A Design Approach Based on Historical Test Data and Bayesian Statistics

Abstract

Testing is still the final verification for a design even though there are substantial number of analytical and simulation methods available. Testing is seen to be also an indispensable part in the foreseeable future. Numerous test data have been generated in many testing institutions over the years and it is clear that future new tests will be conducted. Historical data with similar design and operating conditions can shed light on the current and future designs since they would share some common features when the changes are not dramatic. To effectively utilize the historical data for future design, two steps are necessary: (1) finding an approach to consistently correlate test data obtained from various conditions; (2) Use of Bayesian statistics which can provide a rational mathematical tool for extracting useful information from the historical data. In this paper, the basic Bayesian statistical procedure based on the historical data is outlined. With this information the reduction of sample size number or improving the accuracy and confidence with the same sample size are becoming possible. Examples of utilizing the historical data are also presented and the benefit of using the Bayesian statistics are highlighted.