CHAPTER 3: EXPERIMENTAL AND THEORETICAL...
Transcript of CHAPTER 3: EXPERIMENTAL AND THEORETICAL...
27
CHAPTER 3: EXPERIMENTAL AND THEORETICAL INVESTIGATIONS
3.1 Experimental Investigations
In this part, a detailed description of soil analysis for the subgrade is discussed. It includes
laboratory tests like Gradation (sieve and hydrometer analysis), Atterberg limits, Moisture
density relationships and curves, Maximum dry weight, Optimum moisture content, Specific
gravity and California Bearing Ratio for Soaked and unsoaked samples and filed tests like
Benkelman beam deflection measurement for the finished layer with and without
bioenzymatic stabilization. Samples for the present research work were taken from ten
different areas in and around Thanjavur district. Fifteen numbers of samples were taken with
different composition. The sample locations are Vallam, Budalur, Chennampatti,
Aalavandaan Keni, Munnayampaatti, Mathakkottai, Kuruvadi from Thanjavur district and
Marunkulam from Pudukkottai district.
The clay content of the soils from these 15 locations varies from 4% to 30%. If these
materials are directly used for the formation of roads it attributes to the failure of pavement.
The strength of the local soil can be improved by stabilization. It results in higher CBR
values which in turn reduction in the thickness of the pavement layers for the given load. In
the present research work an attempt has been made to improve the strength of the soil by
adding a particular type of bio-enzyme called Terrazyme.
3.2 Methodology
Study area was chosen as in and around Thanjavur and Pudukkottai districts, in which 15
locations were fixed to collect the soil samples. Samples were collected with sufficient
quantities from each and every location to carryout the laboratory tests. All the samples were
tested for its index properties such as particle size, Atterberg’s limits and strength properties
such as proctor compaction test, unconfined compressive strength and California Bearing
Ratio. CBR tests were done for unsoaked and soaked conditions. The soaking periods
adopted for the present work are 2 weeks and 4 weeks. All the specimens were tested with
28
and without Terrazyme addition to study the effect of bioenzymatic stabilization. Four
different dosages were tried with samples for assessing the strength properties of given
samples.
3.3 Laboratory Tests
3.3.1 Grain size distribution analysis
The grain size distribution is found by mechanical analysis. The components of soils which
are coarse grained were analyzed by sieve analysis and the soil fines by sedimentation
analysis. The grain size analysis or the mechanical analysis is hence carried out to determine
the percentage of individual grain size present in a soil sample.
This test is aim at the determination of the size of the particles of the soil consists. The test
result is illustrated by the grain size distribution curve. In case of the particles with size larger
than 75 µm (retained on sieve no. 200), the particle sizes are determined by “Sieve
Analyses”. While for the particles of the sizes smaller than 75 µm (passed through sieve no.
200), the particle sizes are determined by “Hydrometer Analyses”. The apparatus for grain
size distribution are shown in Figure 3.1.
Figure 3.1 Grain Size Distribution Apparatus
29
In case of the particle with cohesion as the lateritic soil, the sieve analysis method with water
is applied. The soil is put onto a sieve and rinsed by spraying water until water is clear. The
soil sample which retained on the sieve is then air-dried and the particle size is determined by
dry-sieve again. The total mass percent of grains within the selected sieve sizes are
calculated. Subtracting these percentages successively from 100% will give the ordinates of
the grain-size distribution curve. The grain size analysis (coarse and fine) are tabulated and
are given in Table 3.1.
Table 3.1 Grain size distribution analysis - Values
Type of soil Gravel (%) Sand (%) Silt (%) Clay (%)
Soil 01 18.00 62.00 16.00 4.00
Soil 02 17.60 37.40 20.00 25.00
Soil 03 18.40 41.60 20.00 20.00
Soil 04 22.80 62.90 3.37 9.97
Soil 05 6.00 82.00 4.00 8.00
Soil 06 3.00 86.00 6.00 5.00
Soil 07 17.20 44.40 17.40 21.00
Soil 08 11.20 27.80 31.00 30.00
Soil 09 7.20 41.80 24.00 27.00
Soil 10 27.60 48.40 12.00 12.00
Soil 11 9.20 67.40 10.90 12.50
Soil 12 14.40 39.40 20.00 26.00
Soil 13 12.80 75.70 1.50 10.00
Soil 14 13.60 60.80 13.00 12.60
Soil 15 14.00 75.50 1.50 9.00
The details of experimental setup for conducting sedimentation analysis of selected soil
samples are given in Figures 3.2 and 3.3.
30
Figure 3.2 Effect of Hydrometer Immersion
Figure 3.3 Samples Tested in Hydrometer
3.3.2 Specific Gravity
A systematic study of the specific gravity of soils is required for the determination of voids
ratio, degree of saturation which is a very important for compaction point of view. The results
of the tests for different types of samples are tabulated in Table 3.2.
31
Table 3.2 Specific gravity of selected samples
Type of soil Specific Gravity
Soil 01 1.98
Soil 02 2.55
Soil 03 2.25
Soil 04 2.08
Soil 05 2.11
Soil 06 2.15
Soil 07 2.11
Soil 08 1.93
Soil 09 2.15
Soil 10 2.25
Soil 11 2.38
Soil 12 1.61
Soil 13 1.63
Soil 14 1.54
Soil 15 1.45
3.3.3 Atterberg’s Limit
These tests are performed to determine liquid limit, plastic limit and the plasticity of cohesive
soil in order to characterize its condition by water content. The apparatuses for Atterberg’s
Limit Test are shown in Figure 3.4.
Figure 3.4 Liquid Limit and Plastic Limit apparatus
32
Liquid Limit is the water content, normally in percentage, at which the two sides of a groove
Cut in the soil sample contained in the cup of a Casagrande device would touch over a length
of 12 mm after 25 impacts. The procedures of the test are summarized as follow: the soil to
be tested is put in the porcelain bowl, and gradually mixed with distilled water to create a
thick paste. About 50- 80 g of the paste is then transferred to the cup of the liquid limit
device, to a thickness of 12 mm. The surface is leveled by using the spatula and the groove is
then made in soil by using the trenching tool. When the cup is placed onto the shaft and the
whole apparatus is set on a felt pad and crank is then rotated at the rate of about 2 revolutions
per second until the two surfaces separated by the groove touch each other at the bottom of
the cup along an uninterrupted length of 12 mm. The number of impacts required is recorded.
Finally 8-10 g of the paste taken from the groove area is transferred into a can to determine
the water content by drying oven. The test is performed with 4 different water contents. By
plotting the number of blows along the horizontal axis on a logarithmic scale and the water
content along the vertical axis on an arithmetic scale, the relevant points are connected and
this line will give the liquid limit at its intersection with the vertical of the 25 number of
blows. Plastic Limit means the minimum water content, in percentage, of a soil at which
threads of 3-4 mm thickness can be rolled out without crumbling. About 15 g is rolled out
into the thin threads on the glass plate with outstretched fingers. The threads are folded and
rolled out again until at a diameter of 3-4 mm, they start to crumble. For checking the
diameter a short metal rod should be kept at hand. The untreated soil samples consistency
limits test results are tabulated in Table 3.3.
33
Table 3.3 Atterberg’s limits of selected samples
Soil L.L % P.L % P.I % IS Classification
Soil 01 36 29 7 GW
Soil 02 38 32 6 ML – OL
Soil 03 46 40 6 GC
Soil 04 49 43 6 GP
Soil 05 28 22 6 SM – GM
Soil 06 30 25 5 GP
Soil 07 28 22 6 SC
Soil 08 35 30 5 CL-ML
Soil 09 42 36 6 ML
Soil 10 60 54 6 MH – OH
Soil 11 30 25 5 OL
Soil 12 35 27 8 CL
Soil 13 30 21 9 CI
Soil 14 26 17 9 OH
Soil 15 25 17 8 CL – ML
3.3.4 Compaction test
In order to achieve required strengths needed for a long lasting pavement, the properties of
the material in place in the field must simulate what is designed in the laboratory. This is
primarily achieved by compaction to increase the density of the soil. It should also be noted
that compaction not only affects the shear strength of the soil in place, but also the flow
behaviour through the soil (i.e. coefficient of permeability). A high degree of compaction will
minimize settlement and volume change. This is associated with the increase of strength of
the compacted soil layers. As well, compaction of a soil in either wet of optimum or dry of
optimum conditions will affect the soil structure. This soil structure can have dramatic effects
with regards to shear strength and the coefficient of permeability. As such, it is important to
keep control over the compaction in the field to ensure that the soil is at the appropriate
34
design water content. This will ensure the resulting soil structure influence on the engineering
behaviour of soil is known. Inconsistencies also occur in design as a soil is usually only
compacted to 95-98% compaction in the field, in comparison to the 100% compaction
achieved in the laboratory. These soils, supposedly at optimum water content or slightly wet
of, often are compacted on the dry side resulting in a soil structure not accounted for in
design. The details of experimental setup for modified Proctor compaction test are given in
Figure 3.5.
In dynamic or impact compaction, soil is compacted under the blows of a hammer dropped
from a specific height. The soil samples were tested for standard proctor test and modified
proctor test. These tests were performed as per IS: 2720. The results of the tests are presented
in the Table 3.4.
Figure 3.5 Modified Proctor Compaction test apparatus
35
Table 3.4 Maximum Dry Density with Optimum Moisture Content
Type of soil OMC % MDD kN / m3
Soil 01 12 14.53
Soil 02 11 17.3
Soil 03 10 15.79
Soil 04 12 15.5
Soil 05 11 14.03
Soil 06 12 14.22
Soil 07 11 16.2
Soil 08 11 16.2
Soil 09 10 15.7
Soil 10 11 15.8
Soil 11 12 15.6
Soil 12 9.09 14.22
Soil 13 10 14.23
Soil 14 10 14.13
Soil 15 10 14.22
3.4 Strength tests
3.4.1 Unconfined compression Test (UCC)
A native or borrowed soil is compacted to form the sub-grade of a pavement. The compacted
soil that forms the pavement is typically in a state of unsaturated condition. The compacted
soil has a negative pore water pressure, uw, and the pore-air pressure, ua, is typically equal to
the atmospheric pressure conditions. In other words, the matric suction, (ua - uw), is equal to
the negative pore water pressure. The shear strength of unsaturated soils can be interpreted
using the unconfined compression test results extending the shear strength equation for
unsaturated soils. The pore air and the pore water pressures are not measured in a
conventional unconfined compression test during the shearing stage. The shear strength of the
soil can be interpreted in terms of initial matric suction values. The matric suction of the soil
36
specimen can decrease, increase or remain relatively constant during the shearing stage.
However, matric suction is likely to slightly decrease in field compacted samples as the pore
air pressure slightly increases due to compression without significant changes in the pore
water pressures. In other words, the matric suction in soil specimens at failure conditions in
the unconfined compression tests will be slightly less then the initial matric suction. Due to
this reason, it is quite probable that the determined shear strength will be a conservative value
from the unconfined compression test results. For the laboratory prepared soil samples, the
soil was compacted at the desired moisture prior to the compression tests. The cylinders were
placed on the compression testing machine and a load was applied at a constant rate. The
maximum load at failure was recorded. The compressive strength of the sample is calculated
by dividing the maximum load by the cross-sectional area. The experimental setup for the
unconfined compressive strength is given in Figure 3.6.
Figure 3.6 Unconfined compressive strength test apparatus
37
3.4.2 California Bearing Ratio test
The CBR test was originally developed by O.J. Porter for the California Highway
Department during the 1920s. It is a load-deformation test performed in the laboratory or the
field, whose results are then used with an empirical design chart to determine the thickness of
flexible pavement, base, and other layers for a given vehicle loading. Though the test
originated in California, the California Department of Transportation and most other highway
agencies have since abandoned the CBR method of pavement design. In the 1940s, the US
Army Corps of Engineers (USACE) adopted the CBR method of design for flexible airfield
pavements. The USACE and USAF design practice for surfaced and unsurfaced airfields is
still based upon CBR today (US Army, 2001). The CBR determination may be performed
either in the laboratory, typically with a recomputed sample, or in the field. Because of
typical logistics and time constraints with the laboratory test, the field CBR is more typically
used by the military for design of contingency roads and airfields. The thickness of different
elements comprising a pavement is determined by CBR values. The CBR test is a small scale
penetration test in which a cylindrical plunger of 3 in2 (5 cm in diameter) cross-section is
penetrated into a soil mass (i.e., sub-grade material) at the rate of 0.05 in. per minute (1.25
mm/minute). Observations are taken between the penetrations resistances (called the test
load) versus the penetration of plunger. The penetration resistance of the plunger into a
standard sample of crushed stone for the corresponding penetration is called standard load.
The California bearing ratio, abbreviated as CBR is defined as the ratio of the test load to the
standard load, expressed as percentage for a given penetration of the plunger.
CBR = (Test load/Standard load)×100
In most cases, CBR decreases as the penetration increases. The ratio at 2.5 mm penetration is
used as the CBR. In some case, the ratio at 5 mm may be greater than that at 2.5 mm. If this
occurs, the ratio at 5 mm should be used. The CBR is a measure of resistance of a material to
38
penetration of standard plunger under controlled density and moisture conditions. The test
procedure should be strictly adhered if high degree of reproducibility is desired. The CBR
test may be conducted in re-moulded or undisturbed specimen in the laboratory. The test is
simple and has been extensively investigated for field correlations of flexible pavement
thickness requirement. The test apparatus figure 3.7 is shown below.
Figure 3.7 California Bearing Ratio test apparatus
Test Procedure: The CBR test is carried out on a compacted soil in a CBR mould 150 mm in
diameter and 175 mm in height, provided with detachable collar of 50 mm and a detachable
perforated base plate. A displacer disc, 50 mm deep to be kept in the mould during the
specimen preparation, enables a specimen of 125 mm deep to be obtained. The moulding dry
density and water content should be the same as would be maintained during field
compaction. To simulate worst moisture condition of the field, the specimens are kept
submerged in water for about 4 days before testing. Generally, CBR values of both soaked as
well as unsoaked samples are determined. Both during soaking and penetration test, the
39
specimen are covered with equal surcharge weights to simulate the effect of overlying
pavement or the particular layer under construction. Each surcharge slotted weight, 147 mm
in diameter with a central whole 53 mm in diameter and weighing 2.5 kg is considered
approximately equivalent to 6.5 cm of construction. A minimum of two surcharge weights
(i.e. 5kg surcharge load) is placed on the specimen. Load is applied on the penetration piston
so that the penetration is approximately 1.25mm/min. The load readings are recorded at
penetrations, 0, 0.5, 1, 1.5, 2, 2.5, 3, 3.5, 4, 4.5, 5, 5.5, 6, 6.5, 7, 8, 9, 10, 11, 12, and 12.5mm.
The maximum load and penetration is recorded if it occurs for a penetration of less than 12.5
mm. The curve is mainly convex upwards although the initial portion of the curve may be
concave upwards due to surface irregularities. A correction is then applied by drawing a
tangent to the curve at the point of greatest slope. The corrected origin will be the point where
the tangent meets the abscissa.
40
3.4.3 CBR Swell test
To stimulate worst moisture conditions of the fields, the specimens are kept submerged in
water for 4 days before testing (soaked CBR). Both during soaking and the penetration test
the specimen is covered with equal surcharge weights to simulate the effect of overlying
pavement on the particular layer under construction. The results of the penetration test are
plotted in the form of a penetration-load graph. The results of the CBR test (soaked and un
soaked) are tabulated in subsequent chapters.
3.5 Field Tests
Based on the results obtained in the laboratory a test track has been constructed for a length
of 45m in SASTRA University campus. A Non destructive testing using Benkelman Beam
deflection studies were carried out on the test track to measure the rebound deflection due to
the applied load after second week and fourth week of construction. This chapter enlightens
the details of this study.
3.5.1 Scope
This test procedure covers the determination of the rebound deflection of a pavement under a
standard wheel load and tire pressure, with or without temperature measurements.
Figure 3.8 Diagram showing the dimensions of Benkelman beam deflection device
41
3.5.2 Equipment
Basic equipment shall consist of:
The beam must be fitted with a satisfactory locking device designed to secure the beam when
moving to a new site and a suitable vibrator mounted at the pivot point. In sunny weather the
beam may pass from shade into sunshine as the vehicle moves away. Therefore a shield
should be used.
A truck or trailer with an axle load of 8.20 ±0.15 tonnes equally distributed on two dual tyred
wheels operating at the inflation pressure necessary to give a tyre contact area of 0.048 ±
0.0002 m2.* The tyres shall preferably be 10.00 x 20, 12 ply with tubes and rib treads.
A tyre pressure gauge graduated in 20 Kpa divisions or smaller.
A thermometer with a range of 0-6°C in 1°C divisions.
A mandrel suitable for making a 100mm deep hole in the pavement for inserting the
thermometer. The diameter of the hole should be 13mm.
A can containing either glycerol or oil for filling the thermometer hole.
Figure 3.9 Photo showing deflection measurement using Benkelman beam
42
3.5.3 Procedure
Deflections shall be measured as follows:
(a) The test point shall be preselected and marked. For highway pavements, test points
shall be located at the distances from the edge of the lane given in Table 3.5.
Table 3.5 Selection of test points
Lane width (m) Distance from the edge (m)
2.8 or less
3.0
3.2
3.4
3.6 or more
0.5
0.6
0.7
0.8
0.9
(b) The tyre pressure should be checked before the first test and then at intervals not
exceeding three hours.
(c) The truck shall initially be positioned with the test wheel between 100 and 150mm to the
rear of the test spot, i.e. position A.
(d) The probe of the beam shall be inserted between the dual tyres of the test wheel with the
toe located on the test spot.
(e) The locking device shall be released and the rear of the beam adjusted so that the plunger
is in contact with the dial gauge.
(f) The dial gauge shall be set to read between 9 and 11mm (the actual reading need not be
recorded) and the vibrator set in operation.
(g) The truck shall be moved forward at creep speed so that the test wheel passes over the test
spot and continues advancing to position 8 which is 2.7 ± 0.1 metres beyond the test spot.
(h) The START READING, S, is the maximum dial gauge reading occurring during this
movement of the truck from position A to position B, and will normally occur as the wheel
passes over the test spot. This reading shall be recorded.
43
(I) The INTERMEDIATE READING, I, is that figure indicated by the dial gauge at the
moment the truck stops with the test wheel in position B. This reading shall be recorded.
(j) The truck shall be moved forward until the test wheel is in position V which is not less
than 10 metres from position B.
(k) The FINAL READING, F, is that figure indicated by the dial gauge when the truck has
stopped in position C. This figure shall be recorded.
Temperature measurements must be made when the top layer of the pavement consists of
40mm or more of bitumen bound material. The following procedure should be followed:
(a) A hole should be made with the mandrel to a depth of 40mm or to such a depth that it
does not break through the bitumen bound material.
(b) The hole should be filled with glycerol or oil and the thermometer inserted.
(c) The temperature should be recorded at least hourly, or at decreasing time intervals down
to 15 minutes when successive temperatures differ by more than 3°C.
No beam readings should be made outside the pavement temperature range of 5°C to 30°C
when the top layer of the pavement consists of 40mm or more of bitumen bound material.
3.5.4 Calculation
The rebound deflection of the pavement shall be calculated in the following manner:
(a) Two pavement rebound indicators shall be established by subtracting the intermediate and
final readings from the start reading, i.e.: (S - I) and (S - F)
(b) If the indicators so obtained agree within 0.03mm the true rebound deflection at
temperature T shall be calculated as: XT = 2(S-F)
(c) If the indicators (S - I) and (S - F) differ by more than 0.03mm, the initial shape of the
bowl has been such as to influence the front support legs of the instrument and the
calculations shall be adjusted as follows:
X(T) = 2(S - F) + 5.82 (I - F)
44
(d) The pavement rebound deflection at a standard temperature of 20°C shall be calculated
from the above figure by applying the formula:
X 20 = XT + [(20 – t)/110]
Where X20 = temperature corrected rebound deflection in millimetres
and t = temperature in degrees Celsius 40mm below the surface of the pavement
3.5.5 Reporting
All reports shall include the following:
(a) The test location (preferably by SH No, Route Position, lane, and distance from lane
edge).
(b) The actual axle load used.
(c) The rebound deflection of the pavement (to 0.01mm).
(d) The date and time of readings.
Where temperature readings are required:
(a) The pavement temperature.
(b) The depth at which the temperature is recorded.
(c) The time the temperature is taken.
3.6 Enzymes as a stabilizer
Terrazyme is a Natural, nontoxic liquid, formulated using vegetable extracts. It improves
engineering properties of soil. This organic stabilizer is suitable to all types of soil Enzymes
catalyze reactions between the clay and the organic cations and accelerate the cationic
exchange process to reduce adsorbed layer thickness. The use of enzyme as an organic
stabilizer enhances the replacement of conventional granular base and granular sub base
which is commonly practiced in the highway project.
45
3.7 Properties of Terrazyme
Table 3.6 Properties of Terrazyme
Properties Values
Specific gravity pH value Appearance / odour Total dissolved solids Cation exchange capacity Hazardous content Boiling point Evaporation Rate Solubility in water Melting point Reactivity data
1.05 3.50 Dark brown Non-obnoxious 19.7ppm 3.87% None 212˚F Same as water Complete Liquid Stable
3.8 Dosage of Enzyme
CBR and economics are considered the most important information in selecting dosage for
stabilization. The laboratory CBR values for stabilized samples may be useful in selecting a
different dosage of enzyme. Based on the research studies reported, a dosage may also be
fixed in per|m3 of soil (volume in ml is required to treat 1m3 of soil). The base line dosage for
moderately to mild plastic soil (PI=5-12) the dosage will be 1 liter of Terrazyme will treat 27
to 33 m3 of soil mix. Testing shows a strong correlation between increasing Terrazyme
dosage and final CBR results on low strength soils. The bio enzyme was purchased from M/s
Avijeet agencies, Chennai, Based on trial CBR values, Plasticity characteristics the following
dosage of Enzyme was selected for the study.
3.9 Calculation of dosage
01. 200 ml will treat 2.00 m3 of soil.
02. 200 ml will treat 2.50 m3 of soil.
03. 200 ml will treat 3.00 m3 of soil.
04. 200 ml will treat 3.50 m3 of soil.
46
Volume of CBR mould : 3886 cm3 = 3886 cc
1) Treatment for 2.00 m3
2x106 : 200 ml
1000 : [200 x 100/ 2 x 106]
1000 cc : 0.10 ml
[(0.10/1000)x 3886] = 0.388 ml
2) Treatment for 2.50m3 :
2.5x106 : 200 ml
1000 : [200 x 100/ 2.5 x 106]
1000 cc : 0.08 ml
[(0.08/1000)x 3886] = 0.310 ml
3) Treatment for 3.00 m3
3x106 : 200 ml
1000 : [200 x 100/ 3 x 106]
1000 cc : 0.066 ml
[(0.066/1000)x 3886] = 0.256 ml
4) Treatment for 3.50 m3
3.5x106 : 200 ml
1000 : [200 x 100/ 3.5 x 106]
1000 cc : 0.057 ml
[(0.057/1000)x 3886] = 0.221 ml
The dosage in percentage are shown in the Table
Table 3.7 Dosage in percentage
SNo. Dosage No Percentage
01.
02.
03.
04.
Dosage 1
Dosage 2
Dosage 3
Dosage 4
6%
7%
8%
10%
47
3.10 Construction details of the test track
3.10.1 Computation of design traffic
The design traffic is considered in terms of the cumulative number of standard axles (in the
lane carrying maximum traffic) to be carried during the design life of the road. This can be
computed using the following equation:
Where
N = cumulative number of standard axles to be catered in the design in terms of msa.
A = Initial traffic in the year of completion of construction in terms of the number of
commercial vehicles per day = 400
D = Lane distribution factor = 0.75
F = Vehicle damage factor = 2.5
n = Design life in years = 15 years
r = Annual growth rate of commercial vehicles = 0.075
N = 365
= 7.14 msa
Hence the load factor of 1-10 msa is taken in the design.
3.10.2 Construction of test track
• The foundation layer is prepared to the required grade and camber and the dust and either
loose materials are cleaned.
• The foundation layer is compacted.
• Upon the foundation layer the sub base layer is formed with silty gravel which is spread
over the formation.
48
• The soil is mixed, leveled and rotate using soil tiller and soil rotator.
• The enzyme added are tested in the laboratory, the liquid limit is 22.4, plastic limit is
17.50, plasticity index is 4.90 after 2nd week and the corresponding values are 20.2, 16.40,
3.8 respectively after 4th week and these are optimum enzyme dosage values which is 6%
as per Dosage table.
• The Terrazyme is diluted in water is sprayed over the road formation (at OMC)
• The soil is again mixed thoroughly by soil rotator to make sure that the enzymes diluted
in water are mixed through the soil and initiate the process of cation exchange.
• The surface is rolled
• After compaction a spray of water containing a light concentration of Terrazyme can be
used under extremely dry and hot conditions to enhance the curing.
• This furnishes the sub base layer.
3.10.3 Laying of Base course Layer
• The next layer is done with 70% soil (silty gravel) and 30% aggregate (the size of
aggregate is more than 25mm) thoroughly.
• It is spread on the existing layer after mixing thoroughly.
• Soil tiller and soil rotator is run over the formation.
• After mixing of the soil thoroughly the enzymes is diluted in water and mixed through the
soil to initiate the process of cat-ion exchange.
• The formed surface is rolled.
• After compaction, a spray of water containing a light concentration of Terrazyme can be
used under extremely dry and hot conditions to enhance the curing.
• The rolling is continued until the formation may set finally.
The following photos taken at test track site and is self explanatory in terms of construction
methodology using Bio Enzyme (Terra Zyme)
49
Figure 3.10 Details showing the process of ploughing
Figure 3.11 Details showing the process of mixing of soil
Figure 3.12 Details showing the Rotator in process
50
Figure 3.13 Details showing the process of Rolling
Figure 3.14 Details showing the process of Levelling
51
Figure 3.15 Details showing the process of Uniform Mixing
Figure 3.16 Details showing the process of mixing after enzyme application
52
Figure 3.17 Details showing the process of finishing the track
53
3.11 Deflection measurement by Benkelman beam
After 14 days of application of Bio enzyme ( Terrazyme)
Deflection at the first point : 0.01mm
Deflection at the Intermediate : 0.004mm
Deflection at the last point : 0.002mm
Mean deflection = ΣD/n
ΣD = 0.01+0.004+0.002 = 0.016 mm
n = 3.
D = 0.016/3 = 0.005mm
Standard deviation = Sqrt (Σ[D-D]2/[n-1]) = 0.0059 mm
Characteristics Deflection DC = D+ σ = 0.005+0.0059
= 0.0109mm.
Deflection after temperature correction : DC - 0.0065σ t
=0.0109 - (39-35)*0.0065
= 0.015 mm
Over lay thickness of Granular material =392 mm
After 28 days of application of Terrazyme
Deflection at the first point : 0.005mm
Deflection at the Intermediate : 0.0025mm
Deflection at the last point : 0.001mm
Mean deflection = 0.005+0.0025+0.001 = 0.0085
D = 0.0085/3 = 0.0028mm
Standard deviation = Sqrt (Σ[D-D]2/[n-1]) = 0.0072 mm
Characteristics Deflection DC = D+ σ = 0.0028+0.0072 = 0.01 mm.
54
Deflection after temperature correction : DC – 0.0065σ t
=0.01- 4*0.0065 = 0.016 mm
Over lay thickness of Granular material =376 mm
For normal road Contruction using WBM
Deflection at the first point : 0.092mm
Deflection at the Intermediate : 0.098mm
Deflection at the last point : 0.108mm
Mean deflection = 0.092+0.098+0.108= 0.298 mm
D = 0.298/3 = 0.099mm
Standard deviation = 0.008 mm
Characteristics Deflection
DC = D+ σ
= 0.099+0.008
= 0.107 mm.
Deflection after temperature correction :
DC – 0.0065σ t =0.107-4*0.0065 = 0.081 mm
Over lay thickness of Granular material =480 mm
3.11.1 Cost optimization
The comparison of pavement design and cost of construction based on IRC:37-2001 and new
approach of construction by substituting bio-enzyme stabilized/treated soil layer in place of
granular base layer and change in cost of construction have been worked out in general for
soil with soaked CBR of 4 percent for design traffic of 1-10 msa. The following
considerations were made for the design and cost comparison.
• The conventional design is based on IRC:37-2001.
55
• The conventional cost is based upon the average schedule of rates prevailing in the southern
states of India for the year 2010 -2011.
The designed composition of the stabilized soil with composition of crust thickness and cost
comparison is given in Table 5, which clearly indicates that granular layer is replaced by the
soil layer stabilized with Bio-enzyme and overall cost of construction is reduced by more
than 26 percent.
Table 3.8 Overlay thickness and Cost comparison
Sl.No Curing Period (weeks)
Over layer thickness in mm
1. 2 weeks 391 2. 4 weeks 376
56
Track details:
Length of the road=10 m width of road=3. 05 m
Table 3.9 Track Details
Rate in Rs. Thickness
in mm
Cost in Rs. Rate in Rs. Thickness
in mm
Cost in Rs.Materials
Used
Conventional design Bioenzyme based design
Moorum+
Enzyme
- - - 1400/m3 200 8540.00
Sand Gravel
Mix
650.00/m3 460 9119.50 - - -
WBM-
Grade 2
2000.00/m3 83 5063.00 1513/m3 83 3830.00
WBM-
Grade 2
2000.00/m3 83 5063.00 1513/m3 83 3830.00
WBM-
Grade 3
2000.00/m3 84 5124.00 1513/m3 84 3876.00
DBM 5350.00/m3 100 16317.50 5350/m3 50 8158.00
Seal Coat 230.00/m2 40 7015.00 230/m2 40 7015.00
Total 850 47,702.00 540 35,250.00
57
3.12 THEORETICAL INVESTIGATIONS
3.12.1 Mathematical Model
An important objective in mathematical model is to predict the value of a dependent random
variable based on the values of other independent variables by establishing a functional
relation of a statistical nature. A predictive model generates a new outcome (the value of the
dependent variable) based on the previous state of independent variables. Regression
analysis, as an example, is a predictive time-series model. Given a least squares linear
interpretation of data points x1, x2, x3, . . . , xn, the regression model predicts the value of
point xn+1 (and a vector of subsequent points with varying degrees of accuracy).
3.12.2 Criteria Used to Select a Model
The following criteria were used to select a model:
(i) Minimize mean sum square errors (MSE): The smallest MSE will result in the narrowest
confidence intervals and largest test statistics. The model with the smallest MSE involving
the least number of independent variables can generally be considered as the best model
(ii) Maximize the Coefficient of Determination (R2): R2 is a measure of how well the
estimated model fits the observed data. The best model selected is generally the one with the
largest R2.
(iii) Minimum increase of R2: The best model is selected as the model associated with the
smallest increase in R2 with the addition of an extra variable.
(iv) Mallows Cp statistic: The best model is usually thought to have a Cp value closest to p,
where p is the number of regression coefficients. Models associated with Cp greater than p are
usually thought to be biased or misspecified models
3.12.3 Correlation and Prediction
If the correlation between two variables X and Y is zero, we can assume that they have a
random relationship to each other. That is, a knowledge of X tells us nothing about Y and a
58
knowledge of Y tells us nothing about X (therefore, predicting X from Y or Y from X, we
can do no better than a random guess). A nonzero correlation between X and Y implies that if
we know something about X, we can know something about Y (and vice versa). A measure
moving toward +1 indicates a positive correlation, which means that as X moves up Y also
moves up. A measure moving toward −1 indicates an inverse correlation, which means that as
X moves up Y moves down. The greater the absolute value of the correlation, the more
accurate is the prediction of one variable from the other. If the correlation between X and Y is
either −1 or +1, perfect prediction is possible. However, a near-zero correlation between X
and Y does not necessarily mean that the two variables are uncorrelated. It is possible, that X
and Y share a nonlinear relationship.
3.12.4 The Regression Equation
From the database of observations, a data mining engineer can determine the degree of
correlation between two variables X and Y and then use linear regression techniques to find
the equation that best fits the trend of the relationship. A linear regression produces an
equation of the form Yi = bXi + a, where a is a constant and b is the slope of the line. Linear
regression equations are powerful tools in building models from data for which dependencies
can be precisely determined (or closely approximated). Linear regression can be used for
some nonlinear data sets by converting the parameters to a logarithmic scale. Linear
equations are often used in time-series data to predict Y (t) given X(t).
3.12.5 Application in the Present Work
In the present work, an application of different methods (simple–multiple analysis and
artificial neural networks) for the estimation of the California bearing ratio (CBR) from
various parameters such as sieve analysis, Atterberg limits, maximum dry density and
optimum moisture content of the soil is done. The resistances of granular soils, which are
used in the superstructure, foundation and subgrade layers, are usually tested by CBR
59
(California bearing ratio), which is a well accepted and still extensively used parameter.
Regression analysis and artificial neural network estimation indicated strong correlations (R2
= 0.80–0.95) between the sieve analysis, Atterberg limits, maximum dry density (MDD) and
optimum moisture content (OMC). It has been shown that the correlation equations obtained
as a result of regression analyses are in satisfactory agreement with the test results. It is
recommended that the proposed correlations will be useful for a preliminary design of a
project where there is a financial limitation and limited time.
A prediction model for CBR is an important technique in pavement analysis area. Prediction
based on statistical methods are described in this chapter. The artificial neural network
concept is used to generate the prediction model and verified with the field testing results.
Using CBR test data, proper models are developed and results are validated with performance
measure. A comparison is also made between statistical regression analysis and neural
network model and performance measures are tabulated.
3.12.6 Existing CBR Prediction Methods
The California Bearing Ratio is important test for the design of flexible pavements. This test
is performed on all types of soil from clay to gravel. In this work , the CBR of the soil is
predicted from index test values and physical property measurements of the soil. In India,
variety of soil has been commonly used for the formation of sub grade and sub base courses.
Each soil in nature, varies in physical properties, grain structure etc., which poses problems to
find its suitability for the pavement construction. The CBR test is a physical test involves
complicate procedure and the accuracy of the test relay upon standard laboratory test
conditions. If CBR test value is correlated from physical properties of the soil and this
approach will be helpful in selecting varied, more number of samples and different types of
soil samples based on correlation between CBR and physical properties of the soil which
helps us to select the best soil samples for the pavement construction . In this work, an
60
attempt has been made to use statistical methods namely, simple linear regression, multiple
linear regression, and an Artificial Neural Network to predict the CBR values from index
properties of soil. California bearing ratio (CBR) is defined as the ratio which is between the
resistance against the sinking of penetration piston into the soil with 1.27 mm/min (0.05 in.
/min) velocity and the resistance is shown by a standard crushed rock sample for the same
penetration depth. CBR is actually an indirect measure which indicates comparison of the
strength of subgrade material to the strength of standard crushed rock mentioned in
percentage values. The survey reveals the properties of the soils related with the CBR values.
An attempt is made to develop a correlation equation between CBR and other parameters like
liquid limit, plastic limit, optimum moisture content. The data set used in this study is given
below:
61
Table 3.10 Properties of Soil without adding enzyme
Prop. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
MDD (kN/m2) 14.53
17.27 15.79 15.5 14.03 14.2
16.19
16.19
15.69
15.79 15.6 14.22 14.23 14.13 14.22
OMC (%) 12 11 10 12 11 12 11 11 10 11 12
9.09 10 10 10
UCC (kN/m2)
171.68
78.48
112.80
176.58
183.90
127.13
93.20
88.29
90.20
66.20
88.20 111.59
110.36
63.765
83.385
CBR (%) 8 2.2 9 10 7.5 8.3 6.5 5 4.5 10.9 4.5 8 7.5 5 4.2
Gravel (%) 8 2.2 9 10 7.5 8.3 6.5 5 4.5 10.9 4.5 8 7.5 5 4.2
Sand (%) 62 37.4 41.6 62.9 82 86 44.4 27.8 41.8 48.4 67.4 39.6 75.71 60.8 75.5
Clay (%) 4 25 20 9.97 8 5 21 30 27 12 12.5 26 9.99 12.6 9.01
Silt (%) 16 20 20 4.37 4 6 17.4 31 24 12 10.9 20 1.5 13 1.5
LL (%) 36 38 46 49 28 30 28 35 42 60 30 35 30 26 25
PL (%) 29 32 40 43 22 25 22 30 36 54 25 27 21 17 17
PI 7 6 6 6 6 5 6 5 6 6 5 8 9 9 8
SG 1.98 2.55 2.25 2.08 2.11 2.15 2.11 1.93 2.15 2.25 2.38 1.609 1.633 1.543 1.452
CBR (Soaked
) 2.77 1.05 2.37 3.44 2.37 3.17 2.37 1.45 1.58 1.19
1.453
3.08 2.14 1.45 1.05
62
3.12.7 Methods Used
The following methods are used to develop a model
1. Simple Linear Regression (SLR)
2. Multiple Linear Regression (MLR)
3. Neural Network
4.7.1 Simple Linear Regression
The preliminary laboratory investigation are carried out and the results were studied. These
were analyzed statistically. CBR is one of the common test used to predict the strength of the
sub grade. The soil properties viz, liquid limit, plastic limit, maximum dry density, optimum
moisture content, unconfined compressive strength, the percentage of gravel, sand , clay ,silt
etc., are the parameters are treated as independent variables(Input to SLR). The CBR be the
output parameter and it is a dependent variable. Correlation coefficient may be used for
selecting the most relevant parameter to predict the CBR value. The dependent and
independent variable selected may follow the linear model such as Y=aX+b, where Y is the
output dependent variable and X is the independent variable and a,b are constants. Highest
correlation coefficient is treated as best approximation equation.
In Simple Linear regression analysis, the individual properties are treated as independent
variable and CBR value is treated as predictable variable. The following table gives the
individual parameter linear equation along with its norm residual values. Lowest residual
with clay is identified as the relevant factor to predict CBR [4, 5]. Also the highest correlation
coefficient is treated as the best approximation equation.
63
Table 3.11 SLR Coefficients with Residual & MSE
No. Parameter (a , b) Norm residual MSE
1 Max .Density
(-1.4, 29) 7.2814 4.9223
2 OMC (.61, 16) 8.2778 72.7641
3 LL (.13, 1.8) 7.216 4.7595
4 PL (.13, 2.6) 7.2522 4.7955
5 PI (.04, 6.7) 8.4015 6.4171
6 UCC (.026, 3.9) 7.6486 5.3189
7 Gravel (.15, 4.7) 7.5405 5.1769
8 Sand (.051, 4.2) 7.8463 5.5
9 Clay (.18, 9.8) 6.6443 46.3650
10 Silt (.16, 9.4) 7.1615 35.2878
The minimum norm-residual value obtained with clay content and minimum MSE is with LL.
Individual properties impact is shown in the above table with two indicators such as norm
residual and MSE.
3.12.8 Multiple Linear regression
In Multiple Linear regression analysis, all the properties are used as a collection of
independent variables and CBR value gets computed. After getting those regression
coefficients, the equation is kept as model and used for further unknown test data to
determine the CBR values. In the regression method, the independent variables affecting the
dependent variable are first determined. The normal regression provides the alternative
method for determining, in which independent variables are significant by identifying good
subset models that result in considerably less computing time than that would be required to
calculate all possible regressions.
64
Y= a0+a1X1+a2X2+…..+anXn+e,
Where Y is the dependent variable and here it is the CBR value; a0 is the Y intercept;
a1,a2,….an are the slopes associated with X1,X2,X3….Xn respectively; X1,X2,X3….Xn are
the independent variables; ‘e’ is the error. Other soil properties are assigned as independent
variables.
2 4 6 8 10 12 141
2
3
4
5
6
7
8
9
10
11
12Predicted un soaked CBR by MLR
Samples
Un
Soa
kedC
BR
ActualPredicted
Figure3.18 Predicted Unsoaked CBR by MLR
2 4 6 8 10 12 141
1.5
2
2.5
3
3.5
4
4.5
5Predicted soaked CBR by MLR
Samples
Soa
kedC
BR
ActualPredicted
Figure 3.19 Predicted soaked CBR by MLR
The above charts show the comparison of actual vs. predicted CBR for different soil samples
selected for the present research work under unsoaked and soaked conditions using MLR
analysis.
65
The results of R square value are tabulated in the following table.
Table 3.12 Soaked and Unsoaked R-square
SNo. Sample
Description
Method R-Square
1 Un soaked MLR .7032
2 Soaked MLR .7123
3.12.9 Artificial neural network
Due to difficulties in solutions of the complex engineering systems, in the past decade,
researchers have started to study on artificial neural network (ANN) inspired by the behavior
of human brain and nervous system. Each ANN model can be differently organized according
to the same basic structure. There are three main layers in ANN structure; a set of input
nodes, one or more layers of hidden nodes, and a set of output nodes. Each layer basically
contains a number of neurons working as an independent processing element and densely
interconnected with each other. The neurons using the parallel computation algorithms are
simply compiled with an adjustable connection weights, summation function and transfer
function. The methodology of ANNs is based on the learning procedure from the data set
presented it from the input layer and testing with other data set for the validation. A network
is trained by using a special learning function and learning rule. In ANNs analyses, some
function called learning functions is used for initialization; training, adaptation and
performance function. During the training process, a network is continuously updated by a
training function which repeatedly applies the input variables to a network till a desired error
criterion is obtained. Adapt functions is employed for the simulation of a network, while the
network is updated for each time step of the input vector before continuing the simulation to
the next input. Performance functions are used to grade the network results. In this study,
66
Training Function: Gradient descent with momentum and adaptive learning rate
Adapt Function: Gradient descent with momentum weight and bias learning function
Performance Function: Mean square error (MSE) was used.
In the learning stage, network initially starts by randomly assigning the adjustable weights
and threshold values for each connection between the neurons in accordance with selected
ANNs model. After the weighted inputs are summed and added the threshold values, they are
passed through a differentiable non-linear function defined as a transfer function. This
process is continued, until a particular input captures to their output (i.e., target) or as far as
the lowest possible error can be obtained by using an error criterion. An ANN model can be
differently composed in terms of architecture, learning rule and self-organization. The most
widely used ANNs are the feed-forward, multilayer perceptions trained by back-propagation
algorithms based on gradient descent method (FFBP). This algorithm can provide
approximating to any continuous function from one finite- dimensional space to another for
any desired degree of accuracy. The superiority of FFBP is that it sensitively assigns the
initial weights values and therefore it may yield closer results than the each other. Also this
algorithm has easier application and shorter training duration.
Computing systems Artificial neural networks (ANNs) are , made up of a number of simple
and highly interconnected elements. It process information by their dynamic state response to
external inputs. Enormous ANN models are available. The model accuracy depends on
network architecture (that is it depends on processing element, connection patterns, activation
functions, number of layers, a suitable learning scheme) to determine the optimal parameter
values for that architecture. Based on these approaches, there are two methods: the
destructive and the constructive methods. The destructive method begins with a large
network size, and deletes unnecessary units/ connection, until the "optimum" architecture is
67
obtained based on prediction and processing times. Constructive methods are those that start
with a relatively simple network and add units/connections to reduce the error.
General ANN models are drawn primarily on these features: massive parallelism, non-
linearity, processing by multiple layers of neural net units, and finally, dynamic feedback
among units . An important characteristic of neural networks is their ability to "learn" where
the neural networks generate their own rules, by learning from examples. Learning,
analogous "programming" in computing, is achieved through an adaptive process. Learning
can be divided into three stages: an error convergent stage, when the error decreases rapidly
in the beginning of the learning procedure; a competition stage, when the error stays nearly
constant; and a domination stage, when the error decreases rapidly to a small value.
The constructed architecture in this process is analogous to traditional learning paradigms,
such as back-propagation, for determining the value of parameters within a given network
architecture. The communication flow is unidirectional with no feedback permitted. General
ANN is based on number of principles: Processing elements in the network must be
nonlinear, easily computed, and capable of computing nontrivial functions with a minimal
number of processing elements. The neural network must allow for minimally sufficient
interactions between its processing elements. The architecture of the network (number of
nodes in each layer, number of layers, connection patterns) should be learned by example.
The steps followed in Back propagation method are given below:
Step1: Normalized field testing data are presented to the input layer.
Step2: Corresponding CBR values are assigned with output layer.
Step3: Mean square error is computed and used it for adjusting the weight factor
Step4: Using other set of data model is validated
Step5: Model now ready to accept unknown data set and predict the CBR value.
68
Figure 3.20 ANN Architecture
The main objective of this study is to create model and investigate the performance of the
traditional regression analysis with ANN for prediction of CBR. The input parameters were
trained to establish the best interrelationship between basic soil properties and the parameter
CBR. Performances of the models were examined in terms of some statistical verification
criteria. The best results were produced and listed. The comparison of the estimated CBR
values with the experimental counterparts is shown in Fig. Developed models were evaluated
for the statistical performance using some statistical parameters such as coefficient of
regression (R2), standard deviation (r), standard error (SE) and mean.
The main objective of this chapter is the estimation of California Bearing Ratio (CBR) results
using SLR, MLR and ANN models. High correlation coefficients for CBR (R2 = 0.86) were
obtained from SLR in which percentage of gravel is used as independent parameter. In the
SLR in which the percentage of fine grained is used as independent parameter for CBR, the
69
correlation coefficients were 0.80. MLR analyses were implemented using sieve analysis
results, Atterberg limits, maximum dry density (MDD) and optimum moisture content
(OMC) as independent variables to develop models for the estimation of CBR. R2 value for
CBR is 0.88. The fact that F values calculated herein are higher than the tabulated values
shows that the reliability of MLR method is high. T Tests were also performed for the results
obtained from MLR and it was observed that MLR for CBR produced a good result. It is
expected that the effects of maximum dry density, optimum moisture content and percentage
of gravel and sand are higher than the effects of other data (Atterberg limits and percentage of
fine grained) on CBR in respect of the MLR result. This can be inferred from the fact that t
values are higher than the tabulated t values. In addition, five ANN models, which have
different number of input parameters, were developed for estimation of CBR using a total of
15 test results. It was observed that the best results were obtained with input parameters. The
other models which were developed to see the effects of different soil properties (sieve
analysis results, Atterberg limits, maximum dry density and optimum moisture content) also
produced satisfactory results. It was concluded from all findings herein that use of the basic
soil properties such as grain size, Atterberg limits and compaction parameters appears to be
reasonable in the estimation of California Bearing Ratio. Nevertheless while percentage of
gravel and fine grained were the dominant parameters for the estimation of CBR with SLR,
maximum dry density, optimum moisture content and percentage of gravel and sand became
the dominants with MLR. It is shown that the constructed ANN model also used in predicting
CBR value as in traditional statistical model (SLR and MLR) for predicting CBR.
Table 3.13 Without enzyme: R-square
SNo. Sample Description Method R-Square
1 Unsoaked NN .6970
2 Soaked NN .7023
70
The R-square value may be improved by increasing the samples and modify the activation
function in this ANN model.
3.12.10 CBR in enzyme added samples [COMPARISION CHART]
The following chart shows actual versus predicted CBR for different soils added with
bio enzyme after second week and fourth week curing period under soaked and unsoaked
condition. The tables are listed in chapter 3 experimental investigations
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15150
0.2
0.4
0.6
0.8Actual Vs. Predicted CBR - using Neural Network
Samples
.221
CBR
1 2 3 4 5 6 7 8 9 10 11 12 13 14 1515-0.2
0
0.2
0.4
0.6Actual Vs. Predicted CBR - using MLR
Samples
.221
CBR
ActualPredicted
ActualPredicted
Figure 3.21 Actual Vs. Predicted CBR Using ANN & MLR[second week-dosage1]
71
Figure3.22 CBR for Actual Vs. Predicted using ANN&MLR [2nd week – dosage1]
72
1 2 3 4 5 6 7 8 9 10 11 12 13 14 150
0.2
0.4
0.6
0.8Actual Vs. Predicted CBR - using Neural Network
Samples
.256
CB
R
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15-0.2
0
0.2
0.4
0.6Actual Vs. Predicted CBR - using MLR
Samples
.256
CB
R
ActualPredicted
ActualPredicted
Figure3.23 Actual Vs. Predicted CBR Using ANN & MLR[2nd week-dosage2]
73
Figure 3.24 Actual Vs. Predicted CBR Using ANN & MLR[2nd week -dosage2]
74
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15150
0.2
0.4
0.6
0.8Actual Vs. Predicted CBR - using Neural Network
Samples
.31
CB
R
ActualPredicted
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15150
0.2
0.4
0.6
0.8Actual Vs. Predicted CBR - using MLR
Samples
.31
CB
R
ActualPredicted
Figure 3.25 Actual Vs. Predicted CBR Using ANN & MLR[2nd week -dosage3]
75
Figure3.26 Actual Vs. Predicted CBR Using ANN & MLR[2nd week -dosage3]
76
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15150
0.2
0.4
0.6
0.8Actual Vs. Predicted CBR - using Neural Network
Samples
.38
CB
R
ActualPredicted
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15150
0.2
0.4
0.6
0.8Actual Vs. Predicted CBR - using MLR
Samples
.38
CB
R
ActualPredicted
Figure 3.27 Actual Vs. Predicted CBR Using ANN & MLR[2nd week -dosage4]
77
Figure 3.28 Actual Vs. Predicted CBR Using ANN & MLR[2nd week -dosage4]
78
1 2 3 4 5 6 7 8 9 10 11 12 13 14 150
0.2
0.4
0.6
0.8Actual Vs. Predicted CBR - using Neural Network
Samples
.221
CB
R
ActualPredicted
1 2 3 4 5 6 7 8 9 10 11 12 13 14 1515150
0.2
0.4
0.6
0.8Actual Vs. Predicted CBR - using MLR
Samples
.221
CB
R
ActualPredicted
Figure 3.29 Actual Vs. Predicted CBR Using ANN & MLR[4th week-dosage1]
79
Figure 3.30 Actual Vs. Predicted CBR Using ANN & MLR[4th -dosage1]
80
1 2 3 4 5 6 7 8 9 10 11 12 13 14 150
0.2
0.4
0.6
0.8Actual Vs. Predicted CBR - using Neural Network
Samples
.256
CB
R
ActualPredicted
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15-0.2
0
0.2
0.4
0.6Actual Vs. Predicted CBR - using MLR
Samples
.256
CB
R
ActualPredicted
Figure 3.31 Actual Vs. Predicted CBR Using ANN & MLR[4th -dosage3]
81
Figure 3.32 Actual Vs. Predicted CBR Using ANN & MLR[4th -dosage2]
82
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15150
0.1
0.2
0.3
0.4Actual Vs. Predicted CBR - using Neural Network
Samples
.31
CB
R
ActualPredicted
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15150
0.2
0.4
0.6
0.8Actual Vs. Predicted CBR - using MLR
Samples
.31
CB
R
ActualPredicted
Figure 3.33 Actual Vs. Predicted CBR Using ANN & MLR[4th -dosage3]
83
Figure 3.34 Actual Vs. Predicted CBR Using ANN & MLR[4th week-dosage3]
84
1 2 3 4 5 6 7 8 9 10 11 12 13 14 150
0.2
0.4
0.6
0.8Actual Vs. Predicted CBR - using Neural Network
Samples
.38
CB
R
ActualPredicted
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15150
0.2
0.4
0.6
0.8Actual Vs. Predicted CBR - using MLR
Samples
.38
CB
R
ActualPredicted
Figure 3.35 Actual Vs. Predicted CBR Using ANN & MLR[4th week-dosage4]
85
Figure 3.36 Actual Vs. Predicted CBR Using ANN & MLR[4th week-dosage4]
86
3.12.11 Consolidated results
Table 3.14 R-Square value for ANN & MLR
Fourth Week Second Week
Enzyme
Dosage
in ml NN MLR NN MLR
0.221 0.6902 0.7709 0.7317 0.8003
0.256 0.7962 0.7473 0.7143 0.7807
0.310 0.6627 0.8884 0.7251 0.9714
0.388 0.6073 0.8469 0.7897 0.8309
In this section, Figures are drawn for Fourth week and second week data set. This can again
be computed for Neural network and MLR. Each diagram depicts that comparison between
actual vs. predicted CBR with each enzyme dosage values. Finally, the Table consolidates the
two categories of R-square measure for NN and MLR. Largest R-square obtained at 0.31
dosage of second week data. By using this model, immediate conclusion may be taken by the
engineers to design pavement with optimal enzyme dosage and suitable soil for best result.
Prediction is validated with this model.