You are on page 1of 41

Gibson Stanford (Orcid ID: 0000-0002-2125-3424)

Comparing Single-Phase, Non-Newtonian Approaches to Experimental


Results:
Validating Flume-Scale Mud and Debris Flow in HEC-RAS
S. Gibson1 (corresponding author), I. Floyd2, A. Sánchez1, R. Heath2
1
Hydrologic Engineering Center, Davis, CA, USA. Stanford.Gibson@usace.army.mil
2
Coastal and Hydraulics Laboratory, Vicksburg, MS, USA, Ian.Floyd@usace.army.mil

Running Head/Short Title:


Comparing Numerical Non-Newtonian Approaches to Experimental Results

Abstract

Post-fire debris flows and tailing impoundment failures destroy lives and property. These geologic

hazards—and other similar processes—fall on a continuum between classic Newtonian flood analyses

and geotechnical stability analyses. The US Army Corps of Engineers (USACE) is developing a

Non-Newtonian library (DebrisLib) that includes a suite of rheological and clastic approaches to

hyper-concentrated, mudflow, and debris flow dynamics. The Hydrologic Engineering Center (HEC)

has implemented these non-Newtonian methods into the widely used, public domain open-channel

hydraulics and morphodynamic software, HEC-RAS (River Analysis System). This work presents

part of the verification and validation of these non-Newtonian approaches, applying several

rheological equations to Parsons et al.’s (2001) high concentration flume experiments.

This article has been accepted for publication and undergone full peer review but has not been
through the copyediting, typesetting, pagination and proofreading process which may lead to
differences between this version and the Version of Record. Please cite this article as doi:
10.1002/esp.5044

This article is protected by copyright. All rights reserved.


This study tested the linear Bingham model as well as the turbulent and Bagnold quadratic terms of

the O’Brien equation. HEC-RAS also includes the non-linear Herschel-Bulkley (HB) approach,

which quantifies shear-thickening or shear-thinning processes. The study used these non-Newtonian

models in HEC-RAS to simulate ten of Parsons et al. (2001) flume experiments, which measured the

snout and plug velocity of fluids with high solid concentrations (Cv = 68–74%) and a broad range of

material gradations (d50 = 0.05–1mm, d15 = 0.006–0.1 mm). The experiments also measured and back-

calculated Bingham and HB parameters of the materials, finding HB powers between 0.45 and 1.25

(i.e. fluids that are dilatant, pseudo-plastic, and visco-plastic).

The rheological models incorporated into DebrisLib and implemented in HEC-RAS reproduce

experimental data well for most experiments. The Bingham model generated a plug velocity Root

Mean Squared Error (RMSE) of 0.21 m/s using standard flow parameters and Parsons et al.’s

calibrated parameters, a substantial improvement over the unmodified shallow water flow equations

(RMSE 0.77 m/s). Experiments with strong snout effects tended to generate higher residuals,

especially in the snout velocity. The RMSE associated with the O’Brien equation was larger with the

Parsons et al. fit parameters, but similar (0.23 m/s) with measured parameters. The turbulent

parameter was the largest (often the dominant) parameter in most O’Brien simulations, with the

dispersive stress only proving significant for the coarsest material. DebrisLib had to use a modified

version of the dispersive term to simulate these concentrations. Both the 2D depth-averaged Shallow

Water Equation (SWE) and Diffusion Wave Equation were used to simulate the experiments. The

best results were obtained with the SWE with horizontal mixing.

Introduction

In January 2018, intense rainfall centred on a mountainous, coastal, watershed upstream of

Santa Barbara, California. Under normal conditions, the runoff from this event would have

caused moderate flooding and minor damage. But the Thomas fire had just burned 1,140 km2

of the watershed, exposing hydrophobic sediment on high gradient slopes (Kean et al., 2018).

This article is protected by copyright. All rights reserved.


The mud and debris flows killed 23 people and caused over $200 million of damage.

Wildfire frequency and intensity are rising in the Western United States and around the

world, making mud and debris flows more common. In 2020 alone, more than 100 wildfires

have burned over six million acres across twelve state in the US. These wildfires can

transform otherwise moderate-hazard storms into catastrophic events.

At the same time, mining activities over the last century scattered mine tailing impoundments

across many modern landscapes. Since 2014, two mine-tailings dam failures have killed 219

people in Minas Gerais, Brazil (Silva Rotta et al., 2020) and major tailings dams have failed

in British Colombia, Canada and Australia (Cornwall, 2020). As these structures fill and age

around the world, the risk and hazard associated with them grows.

Post-wildfire mud and debris flows and mine-tailing failures share two risk management

challenges. First, the risks associated with these events are non-stationary increasing with

climate change and aging dams respectively. Second, these events carry so much sediment

load that they change the physics of the flows, making make the common flood warning and

mapping approaches inappropriate. The mud and debris flows from these events require

specialized computational approaches that are not available or widely used in most

production-level flood studies. This work introduced single phase, mud and debris

assumptions into one of the most widely used flood risk and mapping models and tested these

algorithms against high concentration laboratory debris flows.

Most alluvial sediment loads do not significantly affect the physics of the transporting fluid.

Standard river concentrations do not increase water viscosity or energy loss significantly.

Therefore, most flood risk models assume clear water and credibly simulate hydrodynamics

This article is protected by copyright. All rights reserved.


with the shallow water flow equations (SWEs). However, extreme sediment loads can

change the fluid-particle mixture properties. When sediment concentrations are high enough

that the particles start to interact with each other (directly or indirectly), the classic SWEs do

not perform well. Accounting for these departures from the SWEs is critical to predict the

impact and manage the risk associated with these events, which pose serious hazards to life

and property.

The USACE has developed a library of existing rheological and geotechnical approaches to

simulate mud and debris flows (Floyd et al., 2019). This library has been incorporated into

several USACE models, including the Hydrologic Engineering Canter’s River Analysis

System (HEC-RAS), a production-level software widely used to evaluate flood risk.

Before release, the study team tested these non-Newtonian algorithms against a selection of

laboratory experiments. This work documents numerical simulations of the Parsons et al.

(2001) flume experiments with HEC-RAS and DebrisLib. The Parsons et al. (2001)

experiments include a wide range of sediment mixtures (including runs with much higher

concentrations than most laboratory studies) and span the process boundary between mud and

debris flows, making them particularly useful for evaluating model performance how the

suite of algorithms function under different conditions.

This work had four main objectives:

1. Verify the non-Newtonian algorithms included in DebrisLib and HEC-RAS against a

suite of high concentration laboratory experiments that include geophysical flows in

the mud and debris range.

This article is protected by copyright. All rights reserved.


2. Compare the performance of three single-phase, non-Newtonian closures for high

concentration geophysical flows with different grain-class compositions.

3. Evaluate the sensitivity and the relative importance of the individual, internal-shear

components of three non-Newtonian equations (Bingham, O’Brien, and Herschel-

Bulkley).

4. Explore the role of lateral mixing and turbulence parameters on the lateral velocity

distribution of high-concentration, high-viscosity flows.

Methods

This study compared a suite of rheological models that account for inter-particle dynamics

(incorporated in DebrisLib and implemented in HEC-RAS) to the results of the Parsons et al.

(2001) flume experiments. Therefore, the methods document the models applied (Non-

Newtonian Modeling Section) and the experiments used to evaluate them (Parsons et al.,

(2001) Flume Experiments) in turn.

Non-Newtonian Modelling

Incorporating Inter-Particle Effects in the Shallow Water Equations

Most hydraulic and sediment transport models simulate fluid flow with the shallow-water

flow equations, which pair the continuity and momentum equations to solve for two

variables: velocity and water surface elevation. While, HEC-RAS has incorporated non-

Newtonian effects in both the one-dimensional (1D) and two-dimensional (2D)

hydrodynamic models this paper only presents the 2D model theory and results. HEC-RAS

2D solves either the depth-averaged Shallow Water Equations (SWE) or the Diffusion Wave

Equation (DWE). The SWE model solves volume and momentum conservation equations

This article is protected by copyright. All rights reserved.


and includes temporal and spatial accelerations as well as horizontal mixing, while the DWE

model ignores acceleration and mixing processes, making it less accurate and more

computationally efficient. Users can select SWE or DWE equations and usually chose based

on a trade-off between accuracy (which favors SWE) or run time and stability (which favor

DWE). In many prototype-scale, clear water applications, the DWE results are very close to

the SWE result and substantially faster. However, that was not the case in these simulations

(see section on Turbulence Closure and Lateral Mixing).

The 2D volume conservation of the water-solid mixture is:

𝜕𝜂
+ ∇ ∙ (ℎ𝑉) = 𝑞
𝜕𝑡

where 𝜂 is the mixture surface elevation, t is time, ℎ is the water depth, 𝑉 is the velocity

vector, and q is a source or sink term, to account for external and internal fluxes. The depth-

averaged momentum conservation equations is:

𝜕𝑉 1 𝜏 𝑉
+ (𝑉 ∙ ∇)𝑉 = −𝑔∇𝜂 + ∇ ∙ (𝑣𝑡 ℎ∇𝑉 ) − 𝑐𝑓 𝑉 −
𝜕𝑡 ℎ 𝜌𝑚 ℎ |𝑉 |

where 𝑔 is the gravitational acceleration, 𝑣𝑡 is a turbulent eddy viscosity, 𝑐𝑓 is a bottom

friction coefficient, 𝜏 is the non-Newtonian internal stress, 𝜌𝑚 is the density of the water-

solid mixture, and |𝑉 | is the magnitude of the velocity vector. he second term on the right-

hand side or the momentum equation represents the horizontal mixing due to turbulence. For

debris flows, this term also includes horizontal mixing due to particle collisions. The

conservative form of the mixing terms is essential for accurate momentum conservation.

HEC-RAS includes momentum loss from bottom friction with the Manning’s roughness

coefficient:

𝑔𝑛2
𝑐𝑓 = 4/3 |𝑉 |
𝑅

This article is protected by copyright. All rights reserved.


where 𝑅 is the hydraulic radius.

Non-Newtonian processes add an additional loss term to the momentum equation, expressed

as an internal fluid stress 𝜏 in the far right hand term the equation.

HEC-RAS also includes a simplified, unsteady, hydrodynamic model, which replaces

momentum with the Diffusion Wave Equation (DWE):

𝜕𝜂
+ ∇ ∙ (𝛽∇𝜂) = 𝑞
𝜕𝑡

where 𝛽 is a non-linear coefficient which is a function of the bottom friction and non-

Newtonian internal stress:

𝐾 𝑅
𝛽=
𝐴 |∇𝜂|1/2
𝐾
The 𝐴 term (𝐾 is the conveyance, 𝐴 is the vertical area) accounts for losses, including classic

boundary friction and an internal loss term:

−1/2
𝐾 𝑛2 𝜏
= ( 4/3 + )
𝐴 𝑅 𝛾𝑚 𝑅|𝑉 |2

where 𝛾𝑚 is the unit weight of the water-particle mixture.

Both equations incorporate non-Newtonian processes as an internal, fluid, shear stress. Both

equations also divide the internal stress by the density of the fluid and depth/hydraulic radius,

for a term analogous to the friction slope from the boundary friction.

Therefore, if the impacts of inter-particle processes in high-concentration, geophysical flows

can be quantified as a shear stress, they can be incorporated into the flow equations. When

the non-Newtonian stress is equal to zero, the above modified SWE and DWE equations

This article is protected by copyright. All rights reserved.


reduce to the clear-water equations utilized in HEC-RAS. Single phase non-Newtonian

models use rheological models to compute this internal shear stress.

Estimating Stress-Strain Relationships across the Geophysical Flow

Taxonomy

The terminology surrounding geophysical flows can be fraught. Multi-axis gradients

confounded this taxonomy. High-concentration fluid behavior depends on both the

concentration and the gradation of the solid phase, so the taxonomy generally unfolds along

two axes (that can be, but are not necessarily, correlated). Most taxonomies (Coussot and

Meunier, 1996; Philips and Davis, 1991) classify these flows based on some relationship of

the total solid fraction and the percentage of those solids that are fine (i.e. silt and clay, or less

than 63 microns).

Philips and Davis (1991) quantified this multi-axis classification with the ternary taxonomy

in Error! Reference source not found.. They classified mud and debris flows (grey zone in

Error! Reference source not found.) as the events where water and solids in the fluid

transport at approximately the same velocity.

Because these flows behave differently based on the relative solid and fine content, different

rheological and geotechnical models are appropriate to estimate the internal shear (and

momentum loss) for different flows.

Computing Internal Losses with Rheological Models

This article is protected by copyright. All rights reserved.


Table 1 includes a common geophysical flow classification, the dominant inter-particle

processes associated with each, and the rheological model DebrisLib and HEC-RAS use to

simulate these processes. Rheology quantifies material deformation in response to stress, so

rheological models usually express these responses as relationships between stress and strain.

Therefore, identifying an appropriate (or approximate) rheological model for different

geophysical flow regimes can quantify the shear stress associated with the internal losses, and

compute an internal momentum loss.

Standard alluvial flows, with volumetric solid content (Cv) less than 5%, are modelled as

Newtonian flows, which have a linear stress-strain relationship and a zero intercept (i.e. any

stress induces strain). As Cv increases above 5% (hyper concentrated flows), DebrisLib

begins to simulate the flow as a Bingham plastic:

𝑑𝑣𝑥
𝜏 = 𝜏𝑦 + 𝜇 ( )
𝑑𝑧

where 𝜏 is the computed, internal shear stress, 𝜏𝑦 is the yield strength,  is the sediment-laden

𝑑𝑣
viscosity, and ( 𝑑𝑧𝑥 ) is the material strain. Modelling hyper-concentrated flows with a

Bingham model has a long tradition (Jin and Fread, 1997; Coussot, 1997; O’Brien et al.,

1993; Huang and Garcia, 1997). The stress-strain relationship in a Bingham plastic is still

linear (like the Newtonian assumptions of shallow-water flow), but the slope of the

relationship increases to reflect the higher viscosity of the fluid. The Bingham model also

has a non-zero yield strength. This yield strength incorporates a range of stresses over which

the fluid does not move and allows the fluid to “run out,” or come to rest when the internal

particle strength is greater than the applied stress. Prototype debris flows do, in fact, run out,

but Newtonian simulations (with zero stress-strain intercepts) do not.

This article is protected by copyright. All rights reserved.


As concentration increases and solid phase gradations coarsen, turbulence and grain

collisions also affect the fluid physics (Julien, 2010). As inter-particle turbulence and

collisions become important, DebrisLib applies non-linear rheological models (Table 1).

This transition to mudflows and grain flows is not strictly a function of concentration and

grain size. Hydrodynamics also determines when these processes emerge or dominate.

DebrisLib includes two main approaches to simulate these non-linear rheologies associated

with inter-particle turbulence and collisions. First, the O’Brien equation method (O’Brien et

al., 1993) adds two quadratic terms to the Bingham model to account for these processes,

adding viscous forces for mudflow and/or a Bagnold dispersive term for grain flows (which

overlap with debris flows in most taxonomies):

1⁄ −2
𝑑𝑣𝑥 2
𝑑𝑣𝑥 2 0.615 3 𝑑𝑣𝑥 2
𝜏 = 𝜏𝑦 + 𝜇𝑚 ( ) + 𝜌𝑚 𝑙𝑚 ( ) + 0.01𝜌𝑠 (( ) − 1) 𝑑𝑠2 ( )
𝑑𝑧 𝑑𝑧 𝐶𝑣 𝑑𝑧

where  is the computed, internal shear stress, y is the yield strength,  is the sediment-laden

viscosity, m is the mixture density s is the solid density, ds is the representative grain size,

and lm is the Prandtl mixing length. Strain is often computed as some ratio of velocity and

depth (which is a depth-integrated expression of deformation). Julien (2010) expresses strain

𝑑𝑣
( 𝑑𝑧𝑥 ) as 3𝑢/ℎ, three times the ratio of velocity to depth. DebrisLib follows this convention,

using 3𝑢/ℎ for strain in all equations.

By simulating mud and grain flows with quadratic terms, the O’Brien approach assumes

these geophysical flows are dilatant or “shear-thickening” processes—the material becomes

less deformational in response to stress, or each increment of additional stress causes less

strain.

This article is protected by copyright. All rights reserved.


DebrisLib also includes the Herschel-Bulkley (HB) approach to simulate these non-linear

stress-strain processes:

𝑑𝑣𝑥 𝑛
𝜏 = 𝜏𝑦 + 𝐾 ( )
𝑑𝑧

The Herschel-Bulkley approach is more empirical. By diverging from a linear model, the

coefficient (K) in the non-linear term no longer has units of viscosity and cannot be directly

measured. This model also does not try to build processes-based terms from similitude like

O’Brien et al. (1993). HB is more flexible, allowing non-quadratic, non-linear terms (n ≠ 2),

so it can simulate pseudoplastic and “shear-thinning” behaviour by selecting n < 1, or

different magnitudes of dilatant, shear-thickening processes with n > 1. HB also essentially

collapses to the Bingham model (and K) if n = 1. DebrisLib also includes geotechnical

approaches (Coulomb and Vollemy) that account for friction forces as the concentrations

begin to diverge from fluid or plastic models and become more clastic, but these were not

included in this work.

DebrisLib includes a broad array of non-Newtonian algorithms, including the approaches in

one of the first widely-used mud and debris flows models, Flo-2D. The current version of

HEC-RAS (6.0) is a fixed bed model, so this analysis should be applicable to other single-

phase, fixed-bed, non-Newtonian models used for flood risk management (like Flo-2D) but

may be less applicable to smooth particle Lagrangian models (e.g. Dan3D) or models with

active bed sources (e.g. RAMMS) (Schraml, 2015).

Parsons et al. (2001) Flume Experiments

The development team validated DebrisLib within HEC-RAS by comparing computed and

measured plug velocities and lateral velocity distributions. Some of the experiments (Parsons

This article is protected by copyright. All rights reserved.


et al., 2001, Mainalli and Rajaratnum, 1994, Haldenwang et al., 2010) used to test the non-

Newtonian capabilities in HEC-RAS are plotted on the Phillips and Davies (1991) ternary

taxonomy of geophysical flows Error! Reference source not found.. For comparison, two

reservoir flushing events are also included in Error! Reference source not found..

Reservoir flushing events have artificially high concentrations for fluvial flows, concentrating

at least a year of sediment load into a short release window. Both the Spencer Dam flush

(Gibson and Boyd, 2016a,b) on the Niobrara River and the Fall Creek flush (Schenk and

Bragg, 2014; Gibson and Crain, 2019) generated high concentrations (by fluvial transport

standards) and significant deposition in downstream reaches. However, they do not even

approach hyper-concentrated flows on this scale.

The Parsons et al., (2001) experiments had much higher concentrations than most of the

laboratory experiments available and included a broad range of cohesive and cohesionless

particle mixes, including simulations that were classified as both mud flows and grain/debris

flows (Figure 1). The high concentrations (Cv = 68–74%) and wide range of gradations (d50 =

0.2–0.6 mm with fine components between 15.5% and 43%) make these experiments

particularly useful to test the mud and debris approaches in HEC-RAS.

Additionally, the Parsons et al. (2001) experiments provide a good context to test the limits

of the rheological approach. The range of concentrations and gradations used in these

experiments generated a variety of “snout effects,” which can complicate modelling of multi-

phase debris flows with a single-phase, rheological approach. Parsons et al. (2001) also

measured yield strength and viscosity of each material (with a tilt table) and back-calculated

both Bingham and HB parameters from their results.

This article is protected by copyright. All rights reserved.


Parsons et al. (2001) performed 24 experiments (which did not stop before the end of the

flume) with four different slopes, two different flume widths, and seven different material

types. This study simulated ten of those experiments (1a, 2a, 3a, 4a, 5a, 5i, 6a, 7a, 8a, 8b),

selecting all of the experiments conducted with their largest flume (10-m length and 14.6-cm

diameter) at two slopes, 18.6% (10.7o or 0.1857) and 21.3% (12.2o or 0.2113). Parsons et al.

(2001) glued sand to these flumes to introduce a roughness element.

Table 2 lists the experimental parameters and the back-calculated (fit) Bingham parameters

Parsons et al. computed for the ten experiments modelled. Table 2 is ordered from top to

bottom by the observed plug velocity (0.24 m/s to 1.02 m/s). An expanded table of

experimental parameters is included in the Supplemental Materials. The sub-set of modelled

experiments includes the full range of material types (Figure 1). The velocities measured in

the ten modelled experiments span most of the observed range and are well-distributed across

that range.

Fit viscosities were mostly in the 1.5-2.0 Pa-s range, except for experiments 8a and b, which

were nearly twice that (4.1 Pa-s) due to high fine content, and experiment 4, which had

almost no fine material and therefore, almost no measured viscosity (0.06 Pa-s). The HB

parameters computed for these experiments almost all indicate shear-thinning (n < 1), except

for the higher viscosity, higher fine content experiments (8a and 8b).

Parsons et al., (2001) recorded both the “plug velocity” and “snout velocity”, which they

measured by sprinkling black beads across the surface of the fluid and tracking the rate and

deformation of the tracer. They also used these measurements to record a lateral, surficial

velocity distribution (discrete velocities across the flume) for several experiments. The

experiments had relatively distinct “plug flow” and “deformation” zones. The plug flow

This article is protected by copyright. All rights reserved.


zone, which included approximately the central half to two thirds of the flume (~3.5–11.5 cm

of the 14.6-flume width), had nearly constant velocity. Velocities across the transitional

zones dropped from the plug flow velocity to zero at the flume boundary, approximately

linearly. The near-linear velocity distribution across these lateral transition zones led the

authors to compute strain rates from the velocity distributions in these near-wall zones.

Parsons et al., (2001) also provided qualitative descriptions of the “snout effects” classifying

the snout effects as “none,” “some,” “strong,” and “stopped.” The ten simulated experiments

were distributed between the first three classifications, including three with no snout effects,

three with “some,” and four with “strong” snout effects.

Model Setup

The modelling domain include 3,996 cells. Cells were 3-cm long and 1.183-cm wide. The

cell width was selected to fit 12 equal-width cells symmetrically across the 14.7 cm flume.

The Courant limited adaptive time step was used with a user-specified range for the Courant

number of 0.2 to 1. This resulted in time steps between 0.002 and 0.1 s. All simulations used

a Manning’s roughness coefficient of 0.16, which was selected before the study based on the

sand flume lining and was not calibrated. The model also used horizontal turbulent mixing

using relatively standard, clear-water, mixing parameters and did not calibrate these. These

parameters included a longitudinal mixing coefficient of 0.3, a transverse mixing coefficient

of 0.1, and a Smagorinski coefficient of 0.05. These simulations used the SWEs solved the

Eulerian-Lagrangian Method (ELM) unless otherwise specified (see discussion on DWE

below). The study team reduced most of the convergence and computational tolerances in

HEC-RAS, which is standard practice when simulating a flume. Most of the default

tolerances are set for prototype, reach-scale analyses (O(102-104 m)) and should be reduced

This article is protected by copyright. All rights reserved.


for this laboratory scale (O(10-2 m)). The Supplemental Materials of this paper include text

and videos that demonstrate the step-by-step process used to create these flumes and simulate

these flows in HEC-RAS 5.1.

Calculating Results

Parsons et al., (2001) reported two observed results: plug velocity and snout velocity. It is

important to compare those observations to the appropriate simulated results. Because this

study required many simulations and analysed non-typical results for each (not easily

accessible through the interface), the study team developed R scripts to cycle through the

HEC-RAS result from all of the simulations to develop the corresponding “computed”

velocities. HEC-RAS result files are in the Hierarchical Data Format - version 5 (HDF5) and

the Supplemental Materials include sample R code that access these data and compute this

result.

Measured plug velocities represented a relatively consistent velocity measured across the

middle half to two-thirds of the flume. HEC-RAS represented the numerical flume with

twelve cells across its diameter. Therefore, this analysis compared the observed plug velocity

to a computed equilibrium average velocity of the middle six cell faces (excluding three cells

on either side with boundary effects; See Lateral Velocity Results Section for justification of

this assumption). The analysis computed this average velocity at the longitudinal center of

the flume (5 m from upstream boundary), 59 seconds into the simulation (after the model

achieved steady-state equilibrium).

The analysis compared the measured snout velocity to a computed “front velocity,”

determined by recording the elapsed time at the first occurrence of a non-zero depth at the

This article is protected by copyright. All rights reserved.


central, downstream computational cell, and dividing the flume length (10 m) by that arrival

time.

The modelling team also compared the measured lateral velocity distributions to those

computed by HEC-RAS (with both SWE and DWE) in two low-residual simulations (2a and

7a) - without strong snout effects - to evaluate the model’s lateral velocity performance and

the impact of hydrodynamic solution and the turbulence parameters.

These models were not calibrated. The study team used the non-Newtonian parameters from

the Parsons et al. (2001), and selected other sensitive parameters (flume roughness and

mixing parameters described in the Model Setup Section) based on qualitative experiment

descriptions and previous experience. The results reported use those a priori, best-estimate

parameters. However, the results are sensitive to these parameters, and if the modeling team

had started with different estimates (which would have been possible within the reasonable

range of variation of these parameters) the results could have included systemic error and

calibration may have been necessary.

Results

The ten experiments in Table 2 were simulated in HEC-RAS with standard Newtonian

hydrodynamic assumptions (SWE with parameterization described in Model Setup) as well

as three of the non-Newtonian closures described in Table 1 (Bingham, O’Brien, and

Herschel-Bulkley). Figure 2Figure 2 includes simulation results from Experiment 1a, with

the Newtonian, clear-water, shallow water flow equations (top) and the Bingham simulation

(bottom). Both simulations used SWE and the same, pre-selected, mixing coefficients

(Longitudinal=0.3, Transverse = 0.1, and Smagorinski=0.05) but the Bingham simulation

included a linear internal stress term based on Parson et al.’s (2000) ‘fit’ parameters from

This article is protected by copyright. All rights reserved.


Table 2 (yield=80 Pa and mixture viscosity=1.48 Pa-s). The Bingham simulations are deeper

and slower (in terms of both front and average velocity), and have a similar snout shape as

observed in the experiments.

Bingham Plastic Velocities for All Experiments

Observed plug velocities and snout velocities are plotted against their comparable, computed

velocities in Figure 3. These results include best-estimate a priori hydraulic parameters

described in the Model Setup Section, and the Parsons et al. (2001) back-calculated Bingham

parameters from Table 2. Results include standard Newtonian clear-water flow (SWE) and

the Bingham simulations. The standard hydrodynamic equations overestimate the average

velocity substantially (average error 0.98 m/s, RMSE = 0.99 m/s). The Bingham simulations,

fall along the line of unity, with an average error of 0.047 m/s and a RMSE of 0.21 m/s.

Newtonian snout-velocity errors were lower than average water velocity results (average

error = 0.74 m/s, RMSE = 0.77 m/s), and higher Newtonian snout velocities were closer to

the observed results than lower velocities. The Bingham snout-velocity results were, again,

much closer to the observed values (average error = 0.10 m/s, RMSE = 0.29 m/s) than the

Newtonian results. Non-Newtonian snout velocity residuals were higher than average

velocity residuals. However, stratifying these results by snout effects (Figure 3-Left)

accounts for some of the variation in the Bingham snout results (See Discussion). The four

Bingham simulations that over predicted snout-velocity the most were also the four

experiments with strong snout effects.

In addition to over-estimating velocity the Newtonian results also underestimate the depth of

the flow. By continuity principles, the slower, mud and debris flows run deeper, which is

critical to forecasting the actual flood risk of these events. Parsons et al. (2000) did not

This article is protected by copyright. All rights reserved.


measure flow depth in their experiments, but the HEC-RAS results did generate deeper flume

flows consistent with depths back-calculated from their results. For example, the Bingham

simulation pictured in Figure 2 was more than twice as deep (6.2 cm) as the Clear water

simulation (3.0 cm).

Lateral Velocity Results

HEC-RAS replicates the lateral velocity distribution of the experiments well, with the

Bingham non-Newtonian closure and the default turbulence and mixing parameters. The

observed lateral velocities from experiments 1a and 7a are plotted with computed lateral

Bingham velocities in Figure 4. An R script extracted computed velocities from each of the

twelve cell faces across the middle of the numerical flume in HEC-RAS, 59 s into the

simulation, to capture the equilibrium velocity transect. Experiments 1a and 7a were selected

to evaluate lateral velocity performance because they had low average residuals (Figure 3)

without strong snout effects and bounded most of the velocity variation of the results.

The simulated velocities, with default turbulence and mixing coefficients, are a little more

parabolic than the observed data. They do not have a constant plug-flow zone, but the

variation across the middle half of the flume is low. The simulated wall-velocity transitions

were slightly more gradual in Experiment 1a, but less gradual (higher velocities in marginal

cells) in Experiment 7a. However, most of these residuals are small. On the whole, the

parabolic simulation results with the default mixing parameters and the Bingham plastic

model simulated the lateral velocities well.

O’Brien Quadratic and Herschel-Bulkley Results

The left-hand plug velocity panel of Figure 3 is replicated in Figure 5, with results from the

non-linear rheological models (O’Brien and HB). Because the O’Brien equation adds two

This article is protected by copyright. All rights reserved.


quadratic terms to the Bingham model, it computes more internal losses than the Bingham

model and therefore over-predicts losses while under-predicting velocities when the fit

Bingham parameters are used (average error = –2.4 m/s, RMSE = 0.31 m/s; see Discussion).

HB, on the other hand, generated velocities very close to Bingham, since the HB parameters

were fit to the same conditions as the Bingham parameters for most experiments.

The O’Brien equation generally over predicted internal losses (under predicted velocities)

with the fit Bingham parameters. However, the fit Bingham parameters were generally

higher than the measured parameters, so it may be more appropriate to evaluate the O’Brian

approach with measured parameters (see Section on Corrective Components of the O’Brian

Quadratic in the Discussion).

Discussion

Error Analysis

All the non-Newtonian simulations in Figure 5 out-performed the Newtonian, clear-water

simulations. However, they still diverged from observed velocities in some cases. The

Bingham and HB simulations tended to over-predict the lower velocities and under-predict

the higher velocities. Parsons et al. (2011) computed the Bingham and HB parameters from

the plug velocities, which accounts for the some of the residuals associated with the snout

velocities (Figure 3). The plug-flow errors are smaller, but probably emerge due to the strain

definition in the parameter estimation.

HEC-RAS and DebrisLib used a different strain definition than the experimental team used to

calculate the non-Newtonian parameters. Parsons et al. (2011) measured strain based on

tracer deformation between the flume wall and the plug flow region. HEC-RAS and

DebrisLib compute strain from the ratio of the velocity and depth in each computational cell.

This article is protected by copyright. All rights reserved.


The computed strain in DebrisLib is three times the ratio of the cell velocity to the fluid depth

in the cell (3𝑢/ℎ). This is a common computational approach to computing strain (Julien,

2011) but yields different results than the direct measurement used to fit the experimental

parameters. This difference in strain relationships explains the computational residuals,

despite using “fit” parameters. It also explains why the Bingham and HB results often have

similar residuals.

The two experiments with the highest residuals used unique materials. Experiment 3a stands

out with the highest positive residual (i.e. computed velocity > observed velocity).

Experiment 3a was one of the special-case experiments, which used the coarsest material,

with low yield strength and almost no viscosity, but strong snout effects. Experiment 6a has

the largest negative residual (computed velocity < observed velocity). The material in

Experiment 6a is similar to the material in experiment 3a but with an additional 3% clay.

This increased the Bingham parameters by 20%, which slowed the simulation significantly.

However, the effect on the numerical flume was bigger than the observed impact.

Experiment 6a is also the only experiment with an HB power significantly higher than 1

(indicating shear-thickening). Finally, despite strong snout effects, Experiment 6a was one of

the only experiments where the measured snout velocity exceeded the plug velocity (the other

was 7a, making the two Usnout > Uplug experiments the highest snout velocity experiments).

So, despite strong snout effects, the model under-predicted plug velocity and over-predicted

snout velocity.

Snout Effects

Two interesting observations emerge from stratifying results by snout effects in the left panel

of Figure 3. First, the higher snout velocity residuals support Iverson’s (2003) assertion that

This article is protected by copyright. All rights reserved.


single-phase, fixed rheology models tend to over-predict front velocities and under-predict

arrival times when fluid concentration and gradation are longitudinally heterogeneous. This

single-phase, rheological approach will introduce arrival time errors with strong snout effects,

but from an emergency management perspective, these are conservative errors. Second,

while the experiments with strong snout effects also had higher average velocity residuals,

these average velocity errors were much smaller than the snout velocity errors (Table 3).

Rheological models are single-phase heuristics for multi-phase flows. They do not capture

the inter-particle effects explicitly and cannot account for discrete solid phase processes—

particularly localized processes like snout formation (Iverson, 1997, 2003) where the solid

phase becomes periodically dominant. However, they can still provide a helpful framework

to transfer flow properties from similar events to improve mud- and debris risk estimates.

Single phase models are approximate but they are easier to use and parametrize and scale to

production-level risk management studies.

Turbulence Closure and Lateral Mixing

HEC-RAS and DebrisLib replicated the lateral velocity distribution relatively well (Figure 4)

with relatively standard turbulence parameters. However, without horizontal turbulent

mixing, the lateral velocity distribution diverged significantly from the measured data. Figure

6 replicates the observed and computed lateral velocities from Figure 4, but also includes

Bingham simulations without horizontal mixing and with the DWE (which does not include

inter-cell turbulent momentum transfer).

Both the DWE and SWE simulations without turbulent mixing concentrate flow in the center

of the flume, diverging much more from the observed plug-flow condition than the SWE

solution with horizontal mixing.

This article is protected by copyright. All rights reserved.


Corrective Components of the O’Brien Quadratic

Parsons et al., (2011) reported two sets of Bingham parameters: those which they measured

directly with tilt tests, and others which they fit to the observed velocities. In most cases, the

fit parameters combined for a higher internal shear value than the measured values (see the

analysis of measured vs. fit parameter results in Supplemental Materials). The measured

Bingham and HB parameters did not account for all the internal losses.

The results in Figure 5 use the fit parameters. The O’Brien quadratic consistently over-

predicted internal losses (systematically under-predicted velocities) with the fit parameters

(Figure 5). However, the O’Brien model includes quadratic (dispersive and turbulent) terms

to quantify additional processes. Re-running the O’Brien simulations with the (mostly lower)

measured parameters improved the RMSE from 0.31 m/s (fit) to 0.23 m/s (measured), which

is on the order of the Bingham residuals (0.21 m/s) (Figure 7).

By fitting parameters to Bingham and HB that compute more internal loss than the measured

parameters, Parsons et al. (2000) may be accounting for additional internal loss processes that

these models do not include. O’Brien’s additional, quadratic terms were not measured and

cannot be evaluated empirically, but they move the results in the right direction when paired

with the actual, measured parameters.

However, the O’Brien quadratic still consistently over-predicted losses (i.e. under-predicted

velocities) for experimental flows faster than 0.7 m/s. Larger losses at higher velocities are

likely a function the squared strain terms ((3𝑢̅/ℎ)2 ) in the O’Brien equation (assuming

substantial shear-thickening), while Parsons et al. (2001) found that most of these materials

were characterized by shear-thinning (HB n < 1).

This article is protected by copyright. All rights reserved.


Because of the high concentration and coarse sediment components, these experiments have

high Bagnold and Friction numbers, which will drive these materials out of a simple linear,

plastic model (Iverson, 1997; Julien, 2010). At lower velocities, it does appear that adding

these theoretical terms to the measured Bingham parameters accounts for the adjustments

required to fit the Bingham equation to the observed experiment results. However, at higher

velocities, squaring the 3𝑢̅/ℎterm seems to over-predict loss.

Relative Contributions of Components of the Loss Equations

Figure 8 plots the internal shear stress computed by DebrisLib for each simulated experiment

with the three main approaches: Bingham (top), O’Brien (middle), and HB (bottom). The

figure also separates the relative contributions of the components of each shear approach.

Bingham and O’Brien use the same yield and viscosity parameters, so the yield components

are identical between the simulations. However, the viscous component is a function of the

fluid velocity. The O’Brien simulations have lower velocities because of the additional

internal loss components, so their viscous components are smaller despite identical

viscosities. The Yield component was larger than the viscous component in all of the

experiments except 5 (where the fit yield strength of 14 Pa was much lower than the

measured yield strength of 67 Pa, as well as the fit yield strengths from all other experiments)

and 8 (which used the fine material with the highest measured and fit viscosity).

Turbulent stress was the largest term in most of the O’Brien simulations. The dispersive term

was only significant in the coarsest experiment (see Section 0). The HB simulations had

higher non-linear (quasi-viscous) terms and smaller yield components. Therefore, K was

This article is protected by copyright. All rights reserved.


substantially larger than  in most cases, to compensate for the smaller yield stress and the

HB power (n < 1).

Applying the Bagnold Dispersive Stress to Well Graded of Natural Materials

O’Brien et al.’s (1993) Bagnold (1954) approach the dispersive term (which accounts for

particle collisions) only substantially contributed to internal losses for the coarsest

experiment (4a). However, the study team encountered an obstacle applying this term for

these high-concentration flows of well-graded (poorly sorted) materials.

This term returns counter-intuitive results (and can become unstable) for high-concentrations

of natural, well-graded (poorly sorted) materials. The Bagnold term computes the ratio

between the volumetric concentration and the maximum possible volumetric concentration

derived from packing theory of uniform spheres (i.e. Cv,max = 61.5%).

3𝑢̅ 2
0.01𝜌𝑠 𝑑𝑠2 ( ℎ )
𝜏𝐷𝑖𝑠𝑝𝑒𝑟𝑠𝑖𝑣𝑒 = 2
1⁄
0.615 3
(( 𝐶 ) − 1)
𝑣

This term becomes indeterminate at Cv = 61.5% and is inversely related to grain size at

higher concentrations, including all concentrations used in this experiment. The blue lines in

Figure 9 plot this term for the smallest and largest d50 measurements and the range of

concentrations (grey box) used in this study.

DebrisLib changed Cv,max to 84% to move it closer to actual, observed maximum

concentrations associated with events composed of natural materials (Rickenmann, 1991;

Rickenmann, 2001; Ancey, 2007). Changing the maximum concentration to 84% shifts the

equation so that the collision losses increase monotonically over the observed range in these

This article is protected by copyright. All rights reserved.


experiments (brown lines in Figure 9). However, the decision to leave the rest of the

equation intact (particularly the 0.01 coefficient), without compensating for this change, is

speculative.

The quadratic sensitivity of the Bagnold term to grain size (d) also introduces some

complexities. First, it is difficult to collapse a well-graded (poorly sorted) material into a

representative grain size. Selecting a representative grain size for these experiments is

speculative, but it becomes more difficult in a natural debris flow, which can have high clay

content and carry a significant component of particles larger than one meter. Second, as a

dimensional parameter, the grain size term does not scale to flow depth. Therefore, it may

not be equally applicable to both a 15 cm flume and a prototype event.

However, in the world of risk management, modellers have to simulate these events without

the benefit of laboratory measurements or HB parameters (which cannot be directly

measured, only fit), and must compute losses explicitly. In these scenarios, making use of a

theoretical model and an easily measured parameter (grain size) has significant advantages.

Conclusion

The rheological, non-Newtonian approaches in DebrisLib improved simulations of

experimental mud and debris flow, common in post-wildfire flooding and impoundment/dam

breach failures. All non-Newtonian approaches had substantially lower RSME than the

Newtonian SWE. Simulated snout velocities were closer to measured velocities for the

experiments with less pronounced snout effects. The model also simulated the lateral

velocity distribution well (despite a more parabolic solution through the plug flow zone) with

standard hydraulic and mixing coefficients. However, the computed results diverged from

the observed lateral velocity distribution without the mixing models (or with the DWE). The

This article is protected by copyright. All rights reserved.


study team encountered some challenges applying the quadratic terms of the O’Brien

equation, but also recognized the value of a theoretical model that reduces requirements for

parameters that are difficult (or impossible) to quantify in real-time risk management

scenarios.

Post-wildfire and dam breach disasters are becoming more common. A validated, widely

applied software with mud and debris flow capabilities will help the flood risk community

manage these risks. These non-Newtonian algorithms are now available in HEC-RAS, and

DebrisLib can be connected to other research and production-level software. However, wider

application of these algorithms highlights the need to interrogate the assumptions in each

approach, and evaluate the terms for each scenario.

Acknowledgements

This work was funded by the Post Wildfire Flood Risk Management Research Unit in the US

Army Corps of Engineers Flood and Coastal Storm Damage Reduction Research and

Development Program (F&C). The authors would like to thank Dr. Parsons for providing

experimental files and the lateral velocity data plotted in Figure 4 and Figure 6.

Data Availability

The data sets used in this study will be released with HEC-RAS version 6.0 and are available

from the corresponding author on reasonable request.

Conflict of Interest: None.

Supplemental Materials

The supplemental materials include a detailed table of experimental parameters, an analysis

of the measured vs. fit Bingham parameters, and detailed, replicable, step-by-step methods

This article is protected by copyright. All rights reserved.


that describe the process of modelling and simulating these experiments in HEC-RAS. They

also include three videos that demonstrate how to create and simulate one of these semi-

circular flume models (e.g. https://youtu.be/kAZhWw-j0HI).

References

Ancey, C. (2007) Plasticity and geophysical flows: a review. Journal of Non-Newtonian

Fluid Mechanics. 142(1-3), 4-35.

Bagnold, R. A. (1954) Experiments on a gravity-free dispersion of large solid spheres in a

Newtonian fluid under shear. Proceedings of the Royal Society of London. Series A.

Mathematical and Physical Sciences, 225(1160), 49-63.

Cornwall, W. (2020) Catostrophic failures raise alarm about dams conaining muddy mine

wastes, Science Magazine, doi:10.1126/science.abe3917.

Coussot, P. and Meunier, M. (1996) Recognition, classification and mechnical description of

debris flows Earth Science Reviews, 40, 209-227.

Coussot, P. (1997) Mudflow rheology and dynamics. CRC Press Book, Rotterdam.

Netherlands, 263 p.

Gibson, S. and Boyd, P. (2016) Monitoring, measuring, and modeling a reservoir flush on the

Niobrara River in the Sandhills of Nebraska. Proceedings, River Flow 2016, ed

Constantinescu et al., 1448-1455.

Gibson, S. and Boyd, P. (2016) Designing Reservoir Sediment Management Alternatives

with Automated Concentration Constraints in a 1D Sediment Model. Proceedings

International Symposium on River Sedimentation.

This article is protected by copyright. All rights reserved.


Gibson, S. and Crain, J. (2019) Modeling sediment concentrations during a drawdown

reservoir flush: Simulating the Fall Creek operations with HEC-RAS. Regional

Sediment Management Technical Note, TN-RSM-19-7, 10p.

Schenk, L.N., and Bragg, M. (2014) Assessment of Suspended-Sediment Transport, Bedload,

and Dissolved Oxygen during a Short-Term Drawdown of Fall Creek Lake. Oregon:

Winter 2012–13. Open File Report 2014. Reston, VA: U.S. Geological Survey.

Floyd, I., Gibson, S., Heath, R., Ramos-Villanueva, M., Pradhan, N. (2019) Development of

'Debris Library' and 1D HEC-RAS and 2D Adaptive Hydraulics Linkage-Architecture

for Predicting Post-Wildfire non-Newtonian Flows. Federal Interagency SEDHYD

2019, 24-28.

Iverson, R.M. (2003) The Debris Flow Myth. 3rd International Conference on Debris-Flow

Hazards Mitigation: Mechanics, Prediction, and Assessment, 303-314.

Jin, M., & Fread, D. L. (1997) One-dimensional routing of mud/debris flows using NWS

FLDWAV model. Debris-Flow Hazards Mitigation: Mechanics, Prediction, and

Assessment, 687-696.

Haldenwang, R. (2003) Flow of non-Newtonian fluids in open channels. Dissertation, Cape

Technikon, 384 p.

Huang, X. and Garcia, M. H. (1997) A perturbation solution for Bingham-plastic mudflows.

Journal of Hydraulic Engineering, 123(11), 986-994.

Iverson, R.M. (1997) The Physics of Debris Flows. Reviews in Geophysics, 35(3), 245-296.

This article is protected by copyright. All rights reserved.


Kean, J.W., Staley, D.M., Lancaster, J.T., Rengers, F.K., Swanson, B.J., Coe, J.A.,

Hernandez, J.L., Sigmand, A.J., Allstadt, K.E., Lindsay, D.N. (2019) Inundation, flow

dynamics, and damage in the 9 January 2018 Montecito debris-flow events,

California, USA: Opportunities and challenges for post wildfire risk assessment,

Geosphere, 15, doi.org/10.1130/GES02048.1.

Mainali, A. and Rajaratnam, N. (1994) Experimental study of debris flows. ASCE Journal of

Hydraulic Engineering, 120(1), 104-123.

O’Brien, J.S., Julien, P.Y., and Fullerton, W.T. (1993) Two dimensional water flood and

mudflow simulation. Journal of Hydraulic Engineering 119(2):244-261.

Parsons, J.D., Whipple, K.X., and Simoni, A. (2001) Experimental Study of the Grain-Flow,

Fluid-Mud Transition in Debris Flows, Journal of Geology, 109, 427-447.

Phillips, C.J. and Davies, T.R. (1991) Determining rheological parameters of debris flow

materials. Geomorphology, 4, 1010-110.

Rickenmann, D. (1999) Empirical relationships for debris flows. Natural Hazards, 19(1), 47-

77.

Rickenmann, D. (2001) Comparison of bed load transport in torrents and gravel bed streams.

Water resources research, 37(12), 3295-330.

Schraml, K., Thomschitz, B., McArdell, B.W., Graf, C., Kaitna, R. (2015) Modeling debris-

flow runout patterns on two alpine fans with different dynamic simulation models,

Natural Hazards and Earth System Sciences, 15, 1483-1492, doi:10.5194/nhess-15-

1483-2015

This article is protected by copyright. All rights reserved.


Silva Rotta, L.H., Alcantara, E., Park, E., Galante Negri, R., Nina Lin, Y., Bernardo, N.,

Goncalves Mendes, T.S. , Sousa Filho, C.R. (2020) The 2019 Brumadinho tailings

dam collapse: Possible cause and impacts of the worst human and environmental

disaster in Brazil, International Journal of Applied Earth Observation and

Geoinformation, 90, 102119, doi.org/10.1016/j.jag.2020.102119.

This article is protected by copyright. All rights reserved.


Figure 1: Philips and Davis (1991) ternary classification of geophysical flows with the
Parsons et al. (2001) data (as well as other hyper-concentrated experiments and two reservoir
flushes for context). The Parsons et al. (2001) points include all of the experiments in that
study while the Parsons (simulated) experiments are those modelled in this study.

This article is protected by copyright. All rights reserved.


Figure 2: HEC-RAS Simulation results of the upper 2 meters of the numerical flume after 2
seconds of experiment 1a with water (top) and a Bingham approach (bottom). The figure
also includes an image from one of the experiments from Parsons et al. (2001).

This article is protected by copyright. All rights reserved.


Figure 3: Comparison between observed plug and snout velocities with comparable HEC-
RAS results, including standard Newtonian shallow-water flow simulations, and Bingham
plastic simulations. The Bingham results are labelled with the experiment number in the
average velocity plot (left) and with the observed snout effect in the snout velocity plot.

This article is protected by copyright. All rights reserved.


Figure 4: Lateral velocity distribution across the flume in experiments 1a and 7a, plotted with
the lateral velocity distribution computed at the cell faces across the numerical flume in HEC-
RAS with the Bingham simulation.

This article is protected by copyright. All rights reserved.


Figure 5: Observed plug velocities plotted against the non-Newtonian, shallow-water flow
simulation for the ten experiments with three approaches to computing internal loss terms.

This article is protected by copyright. All rights reserved.


Figure 6: Measured lateral velocities from experiments 1a and 2a, with HEC-RAS Bingham
simulations, using three different hydrodynamic approaches. Horizontal mixing account for
lateral momentum exchange due to turbulence and particle collisions and spread out the
velocity distribution, reproducing experimental results better.

This article is protected by copyright. All rights reserved.


Figure 7: O'Brien simulations performed better with the measured parameters (▲), than the
fit parameters (∆). Bingham results (○) included for comparison. Lines link simulations of
the same experiment.

This article is protected by copyright. All rights reserved.


Figure 8: Shear components from simulations of the ten experiments with the Bingham (top),
O’Brien (middle), and Herschel-Bulkley approaches, with the fit parameters.

This article is protected by copyright. All rights reserved.


Figure 9: Theoretical Bagnold term with finest and coarsest grain size material (blue) and
with a higher, natural-materials, maximum volumetric concentration (Cv,max = 84%, brown).
The grey box indicates the range of the simulated experiments.

This article is protected by copyright. All rights reserved.


Table 1: Geophysical flow classification used in DebrisLib (and HEC-RAS) including the
incremental, additional terms added to account for each process, the rheological model
inferred by those terms, and the quantitative threshold used to identify the process.

*Iverson (1997) recommends Nfri>2,000 for this transition.


** 𝐴 = 𝜋𝑟 2
𝑑𝑣 3𝑢̅
𝛾̇ =strain ( 𝑑𝑧𝑥 ). DebrisLib follows O’Brien and Julien, quantifying strain with ( ℎ )

This article is protected by copyright. All rights reserved.


Table 2: Parameters for modelled experiments (ordered by observed plug velocity) from Parsons et al. (2011).

Observed Snout

Slope (degree)
Observed Plug

Yield Strength
Velocity (m/s)

Velocity (m/s)

Viscosity ()
Snout Effect
Experiment

Fit (Pa-s)
Flow (L/s)

Fit - (Pa)
d50 (mm)
Parsons

Cv (%)
Important Feature
1a 0.24 0.22 1.89 69.2 10.7 None 0.2 98 1.92 ↑ Yield/Finest Material
8a 0.40 0.32 2.0 68.2 10.7 Some 0.2 77 4.1 Highest Viscosity
Coarsest material, Very low
4a 0.41 0.3 2.75 73.1 12.2 Strong 1.3 67 0.06

8b 0.42 0.29 2.27 68.2 12.2 None 0.2 77 4.1 Highest Viscosity
3a 0.65 0.45 4.48 73.2 12.2 Strong 0.6 90 1.75
2a 0.83 0.57 3.91 71.8 10.7 None 0.43 80 1.48
7a 0.91 0.99 6.0 68.9 10.7 Some 0.58 68 1.5
5a 0.94 0.61 5.54 72.7 12.2 Strong 0.53 14 0.99 ↑ Flow, ↓Yield, ↑ d
5i 1.02 0.77 6.49 72.7 12.2 Strong 0.53 14 0.99 Highest Flow, ↓Yield, ↑ d
↑ Flow, Highest Yield
6a 1.02 1.1 5.73 71.4 12.2 Some 0.56 124 0.71
Strength
*5i corresponds to 5a2 in the paper
Bold values are “special case” coarse or fine experiments.

Table 3: Bingham simulation residuals for Average Velocity (left panel of Figure 3) and Front

Snout Effect n RMSE of Average/Plug RMSE of Front/Snout

All 10 Velocity
0.21(m/s) Velocity
0.29(m/s)
None 3 0.13 0.06
Some 3 0.21 0.31
Strong 4 0.26 0.37
n=# of experiments associated with this snout effect

This article is protected by copyright. All rights reserved.

You might also like