Iceberg dynamics

(4 comments)

Prompted by this tweet and campaign for icebergs to be depicted in their most stable equilibrium orientation, here is a Python script modelling the dynamics of a two-dimensional iceberg which starts in an arbitrary orientation and position and relaxes under gravitational and buoyant forces to its most stable configuration. A cork floats "on its side": with its longest axis parallel to the water's surface (it doesn't bob around with its longest axis vertical), and an iceberg does the same.

Annotated iceberg picture

The centre of mass ($G$) and the centre of buoyancy ($B$, the centre of mass of the displaced water) are indicated by the red and blue dots respectively. The motion of the iceberg (treated as a rigid body) is dictated by the force of gravity ($\boldsymbol{F}_g$) and an opposing buoyant force ($\boldsymbol{F}_b$) due to the displaced water. In addition to any translational force moving the centre of mass of the iceberg, where these forces are not colinear they exert a torque, $\boldsymbol{\tau} = \boldsymbol{r} \times \boldsymbol{F}_b$, where $r$ is the vector from $G$ to $B$, which tends to rotate the iceberg.

Iceberg simulation

There are few things one can customize in the code: the shape of the iceberg (defined as a sequence of vertex coordinates), the initial position ($h$) and orientation (call rotate_poly to rotate the iceberg to a given angle), and the friction applied to damp the rotation and translation of the iceberg in water (edit the apply_friction function). I'll tidy up and post the code on GitHub when I get a moment.

import numpy as np
import matplotlib.pyplot as plt
from matplotlib import animation

# Plot limits: ideally the figure will not have to rescale during the animation
xmin, xmax = -5, 20
ymin, ymax = -15, 10
width, height = xmax - xmin, ymax - ymin

# The area density of sea water and ice.
rho_water, rho_ice = 1.027, 0.9
# Acceleration due to gravity, m.s-2
g = 9.81

def canonicalize_poly(xy):
    """Shift the (N+1,2) array of coordinates xy to start at minimum y."""
    #if get_area(xy) > 0:
    #    xy = xy[::-1]
    idx_ymin = xy[:,1].argmin()
    xy = np.roll(xy[:-1], -idx_ymin, axis=0)
    return np.vstack((xy, xy[0]))

def rotate_poly(xy, theta):
    """Rotate the (N+1,2) array of coordinates xy by angle theta about (0,0).

    The rotation angle, theta, is in radians.

    """
    s, c = np.sin(theta), np.cos(theta)
    R = np.array(((c, -s), (s, c)))
    xyp = (R @ xy.T).T
    return canonicalize_poly(xyp)

def get_area(xy):
    """Return the area of the polygon xy.

    xy is a (N+1,0) NumPy array defining the N polygon vertices, but repeating
    the first vertex as its last element. The "shoelace algorithm" is used.

    """

    x, y = xy.T
    return np.sum(x[:-1]*y[1:] - x[1:]*y[:-1]) / 2

def get_cofm(xy, A=None):
    """Return the centre of mass of the polygon xy.

    xy is a (N+1,0) NumPy array defining the N polygon vertices, but repeating
    the first vertex as its last element. If the polygon area is not passed
    in as A it is calculated. The polygon must have uniform density.

    """

    if A is None:
        A = get_area(xy)
    x, y = xy.T
    Cx = np.sum((x[:-1] + x[1:]) * (x[:-1]*y[1:] - x[1:]*y[:-1])) / 6 / A
    Cy = np.sum((y[:-1] + y[1:]) * (x[:-1]*y[1:] - x[1:]*y[:-1])) / 6 / A
    return np.array((Cx, Cy))

def get_moi(xy, rho):
    """Return the moment of inertia of the polygon xy with density rho.

    xy is a (N+1,0) NumPy array defining the N polygon vertices, but repeating
    the first vertex as its last element.

    """

    x, y = xy.T
    Ix = rho * np.abs(np.sum((x[:-1]*y[1:] - x[1:]*y[:-1]) *
                (y[:-1]**2 + y[:-1]*y[1:] + y[1:]**2)) / 12)
    Iy = rho * np.abs(np.sum((x[:-1]*y[1:] - x[1:]*y[:-1]) *
                (x[:-1]**2 + x[:-1]*x[1:] + x[1:]**2)) / 12)
    # Perpendicular axis theorem.
    Iz = Ix + Iy
    return Ix, Iy, Iz

def get_zero_crossing(pts):
    """Return the coordinates of the zero-crossing in pts.

    pts is a pair of (x, y) points, assumed to be joined by a straight line
    segment. This function returns the coordinates (x,0) at which this line
    crosses the y-axis (corresponding to sea-level in our model).

    """
    P0, P1 = pts
    x0, y0 = P0
    x1, y1 = P1
    if (x1-x0) == 0:
        return x1, 0
    m = (y1-y0)/(x1-x0)
    c = y1 - m*x1
    return -c/m, 0

def get_displaced_water_poly(iceberg, submerged=None):
    """Get the polygon for the submerged portion of the iceberg.

    iceberg is a (N+1,2) array of coordinates corresponding to the iceberg's
    vertexes in its current position and orientation (the first vertex is
    repeated at the end of the array);
    submerged is a boolean array corresponding to the vertexes which are
    under water (<0); if not provided it is calculated.

    """

    if submerged is None:
        submerged = (iceberg[:,1] <= 0)
    nsubmerged = sum(submerged)

    # Partially-submerged iceberg: find where it enters the sea, i.e. which
    # edges cross zero. zc_idx holds the indexes of the vertices *before*
    # each zero-crossing edge.
    diff = np.diff(submerged)
    zc_idx = np.where(diff)[0]
    # Interpolate to find the coordinates of the zero crossing.
    ncrossings = len(zc_idx)
    # We're going to build a polygon for the shape of the displaced water,
    # i.e. the submerged part of the iceberg.
    displaced_water = np.empty((nsubmerged + ncrossings, 2))
    # Loop over the points *before* each crossing in pairs. NB if the
    # iceberg is partially submerged, len(zc_idx) is guaranteed to be even.
    assert not ncrossings % 2
    i = j  = 0
    for idx1, idx2 in zip(zc_idx[0::2], zc_idx[1::2]):
        # All the submerged vertices up to the upwards crossing.
        displaced_water[j:j+idx1-i+1] = iceberg[i:idx1+1]
        # Work out where the crossing vertex should be and add it.
        c = get_zero_crossing(iceberg[idx1:idx1+2])
        j += idx1 - i + 1
        displaced_water[j] = c
        j += 1

        # Now the downward crossing: all the unsubmerged vertices are
        # skipped, and an extra vertex at sea level is added.
        c = get_zero_crossing(iceberg[idx2:idx2+2])
        displaced_water[j] = c
        j += 1
        i = idx2 + 1
    # Copy across any remaining submerged vertexes to displaced_water.
    displaced_water[j:] = iceberg[i:]
    return displaced_water

def apply_friction(omega, dh):
    """Apply frictional forces to the angular and linear velocities."""

    # Hard friction: angular and linear velocities are immediately quenched.
    # after movement.
    # return 0, np.array((0,0))

    # Intermediate friction: reduce the velocities by some fraction.
    return omega * 0.9, dh * 0.6


# Our two-dimensional iceberg, defined as a polygon.
poly = [
(3,0), (3,3), (1,5), (4,7), (0,12), (1,15), (4,17), (6,14), (7,14), (8,12),
(7,10), (7,7), (5,1)
]

# Repeat the initial vertex at the end, for convenience.
iceberg0 = np.array(poly + [poly[0]])
# Centre the iceberg's local coordinates on its centre of mass.
iceberg0 = iceberg0 - get_cofm(iceberg0)
# We might want to start the iceberg off in some different orientation:
# if so, rotate it here.
iceberg0 = rotate_poly(iceberg0, -0.2)

# Get the (signed) area, mass, and weight of the iceberg.
A = get_area(iceberg0)
M = rho_ice * abs(A)
Fg = np.array((0, -M * g))

# We also need the Iz component of the iceberg's moment of inertia.
_, _, Iz = get_moi(iceberg0, rho_ice)

fig, ax = plt.subplots()
ax.set_xlim(xmin, xmax)
ax.set_ylim(ymin, ymax)
# We would prefer equal distances in the x- and y-directions to look the same.
ax.axis('equal')

# The centre of mass starts at this height above sea level.
h = 15
# theta is the turning angle of the iceberg from its initial orientation;
# G is the position of its centre of mass (in world coordinates).
theta, G = 0, np.array((6, h))
# omega = dtheta / dt is the angular velocity; dh is the linear velocity.
omega, dh = 0, np.array((0,0))
# The time step (s): small, but not too small.
dt = 0.1

def update(it):
    """Update the animation for iteration number it."""

    global omega, dh, G, theta

    print('iteration #{}'.format(it))

    # Update iceberg orientation and position.
    theta += omega * dt
    G = G + dh * dt

    # Rotate and translate a copy of the original iceberg into its current.
    # position.
    iceberg = rotate_poly(iceberg0, theta)
    iceberg = iceberg + G

    # Which vertices are submerged (have their y-coordinate negative)?
    submerged = (iceberg[:,1] <= 0)
    nsubmerged = sum(submerged)

    iceberg_in_water = True
    if nsubmerged in (0, 1):
        # The iceberg is in the air above the surface of the sea.
        B = None
        Adisplaced = 0
        alpha = 0
        iceberg_in_water = False

    if iceberg_in_water:
        # Apply some frictional forces which damp the motion in water.
        omega, dh = apply_friction(omega, dh)
        if nsubmerged == len(submerged):
            # The iceberg is fully submerged.
            displaced_water = iceberg
            Adisplaced = A
            B = G
        else:
            displaced_water = get_displaced_water_poly(iceberg, submerged)

            # Area of the displaced water and position of the centre of buoyancy.
            Adisplaced = get_area(displaced_water)
            B = get_cofm(displaced_water)

    # Buoyant force due to the displaced water.
    Fb = np.array((0, rho_water * abs(Adisplaced) * g))

    if B is not None:
        # Vector from G to B
        r = B - G
        # Torque about G
        tau = np.cross(r, Fb)
        alpha = tau / Iz

    # Resultant force on the iceberg.
    F = Fg + Fb
    # Net linear acceleration.
    a = F / M

    # Now plot the scene for this frame of the animation.
    ax.clear()

    # The sea! The sea!
    sea_patch = plt.Rectangle((xmin, ymin), width, -ymin, fc='#8888ff')
    ax.add_patch(sea_patch)

    # The iceberg itself, in its current orientation and position.
    poly_patch = plt.Polygon(iceberg, fc='#ddddff', ec='k')
    ax.add_patch(poly_patch)

    if B is not None:
        # Draw the submerged part of the iceberg in a different colour.
        poly_patch = plt.Polygon(displaced_water, fc='#ffdddd', ec='k')
        ax.add_patch(poly_patch)

        # Indicate the position of the centre of buoyancy.
        bofm_patch = plt.Circle(B, 0.2, fc='b')
        ax.add_patch(bofm_patch)

    # Indicate the position of the centre of mass.
    cofm_patch = plt.Circle(G, 0.2, fc='r')
    ax.add_patch(cofm_patch)

    # Update the angular and linear velocities
    omega += alpha * dt
    dh = dh + a * dt

    ax.set_xlim(xmin, xmax)
    ax.set_ylim(ymin, ymax)
    ax.axis('equal')

ani = animation.FuncAnimation(fig, update, 150, blit=False, interval=100, repeat=True)
plt.show()
Current rating: 5

Comments

Comments are pre-moderated. Please be patient and your comment will appear soon.

John Sharma 3 years, 8 months ago

This is an excellent and informative piece of analysis. Thank you for sharing this, Christian.

Kind Regards,

John

Link | Reply
Current rating: 5

christian 3 years, 8 months ago

I'm glad you found it interesting – thank you for your comment.

Link | Reply
Current rating: 5

John Sharma 3 years, 8 months ago

Welcome, Christian. Observing real-life phenomena, like icebergs, being modelled through efficiently written code has been quite fascinating.

John

Link | Reply
Current rating: 5

Jonas 2 years, 3 months ago

Nice approach,
you may also be interested in the web-applet by Joshua Tauberer https://joshdata.me/iceberger.html

Link | Reply
Current rating: 5

New Comment

required

required (not published)

optional

required