Improving outcomes using Design of Experiments
Amanda French
Chair
Johns Hopkins University Applied Physics Laboratory
Monday, Aug 4: 2:00 PM - 3:50 PM
0611
Topic-Contributed Paper Session
Music City Center
Room: CC-103C
Applied
Yes
Main Sponsor
Section on Statistics in Defense and National Security
Presentations
We build on the work of Neyer in using sequential design to more efficiently identify particular percentiles of interest. We also consider the problem of simultaneously minimizing the variance of multiple outputs with a single design.
Supersaturated Screening Designs (SSDs) are used for factor screening when there are more experimental factors than there are experimental runs. In this case, penalized estimation is commonly applied to analyze these experiments. This work proposes a framework for optimal SSDs in the context of the exact probability of lasso support recovery.
The DOD regularly acquires new, large, complex systems that require reliability testing. Although traditional experimental designs methods can be used to test reliability, each trial of the system may cost millions of dollars. This expense has led to an increasing desire to use results from similar, previous experiments through carefully crafted Bayesian priors. We discuss our refinement of current discrete-time methods, derive a fast approximation for quick analysis, and demonstrate results with an example system.
You have unsaved changes.