| Return to Main
Page Predicting the Compressive JOINT
CSHRP/NEWFOUNDLAND BAYESIAN APPLICATION
Conclusions The evolution of the compressive
strength prediction model shows that Bayesian statistics
works and works well. Model 3 has shown that soliciting
expert judgement is sometimes more valuable than
collecting data. Without Bayesian statistics, the Data
regression for Model 1 would have been chosen as the
regression equation. Sensitivity analysis shows that the
Data regression does not emphasize the importance of the
silica fume concrete variable probably because it was not
coded properly as a categorical variable. Model 2 shows that additional data
isn't always helpful. The Prior data made the coefficient
for fine to total aggregate ratio change its sign.
Although the laboratory data was assumed to be collected
under better quality control conditions, the observations
regarding the fine to total aggregate ratio do not agree
with convention. By adding expert judgement, the
coefficients became more realistic and added information
to the prediction of compressive strengths. However, it
must be remembered that this prediction equation is
acceptable only to the concrete mixes investigated in
this report having a Silica Fume content of 8 %. Changing
any of the variables beyond the working ranges
investigated in this report may change the model
significantly. Presently, silica fume concrete is not
being used on any highway construction projects. Should a
contractor wish to achieve high early strengths, the
option is there to use silica fume concrete. However, it
is recommended that laboratory work should be continued
to determine if other factors are affecting compressive
strength and to solve the conflict about the fine to
total aggregate ratio coefficient. The program XLBAYES was found to be
user friendly. Documentation is easy to understand.
Changes could be made easily and quickly to the
predictions as the Prior, Data and Posterior worksheets
are linked. Incorporating the program into Microsoft
Excel made manipulation of data simple. Bibliography Walsh, C. R., Factors Influencing
fhe Compressive Strength of Normal Portland Cement and
HighPerformance Silica Fume Concretes, Prepared for
Department of Works, Services and Transportation,
Government of Newfoundland and Labrador, St. John's,
Canada, October 1994 Canadian Strategic Highway Research
Program, Training Sessions in Bayesian Methods and
BSTA T. Software, Ottawa, Canada, June 1994 Canadian Strategic Highway Research
Program, Bayesian Software Beta Test Training, Ottawa,
Canada, May 1992 Canadian Strategic Highway Research
Program, MidCourse Workshop on Analysis and
Interpretation of the First Iteration Model Results, Ottawa,
Canada, May 1995 McClave, J.T. and Benson, P.G., Statistics
For Business and Economics, Fourth Edition, Dellen
Publishing Company, San Francisco, U.S.A., 1988
|