their efficiency and sustainability, through advanced modeling and data-driven approaches. My master's and post-master’s research honed my skills in developing optimization models for complex systems, while my Ph.D. research expanded my capabilities in machine learning and predictive analytics. Throughout my academic journey, I have cultivated a deep expertise in optimization, energy supply systems, machine learning algorithms, and predictive analytics. My academic work uniquely positions me at the intersection of optimization and machine learning. It allows me to tackle multidisciplinary problems with a holistic approach, leveraging both optimization techniques and data-driven methodologies. Below, I provide a general overview of my research.
Master and Post-Masters Research: Optimization for Sustainability of Energy Systems
During my master’s degree and post-master position, my research was primarily centered on the optimization of energy systems, with a particular emphasis on sustainable renewable energy supply chains, bioethanol production under WEFL nexus, and coke oven gas utilization for energy production. Across various projects, I consistently aimed to enhance the efficiency, sustainability, and economic viability of various energy systems.
1) Optimization Techniques
- Applied optimization techniques to design and improve energy systems.
- Utilized mixed integer linear programming (MILP) technique to develop models that minimized costs or/and maximized efficiency or/and ensured the sustainability of energy systems.
- Developed MILP models in GAMS software.
2) Renewable Energy Systems
- Proposed strategies for transitioning from conventional energy systems to 100% RES-based systems in South Korea.
- Addressed various challenges from the residential level to the national level.
- Projected future energy demands and structures, using historical data, national statistics, technical factors, political considerations, energy-related variables, and so on.
- Provided a comprehensive framework for the design and operation of energy systems, energy supply chain management, and long-term investment planning.
- Evaluated major cost drivers, cost-effectiveness for CO2 reduction, and the impacts of meteorological change, energy storage, and demand structure on the energy systems.
3) Water-Energy-Food-Land (WEFL) nexus
- Proposed a comprehensive decision model for the design of bioethanol production and supply chain (BPSC) under the WEFL nexus framework.
- Addressed the interdependencies and potential conflicts between various critical resources.
- Evaluated the BPSC system by cost and nexus optimizations in terms of various aspects such as involved technology type, energy supply cost, energy consumption, and so on.
- Provided a more sustainable and resilient solution, balancing resource demands and minimizing environmental impacts.
4) Carbon Utilization Strategies
- Proposed a framework for the systematic analysis and evaluation of carbon utilization strategies, in particular, energy production from coke oven gas.
- Developed process model of technologies involved in the carbon utilization system using a process simulator (ASPAN) to estimate technical and economic parameters.
- Evaluated the carbon utilization strategies with four criteria: product quantity, energy consumption, production cost, and profit.
Ph.D. Research: Data-Driven Predictive Analytics for Smart Manufacturing
My Ph.D. research focused on developing data-driven predictive analytics for tool wear monitoring in precision machining. By leveraging sensor data from a machining process, I utilized machine learning techniques to monitor and predict the degradation of cutting tools in real-time or near-real-time.
1) Signal Processing and Feature Engineering
- Processed sensor signals with filtering or denoising methods to remove noise and irrelevant information and enhance the quality of the signals.
- Extracted meaningful features from processed signals using statistical measures, spectral analysis, wavelet coefficients analysis, or distance metric.
- Identified patterns indicative of tool wear.
2) Fault Detection and Diagnosis
- Developed data-driven models for the classification of wear states, fault detection, and accurate diagnosis of tool wear.
- Implemented various machine learning or deep learning algorithms: SVM, KNN, and CNN for the classification/ SVR, linear regression, decision-tree, gaussian process, and neural network for regression.
- Evaluated impacts of multi-sensor usage and fusion of multi-domain features on the accuracy of tool wear monitoring systems
- Utilized Python and MATLAB software.
3) Tool Wear Prognosis and RUL Prediction
- Developed surrogate models using symbolic regression to predict the remaining useful life (RUL) of cutting tools.
- Compared with prognosis results of Kalman filter and Interacting Multiple Model (IMM) filter models.
- Utilized Automated Learning of Algebraic Models (ALAMO) and MATLAB software.
4) Physics-Informed Machine Learning
- Proposed a hybrid modeling framework integrating data-driven models with physics domain knowledge to develop robust and accurate predictive models.
- Developed a recursive symbolic regression model that integrates both current and historical data for accurate predictions of tool wear.
- Developed a generic model for tool wear predictions across different experimental cases.
- Employed genetic programming for symbolic regression to iteratively refine model structures, enhancing their predictive power and interpretability.
- Utilized GPTIPS which is MATLAB toolbox for genetic programming.