| Return to Main
Page Predicting the Compressive JOINT
CSHRP/NEWFOUNDLAND BAYESIAN APPLICATION Introduction The Department of Works, Services and
Transportation (DWST) of the Government of Newfoundland
and Labrador agreed to participate in the Joint
CSHRP/Agency Bayesian Applications Project. By
participating in this project, the Department hoped to
find useful applications for CSHRP's Bayesian Analysis
and Performance Prediction Programs. Need for a Model Mr. Keith Foster, Manager of Materials
Engineering, suggested that a model to predict the
compressive strength of Portland cement concrete may be
useful to this Department. Knowing early whether the
concrete strengths of highway structures are going to be
low is important. This knowledge would allow for timely
adjustments to the concrete mix designs to correct any
problems before additional concrete is placed. The prediction of compressive strengths
of concrete would also permit better scheduling of
construction activities such as posttensioning,
pretensioning or removal of false work and form work.
These functions depend upon attaining a minimum level of
concrete strength. Although compressive strength tests are
usually conducted at an age of 7 days to predict the
28day strengths (fc), it was thought that it would be
worthwhile to predict strengths using information
gathered from the mix designs and quality control tests. Upon closer examination, it was decided
to sort the data and devise a model only to predict the
compressive strength of silica fume concrete. Silica fume
concrete is a high performance concrete mix that can
produce compressive strengths of 140 MPa or more. The
construction of the Hibernia Gravity Base Structure (GBS)
at Bull Arm has recently introduced this product to
Newfoundland. Staffing The lead analyst for this project was
Mr. Joe English, Pavement Management Engineer. Data
Manager for the first model iteration was Mr. Christopher
R. Walsh, a coop engineering student of Memorial
University of Newfoundland. Expert judgement was provided
by: · Mr. Keith Foster, Manager of
Materials Engineering (DWST) Valuable consulting assistance was
provided by Mr. Mark Nickeson and Mr. Lyle Kajner,
associates with VEMAX Management Inc. of Saskatoon,
Saskatchewan. Methodology A literature search of highway and
concrete publications provided several sources of data
based on the quality control testing of concrete mixes.
Additional data was collected from old DWST project files
for bridge and overpass construction. A search to find an
existing model for predicting the compressive strength of
silica fume concrete was unsuccessful. Review of the data resulted in the
rejection of material external to DWST projects because
the information was not always complete. Variables such
as water cement ratio, air content, and placement
temperature were not always reported. The selected data
for the study was obtained from DWST files from the
following three sources: · Laboratory mix designs for normal
and silica fume concrete · Mix designs and field tests for the
construction of high performance silica fume girders at
Main Brook Bridge · Mix designs and field tests for the
construction of high performance silica fume girders at
Holyrood Pond Bridge The girders were constructed of silica
fume concrete containing 8% silica fume by weight of
cement. Laboratory mix designs involved silica fume
concrete and normal portland cement concrete that
contains no silica fume. As silica fume concrete is a new
technology in Newfoundland, experts were not readily
available. Consequently, it was decided to obtain the
Prior information by the Data Prior Method. For purposes
of the analysis, the laboratory mix designs were
categorized as "old" data, while the database
for the two construction projects was categorized as
"new" data. Three iterations were carried out on the model to predict the compressive strength of silica fume concrete: For the first two models, classical
regression analysis was conducted on the laboratory mix
designs and entered in XLBAYES as Data Priors. In Model
1, mix designs involved silica fume concrete as well as
normal portland cement concrete (0 % Silica Fume). In
Model 2, it was decided to concentrate on silica fume
concrete only. An advantage of a dataderived prior
is that experts do not need to be available. However, a
Data Prior is only as good as the old data. The
laboratory mix designs were selected as Priors because
quality control is higher in the lab than in the field
(the project files). In the third model, a Zellner GPrior
was used. The Prior means were derived from experts and
the laboratory data was only used to determine a
variancecovariance analysis for XLBAYES. |
