4.4. Validation - Multiple Debris Impacts on a Raised Structure - Digital Twin (OSU LWF)

Problem files

Github

4.4.1. Overview

In this digital twin validation example, debris-field wave-flume tests at a NHERI facility, Oregon State University’s Large Wave Flume (OSU LWF), are briefly summarized before demonstrating the use of HydroUQ’s OSU LWF digital twin paired with the Material Point Method (MPM).

Fig. 4.4.1.1 HydroUQ’s desktop GUI for the NHERI OSU LWF digital wave-flume twin.

Details for the experiments are available in various publications. Namely, the work of Andrew Winter [Winter2020] [Winter2019], Krishnendu Shekhar [Shekhar2020] and Dakota Mascarenas [Mascarenas2022] [Mascarenas2022PORTS]. The simulations replicated in this example appeared originally in Bonus 2023 [Bonus2023Dissertation].

Experiments were performed in the NHERI OSU LWF, a 100 meter long flume with adjustable bathymetry, in order to quantify stochastic impact loads of ordered and disordered debris-fields on effectively rigid, raised structure.

Fig. 4.4.1.2 NHERI OSU LWF facility’s experimental schematic is used in this example. Adapted from Winter 2019 [Winter2019], and Mascarenas 2022 [Mascarenas2022].

This example may help to produce a robust database (numerical and physical) from which to eventually be able to extract both the first principles of wave-driven debris-field phenomena and design guidelines on induced forces.

We validate against two very similar (but not identical) physical studies done in the OSU LWF by [Shekhar2020] and [Mascarenas2022], indicating high accuracy of our model and low bias to minor experiment specifications.

Results for free surface elevation and streamwise structural loads are to be recorded for validation at a specified interval.

Qualitatively, an MPM simulation of debris impacts on a raised structure in the OSU LWF is shown below.

Fig. 4.4.1.3 OSU LWF debris impact photos from HydroUQ’s MPM simulations.

It appears similar in the mechanism of debris impact, stalling, and deflection relative to the structure and flow for a similar case in Mascarenas 2022 [Mascarenas2022].

Fig. 4.4.1.4 OSU LWF debris impact photos from Mascarenas 2022 [Mascarenas2022] experiments.

The experiments by Shekhar et al. 2020 [Shekhar2020] are also shown below for comparison. These tests had a slightly different configuration, primarily the debris were located 0.5 meters further upstream from the box and the water level was 0.10-0.15 meters lower than the 2.0 meter datum used in the simulations and Mascarenas 2022 [Mascarenas2022] experiments.

Fig. 4.4.1.5 OSU LWF debris impact photos from Shekhar et al. 2020 [Shekhar2020] experiments.

Similar figures can be made for the whole range of order debris-array experiments done at the OSU LWF. However, this example focuses on teaching you how to replicate the above results.

4.4.2. Set-Up

A step-by-step walkthrough on replicating an MPM simulation result from Bonus 2023 [Bonus2023Dissertation] is provided below.

Open Settings. Here we set the simulation time, the time step, and the number of processors to use, among other pre-simulation decisions.

Fig. 4.4.2.1 HydroUQ Settings GUI

Open Bodies / Fluid / Material. Here we set the material properties of the fluid and the debris.

Fig. 4.4.2.2 HydroUQ Bodies Fluid Material GUI

Open Bodies / Fluid / Geometry. Here we set the geometry of the flume, the debris, and the raised structure.

Fig. 4.4.2.3 HydroUQ Bodies Fluid Geometry GUI

Open Algorithm. Here we set the algorithm parameters for the simulation. We choose to apply F-Bar antilocking to aid in the pressure field’s accuracy on the fluid. The associated toggle must be checked, and the antilocking ratio set to 0.9, loosely.

Fig. 4.4.2.4 HydroUQ Bodies Fluid Algorithm GUI

Open Bodies / Fluid / Partitions. Here we set the number of partitions for the simulation. This is the domain decomposition across discrete hardware units, i.e. Multi-GPUs. These may be kept as their default values.

Fig. 4.4.2.5 HydroUQ Bodies Fluid Partitions GUI

Moving onto the creation of an ordered debris array, we set the debris properties in the Bodies / Debris / Material tab. We will assume debris are made of HDPE plastic, as in experiments by Mascarenas 2022 [Mascarenas2022] and Shekhar et al. 2020 [Shekhar2020].

Fig. 4.4.2.6 HydroUQ Bodies Debris Material GUI

Open Bodies / Debris / Geometry. Here we set the debris properties, such as the number of debris, the size of the debris, and the spacing between the debris. Rotation is another option, though not used in this example. We’ve elected to use an 8 x 4 grid of debris (longitudinal axis parallel to long-axis of the flume).

Fig. 4.4.2.7 HydroUQ Bodies Debris Geometry GUI

The Bodies / Debris / Algorithm and Debris / Partitions tabs are not used in this example but are available for more advanced users.

Open Bodies / Structures. Uncheck the box that enables this body, if it is checked. We will not model the structure as a body in this example, instead, we will modify it as a boundary later.

Fig. 4.4.2.8 HydroUQ Bodies Structures GUI

Open Boundaries / Wave Flume. We will set the boundary to be a rigid body, with a fixed separable velocity condition, that is faithful to the digital twin of the NHERI OSU LWF. Bathmyetry joint points should be identical to the ones used in Bodeis / FLuid.

Fig. 4.4.2.9 HydroUQ Boundaries Wave Flume Facility GUI

Open Boundaries / Wave Generator. Fill in the appropriate file-path for the wave generator paddle motion. It is designed to produce near-solitary-like waves.

Fig. 4.4.2.10 HydroUQ Boundaries Wave Generator GUI

Open Boundaries / Rigid Structure. This is where we will specify the raised structure as a boundary condition. By doing so, we can determine the exact loads on the rigid boundary grid-nodes, which may then be mapped to the FEM tab for nonlinear UQ structural response analysis.

Fig. 4.4.2.11 HydroUQ Boundaries Rigid Structure GUI

Open Boundaries / RigidWalls.

Fig. 4.4.2.12 HydroUQ Boundaries Wave-Flume Facility GUI

Open Sensors / Wave Gauges. Set the Use these sensor? box to True so that the simulation will output results for the instruments we set on this page.

Three wave gauges will be defined. The first is located prior to the bathymetry ramps, the second partially up the ramps, and the third near the bathymetry crest, debris, and raised structure.

Set the origins and dimensions of each wave as in the table below. To match experimental conditions, we also apply a 120 Hz sampling rate to the wave gauges, meaning they record data every 0.0083 seconds.

Fig. 4.4.2.13 HydroUQ Sensors Wave-Gauge GUI

These wave gauges will read all numerical bodies (i.e. particles) within their defined regions at every sampling step and will report the highest elevation value (Position Y) of a contained body as the free-surface elevation at that gauge. The results are written into our sensor results files.

Open Sensors / Load Cells. Set the Use these sensor? box to True so that the simulation will output results for the instruments we set on this page.

Fig. 4.4.2.14 HydroUQ Sensors Load-Cells GUI

Open Outputs. Here we set the non-physical output parameters for the simulation, e.g. attributes to save per frame and file extension types. The particle bodies’ output frequency is set to 10 Hz (0.1 seconds), meaning the simulation will output results every 0.1 seconds. This is decent for animations without taking up too much space. Fill in the rest of the data in the figure into your GUI to ensure all your outputs match this example.

Fig. 4.4.2.15 HydroUQ Outputs GUI

4.4.3. Simulation

We assume that 2 hours are reserved for your simulation. For those using the reduce fluid bulk modulus or reduced resolution, this may be more than necessary.

This simulation was run on the TACC Lonestar6 system. It uses three NVIDIA A100 GPUs on a single node in the gpu-a100 queue. The real-time to complete was 2 hours. The simulated time in the digital twin is 26 seconds.

To retrieve results from the analysis, the analysis must complete and post-process the model output files into an appropriate format before the end of the allotted submission time.

Important

Provide a large amount of time for the Max Run Time field in HydroUQ when submitting a job to ensure the model completes before the time allotted runs out! We recommend 2 hours in this example.

Warning

Only ask for what you need in terms of sensor size, count, and output sampling rate. Otherwise, you will end up with massive amounts of data which can slow simulations due to I/O constraints.

4.4.4. Analysis

When the simulation job has been completed, the results will be available on the remote system for retrieval or remote post-processing.

Retrieving the results.zip folder from the Tools & Applications Page of Design Safe starts by navigating to the designsafe-ci.org website. Login and go to Use DesignSafe / Tools & Applications

Fig. 4.4.4.1 Locating the job files on DesignSafe

Check if the job has finished in the right-side vertical drawer by clicking the refresh icon. If it has, click More info.

Fig. 4.4.4.2 Job status is finished on DesignSafe

Once the job is finished, the output files should be available in the directory which the analysis results were sent to

Find the files by clicking View.

Fig. 4.4.4.3 Viewing the job files on DesignSafe

Move the results.zip to somewhere in My Data/. Use the Extractor tool available on DesignSafe. Unzip the results.zip folder.

Fig. 4.4.4.4 Extracting the results.zip folder on DesignSafe

OR Download the results.zip folder to your PC and unzip to look at the model results.

Fig. 4.4.4.5 Download button on DesignSafe shown in red

Download the results to look at the geometry files of the analysis.

Extract the results.zip folder either on DesignSafe or on your local machine. You will likely want to have a free Side FX Houdini Apprentice installation to view BGEO files.

Fig. 4.4.4.6 File-system view of results zip folder on DesignSafe.

Locate the zip folder and extract it somewhere convenient. The local or remote work directory on your computer is a good option, but note that these files may be erased if another simulation is set up in HydroUQ, so keep a backup somewhere outside the working directories.

HydroUQ’s sensor/probe/instrument output is available in {your_path_to_HydroUQ_WorkDir}/HydroUQ/RemoteWorkDir/results/ as CSV files.

Particle geometry files often have a BGEO extension, open Side FX Houdini Apprentice (free to use) to look at MPM results in high-detail.

Once complete, the simulation data at the three wave gauges (WG1, WG2, and WG3, left-to-right) is as shown below when plotted against experimental trials of Mascarenas 2022 [Mascarenas2022] for the “unbroken” solitary wave case.

Fig. 4.4.4.7 OSU LWF simulated free-surface elevation wave gauges vs. experimental data from Mascarenas 2022 [Mascarenas2022].

The simulation data at the load-cell is as shown below when plotted against experimental trials of Mascarenas 2022 [Mascarenas2022] for the “unbroken” solitary wave case. The experimental streamwise load is the combination of “LC5” and “LC8” in Mascarenas 2022 [Mascarenas2022], as both measured streamwise load on the box to reduce errors from position / slight box apparatus out-of-plane rotation.

Fig. 4.4.4.8 OSU LWF simulated streamwise load-cells vs. experimental data from Mascarenas 2022 [Mascarenas2022].

Though only one case was considered here, if many experimental debris-field cases are run (10+) we can use HydroUQ to perform a sensitivity analysis on the debris-field parameters. This isn’t pursued here-in.

However, the following box-and-whisker charts demonstrate the strength of the numerical replication, as most points fall within experimental interquartile ranges and never outside of the experimental envelope for impact loads.

Fig. 4.4.4.9 OSU LWF simulated first peak debris impact loads vs. experimental data from Mascarenas 2022 [Mascarenas2022].

This complete our HydroUQ validation example for multiple debris impacts on a raised structure in the OSU LWF, Bonus 2023 [Bonus2023Dissertation].

4.4.5. References

Winter2019(1,2)

Winter, A. (2019). “Effects of Flow Shielding and Channeling on Tsunami-Induced Loading of Coastal Structures.” PhD thesis. University of Washington, Seattle.

Winter2020

Andrew O Winter, Mohammad S Alam, Krishnendu Shekhar, Michael R Motley, Marc O Eberhard, Andre R Barbosa, Pedro Lomonaco, Pedro Arduino, Daniel T Cox (2019). “Tsunami-Like Wave Forces on an Elevated Coastal Structure: Effects of Flow Shielding and Channeling.” Journal of Waterway, Port, Coastal, and Ocean Engineering.

Shekhar2020(1,2,3,4,5)

Shekhar, K., Mascarenas, D., and Cox, D. (2020). “Wave-Driven Debris Impact on a Raised Structure in the Large Wave Flume.” 17th International Conference on Hydroinformatics, Seoul, South Korea.

Mascarenas2022(1,2,3,4,5,6,7,8,9,10,11,12,13)

Mascarenas, Dakota. (2022). “Quantification of Wave-Driven Debris Impact on a Raised Structure in a Large Wave Flume.” Masters thesis. University of Washington, Seattle.

Mascarenas2022PORTS

Mascarenas, Dakota, Motley, M., Eberhard, M. (2022). “Wave-Driven Debris Impact on a Raised Structure in the Large Wave Flume.” Journal of Waterway, Port, Coastal, and Ocean Engineering.

Bonus2023Dissertation(1,2,3)

Bonus, Justin (2023). “Evaluation of Fluid-Driven Debris Impacts in a High-Performance Multi-GPU Material Point Method.” PhD thesis. University of Washington, Seattle.