Nagi Gebraeel, Georgia Power Associate Professor and Associate Director of the Strategic Energy Institute, is working on two projects that represent the spectrum of optimization and Big Data problems within the energy industry. The first, funded by General Electric (GE), looks at sensor data from gas turbines that power electricity generators. The turbines are large and quite expensive to both manufacture and maintain, so they are equipped with thousands of sensors constantly measuring whether the turbines are functioning within normal operation parameters by monitoring temperatures, pressures, and vibrations from different sections of the turbines in a process known as condition monitoring. Immense amounts of data are sent to Atlanta, Georgia, where GE’s monitoring and diagnostic center is located. Gebraeel’s research team and collaborators include Shabbir Ahmed, ISyE Dean’s Professor and Stewart Faculty Fellow; Kamran Paynabar, ISyE Assistant Professor; Andy Sun, ISyE Assistant Professor; Edmond Chow, CSE Associate Professor and Director of the Intel Parallel Computing Center in the College of Computing; and Polo Chou, CSE Assistant Professor and Associate Director of the M.S. in Analytics program. In response to the Big Data problem, they are developing a new computational platform to provide detection and predictive analytics for the energy industry. This platform assesses the health and performance of equipment in real-time and monitors trends to determine such things as:

• When to order new spare parts so they don’t linger in inventory, costing money and possibly becoming obsolete.
• Prescribing operational profiles that extend an asset’s life without increasing risk of failure.

By integrating detection, prediction, and optimization capabilities, the new prototype platform could help power companies achieve significant savings. Indeed, a preliminary study shows a 40 to 45 percent reduction in maintenance costs alone. In contrast to the Big Data problem of gas turbines with thousands of sensors is Gebraeel’s project analyzing Big Data from wind farms. A wind farm may have hundreds of turbines, which in turn are equipped with relatively few sensors. Thus — unlike the gas turbines — if one or even several wind turbines stop working, it has relatively little impact on the wind farm’s power production. Furthermore, maintaining wind turbines often requires sending out cranes, or ships in the case of off-shore wind farms, which can be very expensive. So, in contrast to the gas or steam turbines which need to be repaired as soon as one goes down, it’s more cost-efficient to repair several wind turbines at one time. Thus, the optimization model for wind farm maintenance focuses more on opportunistic maintenance. “They are similar in that the objective is the same, and the constraints are the same, but the dynamics of solving and optimizing them are distinctly different. The perspective on the tradeoffs between the cost of maintenance and operation are almost the reverse,” said Gebraeel. “These two settings represent the wide spectrum of configurations within the energy network.” Although the gas turbines present a setting with a high number of sensors on a smaller number of units, while the wind turbines represent a large number of units with relatively few sensors, either way, you end up with a Big Data problem. “ISyE is one of the few engineering disciplines at the intersection of engineering, statistical sciences, and operations research. For example, the analytic algorithms that we develop are all based on statistical methodologies; the optimization models are all based on operations research,” explained Gebraeel. “We are lucky to have some of the best statistics and operations research faculty here in ISyE.” An important aspect of algorithms designed to deal with Big Data in the energy sector is making sure that they are scalable. Having a scalable computing architecture means the algorithms can be scaled as well. “Sometimes we test the limits of the algorithms in the research that we do,” said Gebraeel. His research team uses both actual data from GE as well as simulated data. “Then we go to the hypothetical cases: Let’s blow the number of units and the number of sensors beyond what’s available in today’s industry to study the limits of the algorithms we have re-engineered. It helps us understand the scope of their application.”