Date of Award


Degree Type


Degree Name

Doctor of Philosophy (PhD)


Electrical Engineering and Computer Science


Can Isik


Bayesian Network, Data Fusion/Mining, Decision Support System, Particle Swarm Optimization, Sensor Management, Wind Energy Diagnostics and Prognostics

Subject Categories

Electrical and Computer Engineering


Data fusion technologies are widely applied to support a real-time decision-making in complicated, dynamically changing environments. Due to the complexity in the problem domain, artificial intelligent algorithms, such as Bayesian inference and particle swarm optimization, are employed to make the decision support system more adaptive and cognitive. This dissertation proposes a new data fusion model with an intelligent mechanism adding decision feedback to the system in real-time, and implements this intelligent data fusion model in two real-world applications.

The first application is designing a new sensor management system for a real-world and highly dynamic air traffic control problem. The main objective of sensor management is to schedule discrete-time, two-way communications between sensors and transponder-equipped aircraft over a given coverage area. Decisions regarding allocation of sensor resources are made to improve the efficiency of sensors and communications, simultaneously. For the proposed design, its loop nature takes account the effect of the current sensor model into the next scheduling interval, which makes the sensor management system able to respond to the dynamically changing environment in real-time. Moreover, it uses a Bayesian network as the mission manager to come up with operating requirements for each region every scheduling interval, so that the system efficiently balances the allocation of sensor resources according to different region priorities. As one of this dissertation's contribution in the area of Bayesian inference, the resulting Bayesian mission manager is shown to demonstrate significant performance improvement in resource usage for prioritized regions such as a runway in the air traffic control application for airport surfaces.

Due to wind's importance as a renewable energy resource, the second application is designing an intelligent data-driven approach to monitor the wind turbine performance in real-time by fusing multiple types of maintenance tests, and detect the turbine failures by tracking the turbine maintenance statistics. The current focus has been on building wind farms without much effort towards the optimization of wind farm management. Also, under performing or faulty turbines cause huge losses in revenue as the existing wind farms age. Automated monitoring for maintenance and optimizing of wind farm operations will be a key element in the transition of wind power from an alternative energy form to a primary form. Early detection and prediction of catastrophic failures helps prevent major maintenance costs from occurring as well. I develop multiple tests on several important turbine performance variables, such as generated power, rotor speed, pitch angle, and wind speed difference. Wind speed differences are particularly effective in the detection of anemometer failures, which is a very common maintenance issue that greatly impacts power production yet can produce misleading symptoms. To improve the detection accuracy of this wind speed difference test, I discuss a new method to determine the decision boundary between the normal and abnormal states using a particle swarm optimization (PSO) algorithm. All the test results are fused to reach a final conclusion, which describes the turbine working status at the current time. Then, Bayesian inference is applied to identify potential failures with a percentage certainty by monitoring the abnormal status changes. This approach is adaptable to each turbine automatically, and is advantageous in its data-driven nature to monitor a large wind farm. This approach's results have verified the effectiveness of detecting turbine failures early, especially for anemometer failures.


Open Access