Student: Omar Arafa
Committee: Dr. Lenore Dai and Dr. Heather Emady
Abstract:
Design of Experiments (DoE) is a technique of forming decision frameworks and overall outcome of a process and integral to the Six Sigma methodology. Selecting a robust design that is representative of the data spread has a direct impact on the accuracy of a model. Traditional methods utilize the least squares method for predictive models; however nonlinear datasets are generally misrepresented using these techniques. Neural networks are beneficial in modeling of nonlinear datasets as well as datasets that have non-numeric inputs. Coupled with larger data pools, they can lead to more accurate models than those achieved with the least squares method. To demonstrate the applicability of neural networks, a sample nonlinear dataset for predicting the compressive strength of concrete is validated. A five percent improvement of the R squared coefficient between the predicted and actual result can be achieved with a neural network. This gain can translate to significant boosts in cost savings and efficiency.
Zoom Room:https://asu.zoom.us/j/2012179501
Presentation Time: 12:00-1:00 PM (Arizona Time)
