Prediction and Optimization of Heavy Clay Products Quality - Advanced Materials for Agriculture, Food and Environmental Safety - page 97

case of CaO content to WLF (0.928-0.942) and WA (0.860-0.883). MgO
content is mostly inl uenced by the same dependent parameters, but in a
lower extent then CaO (0.591-0.633). h is is because of the fact that CaO
and MgO are mainly hosted with carbonates, which lower CS and increase
WA and WLF . h e next signii cant inl uence observed is relation of SiO 2
to WLF (around -0.81) and WA (around -0.68). Actually, SiO 2 is pres-
ent within the clay minerals, but also as free quartz, which is mostly i ne-
grained since it had a positive ef ect on CS [2, 25, 26]. Relatively high and
negative correlations are obtained for Fe 2 O 3 - WA (from -0.618 to -0.643)
and Fe 2 O 3 - WLF (around -0.62) due to the increase in CS . Taking all this
into account CS is mostly inl uenced by CaO (negative impact) and Al 2 O 3
content (positive impact).

4.3.2 Analysis of Variance and SOP Models

Analysis of variance (ANOVA) was conducted for obtained SOP models,
and outputs were tested against the impact of input variables, in table 4.3.
ANOVA was performed to demonstrate the inl uences of the inputs on
the outputs, and these are quantii ed by the sum of squares (SOS). Results
concerning WA and CS were shown elsewhere [3], while here testing of
the SOP models for FS , WLF , and VMC is presented. All the variables
that entered into the analysis showed a signii cant ef ect on the outputs,
either as linear or square members, or in the form of a product of two
variables. CS and WA were the most sensitive to the chemical composi-
tion, compared to all other physico-mechanical parameters that charac-
terize the i nal products. Furthermore, the coei cients of determination
( r 2 ) were used to explore the coni dence level of the models. SOPs showed
high r 2 values (0.897-0.913 for compressive strength models, 0.942-0.962
for water absorption, 0.928 for i ring shrinkage, 0.988-0.991 for water loss
during i ring, and 0.941 for volume mass of cubes models). All variables
considered in the ANOVA analysis were also used for the ANN modeling
( See sections 4.3.3 and 4.3.4 ).
h e parameters that most inl uence the compressive strength, statisti-
cally signii cant at p<0.05 (coni dence level of 95%), were the quadratic
terms of temperature and CaO contents, whereas the value of CSB was
the most sensitive to the content of CaO and CSC to the i ring tempera-
ture. h e third-largest impact on CS in both cases was SiO 2 xCaO term.
ANOVA, for most of the observed parameters, showed higher inl uence
in the case of CSC than CSB, coni rming that the compressive strength
of cubes changes more drastically, especially with the i ring temperature.
h e highest ef ect on water absorption showed the content of CaO (square