2015 CSCE Annual Conference Regina - Building on our Growth Opportunities

2015 CSCE Annual Conference Regina - Building on our Growth Opportunities Conference


Title
Improved Sensitivity Analysis in Civil Engineering Design Using Statistical Design of Experiment (DOE) Methodologies

Return to Session

Author(s)
Dr. Leonard Lye, Memorial Uiversity of Newfoundland (Presenter)
Dr. Amgad Hussein, Memorial University of Newfoundland
Mr. Rajib Dey, Memorial University of Newfoundland
Abstract

An engineering analysis or design is seldom complete without conducting a sensitivity analysis. Sensitivity or what-if analysis is the assessment of the consequences of changes due to uncertainty in input factors and/or model parameters on the response of interest, not taking into account information on the probability of these changes.  By quantifying the sensitivity, one can make design decisions by knowing which input factors have the most influence on the output or response. In practice, changing one factor or one parameter at a time is perhaps the most popular and well-known method for conducting sensitivity analysis. This approach has been advocated and described in textbooks on engineering design as well as engineering economics. The advantage of this method is that it is easy apply and the results are easy to understand. However, the one-factor-at-a-time or OFAT method is known to be inefficient and ineffective and in some cases lead to wrong results. In this paper, an improved sensitivity analysis based on statistical design of experiments (DOE) in combination with regression analysis is introduced using several examples from geotechnical and structural engineering, and an example from engineering economics. This paper will show that with the availability of modern DOE software, the DOE approach is easy to apply, efficient, and interpret. The DOE approach is in fact a combined sensitivity and scenario analysis. It will also be shown that the DOE approach provides better information for decision-making than the OFAT and the stochastic risk analysis approaches.