Awesome

Granular Flow Simulation Using taichi_mpm

Simulating sand mass collision using taichi mpm

Input

Using input.json file, granular mass can either be generated randomly in a specified domain, or can be placed manually.

{
    "save_path": "./sand3d_collision/",
      "output_format": {  # Optional
        "timestep_downsampling_rate": 2,  # Downsample timestep, else, default is set to 1.0
        "material_feature": [  # Choose what material feature to include in the input npz
            "friction_angle",
            "elastic_modulus"
        ]
    },
    "id_range": [  # the id of simulations to generate
        0,
        10
    ],
    "domain_size": 1.0,  # the largest domain length (same for all dimension)
    "friction_angle": 35,
    "wall_friction": 0.43,
    "elastic_modulus": 2000000.0,
    "poisson_ratio": 0.3,
    "rho": 1800,
    "sim_space": [  # lower and upper boundary for each dimension 
        [
            0.2,
            0.8
        ],
        [
            0.2,
            0.8
        ],
        [
            0.2,
            0.8
        ]
    ],
    "sim_resolution": [
        32,
        32,
        32
    ],
    "nsteps": 350,  # number of forward steps
    "mpm_dt": 0.0025,  # time between forward steps
    "gravity": [0, -9.81, 0],
    "gen_cube_randomly": {
        "generate": true,
        "sim_inputs": {
            "mass": {
                "ncubes": [1, 2],
                "min_distance_between_cubes": 0.01,
                "cube_size_range": [[0.15, 0.3], [0.15, 0.3], [0.15, 0.3]],
                "cube_gen_space": [[0.11, 0.5], [0.11, 0.50], [0.11, 0.89]],
                "vel_range": [[3, 3], [-2.5, 2.5], [-2.5, 2.5]],
                "nparticle_limits": 15000
            },
            "obstacles": {
                "ncubes": [1, 2],
                "min_distance_between_cubes": 0.01,
                "cube_size_range": [[0.1, 0.1], [0.3, 0.3], [0.1, 0.1]],
                "cube_gen_space": [[1.0, 1.5], [0.10, 0.4], [0.15, 0.85]],
                "nparticle_limits": 7000
            }
        }
    },
    "gen_cube_from_data": {
        "generate": true,
        "sim_inputs": [
            {
                "id": 0,  # id of simulation
                "mass": {
                    "cubes": [
                        [
                            0.2,  # x corner 
                            0.2,  # y corner
                            0.2,  # z corner
                            0.2,  # x length
                            0.3,  # y length
                            0.4   # z length 
                        ],
                        "parcitles.csv"  # custom particle data
                    ],
                    "velocity_for_cubes": [
                        [
                            1.0,  # x vel
                            1.0,  # y vel
                            1.5   # z vel
                        ],
                        [
                             1.0, 
                             0.0,
                             -0.5   
                        ]
                    ]
                },
                "obstacles": null  # may repeat the as what is written in "mass" to add obstacles
            },
            {
              "id": 1  
                ...  # may repeat that is written in id 0 to add more simulations
            }
        ]
    },
    "visualization": {
        "is_realtime_vis": false,
        "is_save_animation": true,
        "skip": 1
    }
}

In the random generating case, if cube_size_range is defined for all dimensions (e.g., [[0.15, 0.3], [0.15, 0.3], [0.15, 0.3]]), the shape of cubes will be randomly generated following the specified values. If cube_size_range is only defined for one dimension (e.g., [0.1, 0.4]), the shape of cubes will be square whose length is random from [0.1, 0.4]

Generate particles from user defined particle files

Simply append the file name to the gen_cube_from_data["sim_inputs"]["mass"]["cubes"] list. The file format is as follows.

x, y
0.143, 0.123
0.243, 0.523
0.343, 0.423
...

Output

The output is saved and .npz file. The code also saves simple .gif animation for the simulation.

Run

python3 run_mpm.py  --input_path="examples/sand2d/inputs_2d.json"

Simulation Example

Note

In taichi, coordinate y-axis corresponds to the height and y-axis corresponds to the plain. However, in matplotlib used in rendering, these axes are the opposite. To deal with this problem, we transpose the axis of the taichi simulation results (positions) when saving .npz files.