Application of Partitioned Iterative Coupling Approach to Multi-physics

: The engineering needs which have to consider the multi-physics are increasing. When the multi-phase problem is treated by monolithic coupling approach, there are some difficulties such as the increase in the number of degree of freedom, the complication of the program code and so on. In order to overcome these shortcomings in considering the new additional phase, we applied the partitioned iterative coupling approach to the Hydro-Mechanical coupled problem in this research. In the partitioned iterative coupling approach, the behavior of single phases can be individually calculated and the interaction between the phases has been considered at an appropriate timing by the middleware. The middleware plays a role to regulate the single-phase analysis. In this research, we carried out the numerical simulation of one-dimensional consolidation by both the partitioned iterative coupling scheme and the monolithic coupling scheme, and then we confirmed that the numerical results are the same. Through the research, it figured out the efficient iterative scheme.


Introduction
When we analyze the behavior of saturated soil, it is necessary to consider the interaction between the deformation of the soil particle and the change in pore water pressure.Therefore, we have to solve simultaneously the governing equation such as the equilibrium equation and continuity equation.
In recent years, the structures and the phenomena to be analyzed has been diversified, and in some cases, pore air, heat, and chemistry also have to be coupled.The field dealing with such problems is called Multi-physics.
When we solve a problem with a large number of coupled phases, the degree of freedom of the simultaneous equations to be solved increases as the number of coupled phases increases.In addition, the program becomes complicated, it becomes difficult to maintain and improve, which may cause a bug.The method solved simultaneously the governing equations describing the coupled problem is called the monolithic method.On the other hand, the method which the governing equations concerning each phase is solved separately, and the convergence calculation is conducted is called the partitioned method (Jahromi, 2007 [1]).If the deformation of soil, pore water pressure, pore air pressure, heat, chemical reaction etc. can be solved separately by the partitioned method, we can connect existing software and use it effectively.As a result, we can overcome the above-mentioned shortcomings of a large number of coupled phase problem.
The objective of this research is to solve the soil-water coupled problem by the partitioned method.In the future, the partitioned method will be applied in the field of Multi-Physics other than soil-water.The applicability of the partitioned method to the one-dimensional consolidation problem is investigated in this research.Then, we compared the solution by the partitioned method and the monolithic method.

Soil-water coupled problem and partitioned method
The governing equations to describe the soil-water coupled problem are as follows.In the following description, compression is defined to be positive.
These equations are the equilibrium equation, the principle of effective stress, the constitutive equation, the strain-displacement relation, Darcy's law and the continuity equation respectively.ρ is density, u is displacement vector, σ is stress tensor, b is body force vector, σ' is effective stress tensor, pw is pore water pressure, 1 is the Kronecker delta, C is constitutive tensor, ε is strain tensor, w is Darcy's velocity, k is hydraulic conductivity tensor, h is total water head, εv is volumetric strain.
The variables are spatially discretized as shown below. where [Nh] and [Bh] are the shape function matrices, the superscript ne and me indicate the variables at the discretized points.By substituting these equations to the weak form of equilibrium equation and continuity equation, the spatially discretized equations of an element are obtained: where matrices are expressed as follows: The time derivative of displacement, the total water head and the time derivative of the total water head are temporally discretized as where θ is constant within a range from 0 to 1.By substituting these equations to spatially discretized equations and summing up all elements, the spatially and temporally discretized weak forms of the equilibrium equation and the continuity equation (the global stiffness equation) are obtained: In order to separate the phase, the terms other than the diagonal terms are shifted to the right-hand side of the equation, we obtain The upper equation describes the deformation problem and the shifted terms are given as the external force which is corresponding with the calculated excess pore water pressure from the lower equation.On the contrary, the lower equation describes the seepage flow problem and the shifted terms are given as the quantity of water flow which is corresponding to the calculated increment of displacement from the upper equation.These equations are solved separately until the solutions are converged.Fig. 1 shows a flowchart of the above mentioned partitioned method for soil-water coupled problem introduced the SOR method (Young, 1971 [2], Masatake, 2002 [3]).In Fig. 1, the matrixes and vectors are replaced to simplify as Fig. 1: Flowchart of the partitioned method for soil-water coupled problem ωu and ωh are the acceleration parameter to be slow the convergence speed for the increment of displacement Δu and the hydraulic head h respectively, and they are applied in order to prevent from the divergence of calculation.

Verification of partitioned method in one-dimensional consolidation
In this section, the applicability of the partitioned method to soil-water coupled problem is verified by comparing the simulated results of one-dimensional consolidation obtained by the partitioned method and the monolithic method.In the simulation, one element is used.The upper side of the element is drained condition and the other sides are set as undrained conditions.The linear elastic constitutive model is applied in the simulation.The Lame's constant λ and μ are set to be 1338.8kN/m 2 and 666.9 kN/m 2 respectively.The coefficient of permeability k is set to be 1.0☓10 -4 m/s.The load of 98 kPa is applied to the upper side of the element at the first time step which the increment of time Δt = 1.0☓10 -3 (day).After the second time step, the displacement at the upper side of the element and the excess pore water pressure at the lower side of the element are calculated by giving the increment of time Δt = 0.0367(day) which is corresponded to the time factor Tv = 0.001 and repeating the step 1000 times.
The convergence calculation is carried out at the first step in the consolidation problem which the elapsed time is 0.001 (day) by using the partitioned method.In this research, when the error rate which is defined as |x (k+1) -x (k) |/| x (k) | ☓100 becomes less than or equals to 1.0☓10 -5 , the solutions are judged to be converged.
By changing the acceleration parameter ωu from 1.0☓10 -6 to 1.0☓10 -3 by each 0.1 and the acceleration parameter ωh from 0.1 to 0.5 by each 0.05, the convergence calculation was performed and the iteration number was obtained.Fig. 2 shows a contour diagram of the iteration number.The interval between contour lines is every 20 iterations, after 200 iterations, every 100 iterations.When ωu = 1.0☓10 -3 , the calculation diverged between 0.4 and 0.5 of ωh.When ωu = 1.0☓10 -3 , the calculation was not converged within 500 iterations.It is understandable results that the calculation is diverged by using a large acceleration parameter and conversely, the number of iteration increases by using a small acceleration parameter.In this case, the optimum acceleration parameter ωuopt, ωhopt becomes 1.0☓10 -4 , 0.45 respectively.Fig. 3 shows the contour diagram of the iteration number in case of Δt = 0.0356 (day) for the second step.By changing the acceleration parameter ωu from 1.0☓10 -4 to 1.0☓10 -1 by each 0.1 and the acceleration parameter ωh from 0.1 to 0.6 by each 0.05, the convergence calculation was performed.The trend in which the divergence occurs when a large acceleration parameter is used and the convergence becomes slower when the smaller acceleration parameter is used is similar to the first step.However, the appropriate acceleration parameter is different from the first step.In this case, the optimum acceleration parameter ωuopt, ωhopt becomes 1.0☓10 -3 , 0.55 respectively.Figs. 4 to 6 show the contour diagram of the iteration number when the order of norm of [Khh] matrix is set to be 10 -9 , 10 -10 , 10 -11 respectively.The contour diagram is shown in Fig. 5 is same as the result of the convergence calculation at the first step.When the order of norm of [Khh] matrix becomes 10 times larger or 10 times smaller, the contour lines move along the axis of the acceleration parameter for the displacement ωu.The appropriate acceleration parameter ωu is 1.0☓10 -5 , 1.0☓10 -4 , 1.0☓10 -3 in case of ||Khh|| = O(10 -9 ), O(10 -10 ), O(10 -11 ) respectively.From these cases, the appropriate acceleration parameter is obviously related with the order of [Khh] matrix.These results suggest that the appropriate acceleration parameter might be possible to be estimated by [Khh] matrix.
A comparison between the results of one-dimensional consolidation obtained by the partitioned method and the results obtained by the monolithic method is shown in Fig. 7.In order to calculate the displacement and excess pore water pressure by the monolithic method, the finite element program which can solve the soil-water coupled problem named DACSAR-I (Iizuka et al., 1987 [4], Takeyama et al. 2015 [5]) is employed.In the partitioned method, the appropriate acceleration parameters ωu and ωh set to be 1.0☓10 -4 , 0.45 for the first step and 1.0☓10 -3 , 0.55 for the second step and after based on the contour diagram shown in Fig. 2 and Fig. 3. From Fig. 7, it is confirmed that the results by the partitioned method and the results by the monolithic method agree with each other.It is found that the disadvantage of the partitioned method that it takes a lot of time because there is processing of reading and writing the file in the iteration process.

Conclusions
In order to solve the soil-water coupled problem separately, the partitioned method was applied.The coupled problem is divided into the deformation problem and the seepage flow problem and are solved separately until the solutions are converged.In order to verify the applicability of the partitioned method to soil-water coupled problem, the simulated results of one-dimensional consolidation obtained by the partitioned method and the results by the monolithic method are compared.The acceleration parameter is adopted in the iterative calculation.The appropriate acceleration parameter is obviously related with the order of the diagonal matrix in the global stiffness matrix.Therefore it is suggested that the appropriate acceleration parameters might be possible to be estimated by the diagonal matrix.It is confirmed that the results by the partitioned method and the results by the monolithic method agree with each other.However, it is found that the disadvantage of the partitioned method that it takes a lot of time because there is processing of reading and writing the file in the iterative calculation.

Partitioned method Monolithic method
Partitioned method Monolithic method