Tensorflow Coupon Models

Branched CNNs were developed in tensorflow to estimate the parameterization of a TePla damage model from a coupon experiment. To read more about the coupon experiment, see our 2023 paper.

Model Inputs & Outputs

Networks trained on coupon data take as input an image (either a density field or a proton radiograph) and a scalar simulation parameter (a detonation velocity \(D_{cj}\)). An example proton radiograph is shown below.

coupon experiment proton radiograph

These models predict values of a parameterized TePla damage model:

  • \(Y_{spall}\) the spall strength,

  • \(log_{10}(\phi_0)\) the log base 10 of the initial porosity,

  • \(\eta\) the overstress parameter,

and the simulation time of an input image \(T_{sim}\).

Running the Coupon Example

To run the scripts on the tensorflow coupon networks, use the following arguments:

  • --PACKAGE tensorflow

  • --EXPERIMENT coupon

  • --MODEL ../examples/tf_coupon/trained_rho2TePla_model.h5 or --MODEL ../examples/tf_coupon/trained_pRad2TePla_model.h5

  • --INPUT_FIELD rho or --IN_FIELD pRad

  • --INPUT_NPZ use any .npz file in ../examples/tf_coupon/data/

  • --INPUT_DIR ../examples/tf_coupon/data/

  • --DESING_FILE ../examples/tf_coupon/coupon_design_file.csv

  • --NORM_FILE ../examples/tf_coupon/coupon_normalization.npz

  • --FIXED_KEY tpl112, --FIXED_KEY idx00110, or --FIXED_KEY None

Coupon Data

The file names of coupon data contain multiple pieces of information about their contents. The two relevent componets are the tpl and the idx:

  • The tpl is followed by a number identifying a simulation ID.

    • The values of \(Y_{spall}, log_{10}(\phi_0), \eta\) are identical across files with idential tpl ‘s.

    • Data is included for tpl112 at all 26 timesteps.

  • The idx specifies what time step the simulation was at.

    • The value of \(T_{sim}\) will be identical across files with identical idx ‘s.

    • Data is included at idx00110 for 31 differnt simulations.

Each .npz coupon data file contains the following fields:

  • sim_time (scalar): ground truth for \(T_{sim}\) prediction

  • pRad (2D array): proton radiograph training field (use -IN_FIELD pRad)

  • rho (2D array): density training field (use -IN_FIELD rho)

  • hr_coupon (2D array): density of the coupon

  • hr_maincharge (2D array): density of the main charge

  • pressure (2D array)

  • temperature (2D array)

  • melt_state (2D array): binary array of if a cell has melted

  • porosity (2D array)

  • eqps (2D array): equivalent plastic stress

  • eqps_rate (2D array): equivalent plastic stress rate

  • eff_stress (2D array): effective stress

  • bulk_mod (2D array): bulk modulus of the material

  • sound_speed (2D array): speed of sound in the material

  • rVel (2D array): veliocty of material in the R-axis direction

  • zVel (2D array): velocity of material in the Z-axis direction

  • Rcoord (1D vector): vector containing position in cm of all cells along the R-axis

  • Zcoord (1D vector): vector containing position in cm of all cells along the Z-axis

Model Architecture

These models have two branches to handle the two disparate input types: an image processing branch, and a simulation parameter processing branch.

tensorflow coupon branched neural network diagram

Model Layers

The layers in model follow the following naming convention:

  • ???_input: input layer

  • batch_normalization(_#): batch normalization layer

  • activation(_#): activation layer

  • interp_##: 2D convolutional layer that (mostly) preserves size, decreases first two dimensions by 2

  • second_reduce_##: 2D convolutional layers that more drastically decrease size

  • dense_?: dense layers

  • post_reduce_conv: 2D convolutional layer that (mostly) preserves size, decreases first two dimensions by 2

  • flatten_conv: 2D convolution layer that reduces the first two dimensions to 1

  • concateate: concatenate layer

  • final_dense: dense layer that produces outputs

  • SIM_PRMS_????: Indicates that the layer is on the network branch that operates on the non-image simulation parameter input; layers without this flag are assumed to be on the main network “trunk” that operates on the image input

  • SIM_PRMS_IMG_???: Indicates the layer where the simulation parameters branch joins the main network “trunk”