Sei sulla pagina 1di 8

ECCM 2010

IV European Conference on Computational Mechanics Palais des Congrs, Paris, France, May 16-21, 2010

Optimization of Ground Improvement Techniques in Tunnelling Using Genetic Algorithms


M. A. Eid 1 , M.Zaki Abd Elrehim 2 , F. El-kashef 3 and G. Swoboda4
1 2 3 4

Civil Engineering Department, Minia University, Minia, Egypt, m_eid76@yahoo.com Civil Engineering Department, Minia University, Minia, Egypt, mostafa_zaki88@yahoo.com Civil Engineering Department, Minia University, Minia, Egypt. m_eid76@windowslive.com Department of Structural Analysis, University of Innsbruck, Innsbruck, Austria, gunter.swoboda@uibk.ac.at

Abstract: With recent developments in efficient deterministic numerical algorithms, coupled with the rapid improvement of computer performance, probabilistic finite element methods that use more realistic soil and physical models are now feasible . The genetic algorithms (GA s ) are one of such probabilistic search methods which are well suited to solve large combinational design problems. The primary reasons for their success are their broad applicability, ease of use, and global perspective. Being a stochastic method, the GA does not need specific information to guide the search, and requires only an evaluation of the objective function value for each decision variable set in order to proceed. Tunnelling through difficult ground needs, in addition to face subdivision, effective solutions based on soil strengthening and reinforcement techniques. These different techniques and solutions aim to improve the soil around the tunnel to minimize the ground deformation, to have longer tunnelling free span and to maintain face stability. The layout and design of these techniques mainly depend on experience. So, there is an increasing need for tools to help in more understanding and more optimization of ground improvement techniques. In the present paper, it is shown how a genetic algorithm can be incorporated into the displacement finite element method to solve the optimization problem. Deterministic Finite Element program, FINAL package, is used to simulate tunnelling operations under different loads with different initial profiles and parameters values for ground improvement process. A Genetic algorithm program is coded and linked to FINAL results to perform the evolution procedures (crossover, selection, mutation etc.). Repeating these stages for many generations, results in the optimum values and parameters used for the ground improvement technique in both of its quantity and material properties. 1. Introduction: Tunnelling work requires a proper attention of ground deformation. Depending on the method of excavation and support, tunnelling may induce considerably different magnitudes of deformation to the surrounding ground. These movements may cause damage to adjacent buildings and other surface structures especially in soft soils. So, prediction and assig nment of control technique of surface and subsurface settlement are very important parts of designers task to insure safe construction [1-3]. In some cases, using jet grout, as a soil strengthening technique, is selected as an economic procedure to control surface settlement [4]. Determination of grouting parameters, dimensions and its

material properties, is used to be previous experience dependent. Applying optimization process on the determination of these parameters can help to get the best performance with minimum costs. The search for a robust optimization algorithm, with good balance between efficiency and effectiveness and able to survive in many different environments, has led to the use of Genetic Algorithms (GA s ). GAs have many advantages over traditional optimization techniques [5]. Among other considerations, they do not need further additional information than objective function or fitness information. Its ability to find out the optimum or quasi-optimum solution, even if the optimization function is noncontinuous, non-differentiable and with any kind of constraints, gives it a privileged position as a powerful tool in non-conventional optimization problems. GA s have been successfully applied to many branches in civil engineering optimization p roblems such as bridge construction management [6], water network design [7], tunnel geometric design [8] and structural optimization [9]. In the present work, the grouting parameters properties optimization problem is treated as a discrete problem using a GA. The GA is coded and combined with developed Finite Element program to analyze NATM tunnel in grouted strengthened soil. The object of this new program is to get the best values for grouting parameters. Best values are those values produce minimum cost within safety limits for surface settlement. This article has been structured as follows: First a general description of GAs is presented. Second, Tunnel structural modeling is demonstrated showing geometrical and geotechnical parameters. Third, the optimization methodology has been presented with outlines about the developed program. Finally, results are analyzed and discussed. 2. Genetic Algorithms: Genetic algorithms are search and optimization procedures that are motivated by the principles of natural genetics and natural selection. They are also referred to as stochastic optimization techniques different from usual mathematical programming [10]. Being considered as stochastic methods, the GAs do not need specific information to guide the search, and r equire only an evaluation of the objective function value for each decision variable set in order to proceed. They typically work with a coding of the decision variables, not with the decision variables themselves. They search simultaneously using a popula tion of decision variable sets, not a single set of decision variables.

Figure 1: A generic framework for genetic algorithm Figure 1 illustrates the generic framework for genetic algorithm. As in a biological system submitted to external constraints, the fittest members of the population are selected to survive and given better chances of reproducing and transmitting part of their genetic heritage to the next generation. A new population is then created by recombination of parental genes. It is expected that some members of this new population will have acquired the best characteristics of both parents and, being better adapted to the environmental conditions, will provide an improved solution to the considered problem. After it has replaced the original population, the new group is submitted to the
2

same evaluation procedure, and later generates its own offsprings. The process is repeated many times, until elite members of a given generation share the same genetic heritage. These members, who are often quite different from their ancestors, possess genetic information that corresponds to the best solution to the optimization problem [11]. 3. Structural Modelling of Tunnelling Problem In this paper, conventional numerical model with plane-strain analysis is used. An optimal tunnel profile is selected to induce minimum stresses on the lining [12]. Figure 2 shows the finite element mesh used for simulation, the applied boundary condition and soil parameters.

Figure 2: Finite element mesh Where; E

elasticity modulus, Poissons ratio, angle of internal friction

soil density, cohesion,

3.1 Geometrical and Geotechnical Parameters The tunnel is constructed in a clayey layer at a depth of 14.0m with no existence for ground water in the tunneling vicinity. The excavated cross section is divided into two parts, heading and invert. Grout is assumed to be distributed in uniform arc shape above the tunnel. Thickness and length of this arc differ according to the grouting amount. 3.2 Finite Element Simulation For the modelling process in the Finite Element program, FINAL package [13], the soil media is modeled using a six node linearly varying strain triangular finite elements (L.S.T) and the shotcrete lining by a six node curved boundary beam elements (Beam6). A half-section mesh is used in the analysis to reduce computation time. Sufficient mesh depth and width, to model soil infinite body, are used. For boundary condition, vertical and horizontal movements are prevented at the bottom of the model while only the horizontal movements are prevented at both sides.
3

3.3 Construction Sequences For simulating the excavation sequences and shotcrete installation, 2D plane strain model with different loading cases is employed. The initial loading case is the geostatic stat of stresses and the existing surface loads. Then modulus of elasticity and soil parameters of clay are increased in specified area to indicate the application of grouting strengthening technique in this area before excavation. Each excavation step is simulated in two steps using stiffness reduction method. In the first step, the stiffness of the soil in the excavated part is reduced by a factor accompanied with stress redistribution at the tunnel zone. In the second step, the excavated part is removed and the temporary lining is installed in its fresh state accompanied with a stress elimination of the excavated soil. Then, the permanent shotcrete/concrete lining is installed after completing the whole cross section. 4. Grout Optimization Problem The grout optimization problem addressed herein is to find out the optimal dimension for grouted area and elastic modulus for grouting material those lead to pre-specified and safe surface settlement to have minimum soil strengthening cost. Hence, the objective function can be stated as: find X R k to minimize f (X) subject to gi (X) 0, i= 1, 2, . . . , n and

X L X XU , j j

j = 1,2,..........., k

where X is the vector of design variables; f (X) is the objective function; gi (X) is the performance constraints; and X L and X U refer to the lower and upper bounds on the design variables respectively. j j The objective function here is the cost of the grout and can be expressed as:

min C = M c L j ( i )T j (i ) ,

j (i ) = 1,2,......n

(1)

where M c is grouting cost per cubic meter, L j ( i ) is the length of grouting arc and T j (i ) is grouting arc thickness taken from n trials. The performance constraints include concern for surface settlement as:

4.1 Design Variables First design variable is grouting arc length. Grouting arc length is limited to two-third tunnel perimeter as a practical value. This length is divided into 19 segments of 0.5 m length each. Thus we have 20 possible values for this length by adding zero length value from the case of no-grout. So, the length ranges from 0 to 9.5 m with 0.5 m step. Second design variable is thickness of grouting arc. Clay layer extends for about 7.5 m above tunnel crown. So, this distance is divided into 15 layers of 0.5 m thickness each. This means that we have 15 possible values for thickness ranges from 0.5 m up to 7.5 m. By adding the case of no-grout, thickness equal zero, we got 16 possible values. These variables are illustrated in Figure 3.

| |

g i = i all 0 ,

i = 1,2,........n

(2)

The third variable is the elastic modulus for grouting material. As clay layer has an elastic modulus of 50 MPa, the grouting modulus range starts from double of this value and increase with 50 MPa increment. A binary integer of 5-bit length is used to present all variables. So, 32 values range from 100 to 1650 MPa are used as search area for elastic modulus in this analysis. Elastic modulus indicates grouting material quality which is related to grouting unit cost. This cost differs from one country to another depending on raw materials prices and technical process costs. In this research, costs are estimated according to prices in Egypt.

Figure 3: Dimensions and limits of grouting arc 4.2 Computation Procedure According to the assumed alternatives for every variable, the total number of possible solutions (chromosomes) = 20 * 16 *32 = 10240. An initial population of 20 chromosomes is considered. Initial population and every new generation pass through three different phases. First, for every randomly generated solution, the structural analysis package FINAL is called to analyze it and calculate maximum surface settlement. Solutions with allowed safe settlement value are kept for intermediate phase and other solutions are to be mutated and re-checked until reaching safe value. All solutions in the new phase are safe and to be evaluated. Roulette selection is applied on these solutions to move to the third and final phase. Each solution has a probability to be selected depends on its fitness (evaluation) value. The final phase represents the generation form involved with other GA operators. Best values are copied to secure their coexisting in next generation according to elitism operator. Onepoint crossover and mutation operators are applied on the binary coded solution in order to form a new preliminary population.

Processing optimization operators and repeating them through generated population leads to convergence toward global optimum. Difficulty of having optimal or quasi optimal solution increases as convergence rate increases. In final optimization step, looking for the very close best chromosome through random mutation is a time consuming process. In this case, changing mutation to local searching technique is more preferable. The algorithm changes its mutation technique at the final 10 generations of every 100 generations to avoid having local minima. Then, termination of the GA process depends on the resulted design variables values stability for more additional 100 generations after final convergence.

Figure 4: Elite members convergence progression For example, the algorithm was applied to optimize grouting variables with 3cm allowable surface settlement as constraint value. Figure 4 shows elite members fitness progression through generations. The resulted optimal values for design variables are 1m, 9.5m and 1450MPa for grouting arc thickness, length and modulus of elasticity respectively. By generation number 190, elite members converged and reached design variables of 1m, 9.5m and 1600 MPa for grouting arc thickness, length and modulus of elasticity respectively. Two steps far from the optimal values. Starting from the second 100 generation, the program switches mutation to local search (uniform mutation) technique at the last 10 generations of every 100 generations as stated before. This leaded elite members to reach the optimal value at generation number 195. Another 100 generations are produced to check solutions stability. 5. Constraints Value Effect The algorithm was applied for different constraint values: 4, 3.5, 3, 2.5 and 2cm allowable surface settlement, 50 attempts for each. The aim of these attempts is to check the algorithm reliability in reaching global optimum and to study the effect of constraints value on convergence speed. Figure 5 summarize results of all performed attempts. It shows algorithms success percentage in reaching optimal design variables and global optimum costs value every hundred generations. Dividing generation range into hundreds demonstrates the speed of the algorithm.

For allowable surface settlement of 4, 3.5, 3, 2.5 and 2 cm, the percentages of reaching global optimum in the first hundred generations were 58%, 20%, 10%, 0 and 2% respectively. The percentages of reaching global optimum in the first two hundreds generations were 96%, 90%, 80%, 72% and 34% respectively. The percentages of reaching global optimum in the first three hundreds generations were 98%, 92%, 100%, 72% and 60% respectively. Increasing limitation, by reducing the allowable value, bind the algorithms searching freedom. This slows down the algorithm but does not stop it from reaching the target. For further limitation, the algorithm will get slower and it may be more visible to use deterministic searching algorithms.

Figure 5: Prospects percentage of reaching global optimum due to different constraints values

6. Conclusions The developed algorithm combining GA and FE has proven to be successful in grouting, as an example of ground improvement techniques, optimization problem. The developed computer program can be easily used to handle optimal solution. Due to its simplicity and ease of coding the GA procedure described here can be applied to a wide variety of optimization problems. It has been shown that it can be used in jet grouting costs minimization. Despite t e h relatively large space of permissible solutions, the procedure converged towards the best possible solution. The work presented here can be considered as a step forward towards analysis and design automation using soft computing. References
[1] Soil Movements Around Tunnel in Soft Soil, C. Lee, B. Wu and S. Chiou, Proceeding of National Science Council, Vol.23, No.2, 1999. [2] Analysis of Settlement Damage to Masonary Structures, H.J. Burd, G. T. Houlsby, L. Chow, C. E. Augarde and G. Liu, Numerical Methods in Geotechnical Engineering, Smith(ed), Balkema, Rotterdam, 1994.

[3] Ground and Building Settlement due to Tunnelling in Hong Kong, R. A. Forth and C. B. B. Thorley, Proceedings of the Fifth International Symposium on Land Subsidence, The Hague, October 1995. [4] Chemical Grouting and Soil Stabilization, R. H. Karol, Marcel Dekker, 2003. [5] Genetic Algorithms in Search, Optimization, and Machine Learning, D. E. Goldberg, Addison Wesley, 1989. [6] Application of Genetic Algorithm to Bridge Construction Management, Natsuaki, Y., H. Furuta, S. Mukandai and K. Yasuda, 3rd International Symposium on Uncertainty Modeling and Analysis, 1995. [7] Genetic algorithms for least-cost design of water distribution networks, D. A. Savic and W. A. Walters, J. of Water Resources Planning and Management, ASCE, 123B, p67-71, 1997. [8] A Genetic Algorithm for Tunnel Design Optimization, M. B. Reed, S. Schenk and G. Swoboda, Genetic and Evolutionary Computation Conference (GECCO), 2005. [9] A Combined Finite Element-Genetic Algorithm for Structural Optimization, M. Z. AbdElrehim, F. A. Elkashef, M. A. Eid, Aion Shams Journal of Civil Engineering, Vol. 1, No.1, 2009. [10] Adaptation in Natural and Artificial Systems, J. Holland, The University of Michigan Press: Michigan, 1975; MIT Press : Cambridge, MA, 1992. [11] Discrete Optimization of Structures Using a Floating Point Genetic Algorithm, N. Turkkan, Annual Conference of the Canadian Society for Civil Engineering, Moncton, Nouveau - Brunswick, Canada, GCM -134/1-8, 2003. [12] Design Optimization of Urban NATM Tunnelling, M. Z. AbdElrehim, A. M. Marwan, 13th International Conference on Structural and Geotechnical Enginnering (IGSGE), Cairo, 2009. [13] Programmsystem FINAL, Finite Element Analysis Linearer und Nichtlinearer Stukturen, Version 7.1, G. Swoboda, Druck Universitt Innsbruck, Austria, 2001.

Potrebbero piacerti anche