Track solidification of a metallic alloy

In this example, we identify and track the solid-liquid (S-L) interface in a nickel-based alloy undergoing solidification. Tracking the solidification over time enables the calculation of the solidification velocity. This is important to characterize the solidified structure of the sample and will be used to inform research into additive manufacturing of metals. The image sequence was obtained by the Center for Advanced Non-Ferrous Structural Alloys (CANFSA) using synchrotron x-radiography at the Advanced Photon Source (APS) at Argonne National Laboratory (ANL). This analysis was first presented at a conference [1].

[1]

Corvellec M. and Becker C. G. (2021, May 17-18) “Quantifying solidification of metallic alloys with scikit-image” [Conference presentation]. BIDS ImageXD 2021 (Image Analysis Across Domains). Virtual participation. https://www.youtube.com/watch?v=cB1HTgmWTd8

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import plotly.express as px
import plotly.io

from skimage import filters, measure, restoration
from skimage.data import nickel_solidification

image_sequence = nickel_solidification()

y0, y1, x0, x1 = 0, 180, 100, 330

image_sequence = image_sequence[:, y0:y1, x0:x1]

print(f'shape: {image_sequence.shape}')

shape: (11, 180, 230)

The dataset is a 2D image stack with 11 frames (time points). We visualize and analyze it in a workflow where the first image processing steps are performed on the entire three-dimensional dataset (i.e., across space and time), such that the removal of localized, transient noise is favored as opposed to that of physical features (e.g., bubbles, splatters, etc.), which exist in roughly the same position from one frame to the next.

fig = px.imshow(
    image_sequence,
    animation_frame=0,
    binary_string=True,
    labels={'animation_frame': 'time point'},
)
plotly.io.show(fig)

Compute image deltas

Let us first apply a Gaussian low-pass filter in order to smooth the images and reduce noise. Next, we compute the image deltas, i.e., the sequence of differences between two consecutive frames. To do this, we subtract the image sequence ending at the second-to-last frame from the image sequence starting at the second frame.

smoothed = filters.gaussian(image_sequence)
image_deltas = smoothed[1:, :, :] - smoothed[:-1, :, :]

fig = px.imshow(
    image_deltas,
    animation_frame=0,
    binary_string=True,
    labels={'animation_frame': 'time point'},
)
plotly.io.show(fig)

Clip lowest and highest intensities

We now calculate the 5th and 95th percentile intensities of image_deltas: We want to clip the intensities which lie below the 5th percentile intensity and above the 95th percentile intensity, while also rescaling the intensity values to [0, 1].

p_low, p_high = np.percentile(image_deltas, [5, 95])
clipped = image_deltas - p_low
clipped[clipped < 0.0] = 0.0
clipped = clipped / p_high
clipped[clipped > 1.0] = 1.0

fig = px.imshow(
    clipped,
    animation_frame=0,
    binary_string=True,
    labels={'animation_frame': 'time point'},
)
plotly.io.show(fig)