Mechanical & Aerospace Engineering
Development of a Dual Number Automatic Differentiation (DNAD) module for nuclear reactor thermal/hydraulic sensitivity studies. Major Professor: Robert Spall
Traditional sensitivity studies use Monte Carlo simulations to approximate the relationships between design parameters and system characteristics. While this approach is sufficient for many applications, it tends to be cost-prohibitive when considering complex systems such as nuclear reactor thermal/hydraulic systems. Alternative approaches (e.g. adjoint methods and complex step derivatives) have been developed that address some of the shortcomings of the traditional Monte Carlo approach, but applying them to numerical algorithms requires extensive source code modification. Dual Number Automatic Differentiation (DNAD) is a method of performing numerical sensitivity studies through the application of advanced programming concepts. Automatic differentiation is a method of computing function derivatives based on the elementary mathematical operations used to numerically evaluate the function itself. Applying DNAD to sensitivity studies has the potential to improve the speed and accuracy of the analysis when compared to the traditional Monte Carlo approach, while requiring significantly less source code modification than the alternatives mentioned earlier.