The main goal of our laboratory is to solve various real-world problems by utilizing data acquired sensors and employing data-driven optimization/control approaches. The target systems that we are interested in are mainly complex engineering systems comprising of many subsystems interacting each other and, at the same time, interacting with stochastic environments.
Renewable energy systems
The performance of renewable energy systems, such as wind farms and solar plants, strongly depends on their immediate operational strategies under varying environmental conditions. In addition, the performance of renewable energy systems degrades continuously over time. The key question, then, is how to optimally manage and operate such systems to increase the system performances (i.e., energy producing-efficiency and life cycle cost).
- For control, we are developing and validating various sequential decision-making algorithms (i.e., data-driven control algorithms) to increase the performance of the wind farm under stochastically varying wind conditions. The figures below shows the implementation of the Bayesian Ascent algorithm to controlling the scaled wind farm in a wind tunnel laboratory.
- For optimum maintenance, we are developing a condition-based maintenance scheme that can optimally schedule the maintenance based on the measured performance of the system. For example, solar panels’ repair schedule can be determined by investigating the variation in the power production performance under various environmental conditions. We will develop a data-driven diagnostic/prognostic model for various energy systems.
- For operation, we are researching about the optimum operation and management of renewable energy systems that are connected to an energy grid, in which the intermittency of the energy source and the stochastic variation of energy load need to be well-balanced. As a way to resolve the imbalance between the energy supply and demand, our laboratory is developing a way to jointly operate a wind farm and an energy storage system. We are exploring various ways to derive the optimum operation of the energy storage system, mainly focusing on data-driven approaches, such as approximated dynamic programming and reinforcement learning.
Smart manufacturing systems
An enormous amount of data is currently being collected from modern construction and manufacturing processes, providing great opportunities to use the data to better understand these complex processes.
- Condition-based maintenance is to diagnose or prognosis a component or machine failure using sensor data collected from the target systems. The figure bellow shows an example of detecting tool wear by analyzing the power consumption data.
- We employ a data-driven approach to understanding better not only the individual operation but also the complex interactions among multiple operations, in an attempt to achieve an efficient use of resources (i.e., scheduling), effective maintenance (i.e., condition-based monitoring), and process control to improve the quality and productivity of the process.
Population density in urban environments is rapidly increasing and expected to be doubled by 2050. Our laboratory is researching into the ways of finding causal relationships among human activities in cities and the performances of various urban systems, such as the traffic network, energy grid, and public transportation, by applying spatiotemporal analysis to urban big data. The main goal is to understand how urban systems perform while interacting with humans under different spatial and temporal contexts.
- Various spatiotemporal analysis tools will be explored and developed to analyze the performances of various urban systems. The enhanced understanding can then be used to optimally operate such systems and to make long-term planning and policy decisions. The figure below shows the operation pattern of New York city taxes depending on time and location, which is drawn by analyzing a large volume of taxi operation data.
- Our laboratory is also highly interested in analyzing the spatial and temporal variations of energy usage by users. The energy usage pattern then can be used to optimally operate energy storage systems connected with energy generation systems. Furthermore, it can be used to design a (cooperative) load shifting scheme for the optimum utilizations of the energy source and infrastructures.