Intelligent back analysis using data from the instrument (poster)
1. Masoud Ghaemi, Hamed zarei, Alireza Khalili , Kaveh Ahangari
hamed.zarei69@gmail.com
Paper ID: 2SMFE10103060207
Intelligent back analysis using data from the instrument
2nd
Iranian Conference on Soil Mechanics and Foundation Engineering
Qom University of Technology, Qom, Iran – October 2015
2nd
Iranian Conference on Soil Mechanics and Foundation Engineering
Qom University of Technology, Qom, Iran – October 2015
ABSTRACT
In this paper a model based on Perceptron multilayer artificial neural network have been presented
for intelligent regressive analysis of Chehel Chai water conveyance tunnel base on monitoring data.
Our input data were 27 parameters categorized in three classes including: tunneling data, geological
data, and average of in situ horizontal stress. For network instruction, data bank of regressive
analysis results of 18 Convergence stations was prepared in 980 classes by using FLAC3D
software.
Then according to network behavior in instructing step, optimum values for medial layers number,
neurons number and, activity functions obtained. By this way a model was mad based on artificial
neural network that was able to regressive analyzing of displacements in future Convergence station
projects in every time of monitoring..
1. INTRODUCTION
Neural Network abilities in learning from widely dispersed and erroneous data caused this
technique to be successful in solving problems related to geotechnical engineering. Although,
teaching a neural network is a time-consuming process, when running at high speed, it can be
time-saving, though. Therefore, it can be taken into account as an appropriate alternative for
time-consuming and complex numerical back analysis of monitoring results.
2. INTRODUCING NARMAB TUNNEL
Chehel Chai Water Conveyance Tunnel is 3175 meters long and 5 meters deep (excavation
diameter). Furthermore, its entry altitude is 2324 meters, in west coast of Chehel Chai River, in
right coast of Narmab River (22675 meters).
Figure 1. Shematic representation of the geological profile of the Chehel Chai Water Conveyance
Tunnel
Parameters 0-128 m 128-190 m 190-219 m 219-580 m
C(MPa(
E(GPa(
K(GPa(
G(GPa(
0.08
45
1.4
0.93
0.56
0.25
0.16
37
2.5
1.6
1
0.25
0.19
36
2.5
1.6
1
0.25
0.25
32
2.5
1.6
1
0.25
Geomechanical properties of tunnel are as follows:
3. BACK ANALYSIS OF MONITORING STATIONS IN NARMAB WATER CONVEYANCE TUNNEL
Back analysis is able to forecast controlling parameters of system through analyzing its output behavior. Back analysis
problems may be solved in two different ways: inverse and direct. In inverse method, mathematical formulation is just
opposite the typical analysis; however, the direct method is based on optimization in which trial values of unknowns are
corrected in a way that the difference between values measured and calculated is minimized. This method can be used for non-
linear relations as well. However, the mentioned method needs a lot of time to carry out repetitive calculations. One variable
optimization searching algorithm technique has been applied. Error function is defined by
Where N: number of points measured; uk: calculated displacement and uk*: measured displacement
Among available optimization methods, univariate and univariate periodic search techniques can find optimal
values of parameters regardless of their initial values.
Numerical back analysis presumptions of Chehel Chai Water Conveyance Tunnel monitoring stations are as follows:
A)Variables: rock mass module of elasticity (Erm) and average horizontal in situ stress (Shav)
B)Back analysis by direct approach with periodic univariate algorithm
C)Rock mass surrounding tunnel is placed in rock classes with close joints. Back analysis was performed using FLAC3D
software.
D)All stations were modelled using Mohr-Coulomb elasto-plastic behavior model.
E)The dimensions of each monitoring station under study equals 10 meters of tunnel length in 5 meters before and after station site.
Figure 2. FLAC3D
model for the Chehel Chai Water Conveyance Tunnel.
3.1 CHOOSING INPUT (Data)
In order to perform intelligent back analysis on Chehel Chai Water Conveyance Tunnel using monitoring results, data were collected from 18 stations of
convergence meter classified as 980 data categories each of which included 29 parameters (a total of 26460 parameters). 27 of 29 parameters, input, are
classified into three general categories of rock mass Geomechanical parameters, tunneling and convergence meter parameters. Moreover, 2 parameters,
output, include rock mass elasticity module and average horizontal in situ stress. The final architecture of the network used in this study are shown in
Fig. 3. Then, they were examined and analyzed for finding non-linear, complex relation between inputs and outputs using multi-layer perception neural
networks with back-propagation learning rule. Glimpsing at collected data, it can be concluded that it may have a wide range of possibilities. Thus, it can
be stated that the network trained using mentioned data is highly generalizable and can be used with high reliability to perform back analysis on future
stations of convergence meter in non-excavated sections of Chehel Chai Tunnel (more than 45% of tunnel route).
Figure 3. The final architecture of the trained ANN
4. VALIDATION OF THE ANN-BASED SOLUTION
The values of rock mass elasticity module and average horizontal in situ stress for 980 different conditions were
calculated using the numerical code FLAC. As mentioned before, the data were divided into two sets of training and test.
The training data were used in training the ANN and obtaining the values of network weights and biases, while the test
data were reserved for examining the accuracy of the ANN in new conditions that has never experienced.
Figure 4. The plots of predicted vs. Target values of rock mass elasticity module and average horizontal in situ stress and
corresponding values of coefficient of determination for training data.
Figure 5. The plots of predicted vs. Target values of rock mass elasticity module and average horizontal in situ stress and
corresponding values of coefficient of determination for test data.
5. CONCLUSION
Equations Results show that the mentioned network has learnt extremely non-linear relations between input and output parameters and regarding instrumentation
sections, it is able to estimate geomechanical parameters of rock mass surrounding the tunnel at different time periods of monitoring with an acceptable approximation.
Furthermore, results show that network designed for performing intelligent back analysis of Chehel Chai Water Conveyance Tunnel NET9 was trained well. As shown
in the figures, in all cases the values were higher than 0.99 that is quite satisfactory.
Table 1. Geomechanical properties of
tunnel