Download - Virtual Process Engineering - Energy
![Page 1: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/1.jpg)
Virtual Process Engineeringin coarse-grained discrete particle methods
Wei GeState Key Laboratory of Multiphase Complex Systems
Institute of Process Engineering, Chinese Academy of Sciences
2017 NETL Workshop on Multiphase Flow Science
2017.08.08
![Page 2: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/2.jpg)
What is virtual process engineeringVirtual experiment and measurement on digital processes in computers
High fidelity: Indistinguishable to the real processesRealtime/quasi-realtime
High flexibility: Virtual and real systems integrated:experiment-measurement-computation-visualization-control integrated
High efficiency: Within reasonable cost and complexity
New R&D and training/educational platformGe et al. 2011 Chem. Eng. Sci. 66:4426-4458; Liu et al. 2012 Chem. Eng. Proc. 108:28-33 2
![Page 3: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/3.jpg)
Whydiscrete particle methods
![Page 4: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/4.jpg)
4
Challenges of multi-scale simulation
DEM
MD MC / DPD
CGP
molecule assembly particle cluster reactor
unit model
meso-scalesno matured theory
direct simulationtooooo costly
![Page 5: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/5.jpg)
5
W(r)
a
i
Locality
Additivity
Similarity
ia ia ai ai2
i ai i
f mf | = D r Wr ρ
∇ ∑
ia ia ai2
i ai i
f mΔf | = 2D Wr ρ∑
Xu et al. 2015 Chem. Eng. Sci. 121:200-216
Advantages of discrete methods
MD MC PPM, DPD LBM, …DEM SPH PIC MaPPM, CG-DEM, …
![Page 6: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/6.jpg)
Software-hardware co-design
6
Particle-based
algorithms
Graph- or Matrix-Based
algorithms
MulticoreMIMD
(e.g., CPU)
ManycoreMIMD
(e.g., MIC)
ManycoreSIMD
(e.g., GPU)
model software hardwaresystem
Lattice-based
algorithms
Ge et al. 2015 Chem. Eng. & Tech. 38:575-584 structural consistency
![Page 7: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/7.jpg)
7
Multi-scale node architecture
ConfigurationArchitecture Layout
2x Intel E56404C @2.6GHz224.9Gflops
Inter Xeon Phi 3120a 57C @1.1GHz
1000Gflops
5x nVIDIA Tesla c2050448C @1.15GHz
515Gflops
PCI-E8GB/s
QDR IB5GB/s
QDR IB5GB/s
Li et al. 2014 Particuology 15:160-169; Ge et al. 2015 Chem. Eng. & Tech. 38:575-584
![Page 8: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/8.jpg)
Rpeak SP:Rpeak DP:Linpack:Mflops/Watt:memory:storage:network:occupied area:weight:max power:
2.26Petaflops ~ 10-8.5 Mole-flops1.13Petaflops496.5Tflops (19th, Top500, 2010)963.7 (8th, Green500, 2010)17.2TB (RAM), 6.6TB (VRAM)76TB (Nastron)+320TB (HP)Mellanox QDR InfiniBand150m2 (with internal cooling)12.6T (with internal cooling)600kW+200kW (cooling)
8
Greenest petaflops supercomputer in 2010
![Page 9: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/9.jpg)
Cases Nparticles
1 7,220
2 120,159
3 240,318
4 480,636
5 961,274
6 1,922,548
7 3,845,096
Configuration State: compact packing (most colliding neighbors) Box:80×40 ×20 m3
Radius of particles: 12.5 mm Time step: 10-4 s Grid size: 2 times of diameter of particles Neighbor searching cutoff: 2 times of diameter of
particles Average neighboring particles in neighbor list: ∼20
Parameters
GPU performance in DEM simulation
packing stateXu et al. 2016 IPE Internal Report
9
![Page 10: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/10.jpg)
Speed of different modelsNparticle*Nstep/time (108)
Relative speed
GPU performance in DEM simulation
10
![Page 11: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/11.jpg)
storagedistribution
11
Full scale simulation of a blast furnace
Ren et al. 2015 Internal Report
![Page 12: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/12.jpg)
Example IDiscrete particle simulation
of gas-solid flow
![Page 13: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/13.jpg)
13
cellcell
1 dd i
iV ∈
= − ∑F fcoupling
CFD-DEM method
solid phase
fluid phase
discrete element method
c dii ij i i
j
dm mdt
= + +∑v f f g
ii ij
j
dIdt
=∑ω T
Navier-Stokes equation
mass conservation
momentum conservation
( ) ( )gg 0
tερ
ερ∂
+∇ ⋅ =∂
u
( ) ( )( )
gg
g d
tp
ερερ
ε ε ερ
∂+∇ ⋅
∂= − ∇ +∇⋅ + +
uuu
τ g F
![Page 14: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/14.jpg)
More efficient discrete methods
Simplifying interactions
14
Upgradingscales ofdescription
Stochastic DPMMonte Carlo motion, individual particles
Particle in cellcontinuum stress laws, representative particles
Smoothed particle hydrodynamicscontinuum stress laws, swarms of particles
Coarse-grained particlesdiscrete element models, swarms of particles
![Page 15: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/15.jpg)
15* Sakai et al. 2010, Int J Numer Meth Fl. 64:1319-1335
The k3 real particles are represented by k coarse-grained particles and
dc=k3d0
Development of coarse-grained models
𝑚𝑚𝑐𝑐𝑑𝑑𝒗𝒗𝒑𝒑,𝒄𝒄
𝑑𝑑𝑑𝑑 = 𝑭𝑭𝒅𝒅,𝒄𝒄 − 𝑉𝑉𝑐𝑐𝛻𝛻𝑃𝑃 + �𝑭𝑭𝒄𝒄 + 𝑮𝑮𝒄𝒄
= 𝑘𝑘3(𝑭𝑭𝒅𝒅,𝟎𝟎 − 𝑉𝑉0𝛻𝛻𝑃𝑃 + ∑𝑭𝑭𝒄𝒄,𝟎𝟎 + 𝑮𝑮𝟎𝟎)
ln𝑒𝑒𝑐𝑐ln𝑒𝑒0
= 𝑘𝑘3
* The modification of micro-dynamics on particles (Similar particles assembly, SPA)
** In addition, the coefficient of restitution is modified as (MSPA):
** Benyahia et al. 2010, Ind Eng Chem Res. 49:10588-10605
![Page 16: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/16.jpg)
16*Liu et al. 2013, 14th International Conference on Fluidization .
Modification of fluid properties for constant Re and Ar (Principle of similarity model, PSM)
𝜇𝜇𝑔𝑔,𝑐𝑐
𝜇𝜇𝑔𝑔,0= 𝑘𝑘2 𝜌𝜌𝑔𝑔,𝑐𝑐
𝜌𝜌𝑔𝑔,0= 𝑘𝑘
𝑮𝑮𝒄𝒄 = 𝜌𝜌𝑝𝑝,𝑐𝑐16𝜋𝜋𝑑𝑑𝑐𝑐3𝐠𝐠 = 𝑘𝑘3𝜌𝜌𝑝𝑝,0
16𝜋𝜋𝑑𝑑03𝐠𝐠 = 𝑘𝑘3𝑮𝑮𝟎𝟎
𝐅𝐅𝐠𝐠,𝐜𝐜 =βc
1 − εc(𝒖𝒖𝒈𝒈,𝒄𝒄 − 𝒗𝒗𝒑𝒑,𝒄𝒄) − 𝛻𝛻𝛻 𝑉𝑉𝑐𝑐 =
β01 − ε0
(𝒖𝒖𝒈𝒈,𝟎𝟎 − 𝒗𝒗𝒑𝒑,𝟎𝟎) − 𝛻𝛻𝛻 𝑘𝑘3𝑉𝑉0 = k3𝐅𝐅𝐠𝐠,𝟎𝟎
Umfc =µg,c
ρg,cdc𝑓𝑓 Ar =
𝑘𝑘2µg,0
𝑘𝑘2ρg,0d0𝑓𝑓 Ar = Umf0
No change to the particle-particle interactions
Development of coarse-grained models
![Page 17: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/17.jpg)
Coarse-graining with statistical equivalence
Lu et al. 2014 Chem. Eng. Sci. 120:67-87
𝑑𝑑ℎ𝑐𝑐 = 𝑑𝑑𝑐𝑐(1 − 𝜀𝜀𝐶𝐶𝐶𝐶𝐶𝐶)1/3
𝑒𝑒𝑐𝑐 = 1 + 𝑘𝑘 𝑒𝑒02 − 1 (1 − 𝜀𝜀𝐶𝐶𝐶𝐶𝐶𝐶)1/3
k3(1-εCGP) real particles are lumped into k coarse-grained particles with
dc=k3d0
𝑮𝑮𝒄𝒄 = 𝑘𝑘3(1 − 𝜀𝜀𝐶𝐶𝐶𝐶𝐶𝐶)𝑮𝑮𝟎𝟎
𝑭𝑭𝒅𝒅,𝒄𝒄 =𝛽𝛽𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
1 − 𝜀𝜀𝐶𝐶𝐶𝐶𝐶𝐶(𝒖𝒖𝒈𝒈 − 𝒗𝒗𝑪𝑪𝑮𝑮𝑪𝑪)𝑉𝑉𝐶𝐶𝐶𝐶𝐶𝐶
𝑭𝑭𝒈𝒈𝒈𝒈𝒈𝒈,𝒄𝒄 = −( �𝑖𝑖=1
𝑁𝑁𝑛𝑛,𝐶𝐶𝐶𝐶𝐶𝐶 16𝜋𝜋𝑑𝑑𝑐𝑐
3)𝛻𝛻𝑃𝑃 = −(1 − 𝜀𝜀𝐶𝐶𝐶𝐶𝐶𝐶)𝑘𝑘3𝜋𝜋6 𝑑𝑑0
3𝛻𝛻𝑃𝑃
17
![Page 18: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/18.jpg)
2 2r
1 ( 1)( )4
E m e∆ = − k v
2 2c 0 g4 ( )f n d gπ ε= Θ
22 2 2ps,p c sys p p p p r,p 0 g( 1) ( )E E f V m N d e gπ ε∆ = ∆ ⋅ ⋅ = − Θk v
22 2 2s,CGP CGP CGP hc CGP r,CGP CGP 0 g( 1) ( )E m N d e gπ ε∆ = − Θk v
2 1/3CGP p CGP1 ( 1) (1 )e e k ε= + − −
Kinetic energy loss:
Collision frequency:
Original system:
Coarse-grained system:
·
Derivation of the coefficient of restitution
·
·
18
![Page 19: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/19.jpg)
19Lu et al. 2014 Chem. Eng. Sci. 120:67-87
dcl,min > dcgp
dcl,min
CD(ug-us)
Ug Us
dcl
εcgp
CD,cgp=CD,EMMS
EMMS
Structure:
Drag:
εcl,max > εcgpdcgp
Critical parameters from EMMS
![Page 20: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/20.jpg)
Simulation of a bubbling fluidized bed
Zhang et al. 2017. Submitted to Powder Tech. 20
![Page 21: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/21.jpg)
21
Flow distribution
21
Interpolation
“Bi-linear”i ui
Ai+1,j Ai,j
Ai,j+1Ai+1,j+1
ui,j
ui,j+1 ui+1,j+1
ui+1,j
Traditional approach ignores meso-scale structure
![Page 22: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/22.jpg)
22
Structure-based flow distribution
22
i
u┴
u∕∕
ε∇
ui
i
u
u
Xu, Ge & Li 2007 Chem. Eng. Sci. 62:2302
![Page 23: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/23.jpg)
23
Slugging: simulation vs experiment
DPM only DPM+EMMS
Xu, Ge & Li 2007 Chem. Eng. Sci. 62:2302
![Page 24: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/24.jpg)
24
Simulation of lab-scale CFB riser
DPM DPM+EMMS
Experiment:Chem. Eng. Sci.49:2413, 1994
Simulation:Xu et al. 2012 Chem. Eng. J. 207:746–757
Up to 10 million particles100mm ID x 6m in 2D, 20x(CPU+GPU)
Speed at ~0.2s per day using ~0.5µs time step
![Page 25: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/25.jpg)
Ge et al. 2011 Chem. Eng. Sci. 66:4426-58; Xiong et al. 2012 Chem. Eng. Sci. 71:422-430 25
Our results:1 million particles, 1 billion lattices entering the scale-independent domain for the first time.
Previous results: ~100 particles
Scale-independent clustering from DNS
![Page 26: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/26.jpg)
Zhou et al. 2014 Chem. Eng. Sci. 116:9-22
Spatial distribution
Magnitude & direction
26
new old
Drag characteristics revealed by DNS
F(Rep,ϕ,|∇ϕ|,θ)
![Page 27: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/27.jpg)
Flowchart for CFD-DEM coupling
CFD DEMcoupling step
only
27Lu et al. 2016 Chem. Eng. Sci. 155:314-337
![Page 28: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/28.jpg)
Overlapping of computing & communication
particle order & storage main tasks
28
Lu et al. 2016 Chem. Eng. Sci. 155:314-337;Xu et al. 2016 Huagong Xuebao 67:14-16 (In Chinese)
![Page 29: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/29.jpg)
Irregular space decomposition
1
23
4
6 7
8
91011
5
29
Xu et al. 2016 Huagong Xuebao 67:14-16 (In Chinese); Xu et al. 2017 IPE Internal Report
![Page 30: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/30.jpg)
3D simulation of a lab-scale riser in DPM14M CGPs for
~7G 80µm RPs120x(CPU+GPU)
@ ~12s/daywith ∆t=10µs
Lu et al. 2014Chem. Eng. Sci.120:67-87
30
![Page 31: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/31.jpg)
Example IIHard sphere/Pseudo Particle
Modeling of high Kn flows
![Page 32: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/32.jpg)
32
activation center
poresize & structure
intrinsickinetics
coupled models, concurrent computing
particle
chemistry
Interface of chemistry & chem. eng.
interfacial diffusion
hydrodynamic behavior
strength & lifespan
chemical engineering
![Page 33: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/33.jpg)
33
Challenges at the “interface”• Lack of scale separation• Strong scale dependence of properties• Strong non-linear effect (e.g., non-Newtonian, slip…)• Strong non-equilibrium effect (e.g., anisotropic)
Continuum methods are invalidDiscrete methods are expensive
Efficient algorithms & supercomputing
![Page 34: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/34.jpg)
Shen & Ge, Phys. Rev. E, 2010
34
Two categories of MD methods
Model: Soft Sphere (SS) Hard Sphere (HS)
Algorithm time-driven(synchronized)
event-driven(asynchronized)
Pros scalable accurate (machine err.)efficient (dilute)
Cons inefficient (dilute)inaccurate (num. err.)
non-scalable( 𝑁𝑁 log𝑁𝑁 → 𝑁𝑁 )
![Page 35: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/35.jpg)
35
Begin Initialization Construct FEL&CBT
Winner from CBTValid
Traverse
TimeUpdate FEL&CBT
Move to new cell Collision
Finish
Y
N
Y
N
YN
Hard sphere model and algorithm
![Page 36: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/36.jpg)
Miller et al., 2003. Computational Physics 193, 206-316
dash line: Available HS
dash-dot line: Ideal HSparallelized perfectly
Scalability of the event-driven algorithm
36Zhang, Shen & Ge et al. 2016 Molecular Simulation 42: 1171-1182
![Page 37: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/37.jpg)
t t+1 a t+1 b t+2
Pseudo-particle modeling for dilute gas:a combination of SS & HS
Ge & Li 2003 Chem. Eng. Sci. 58: 1565-1585; Chen & Ge 2010 Particuology 8:332-34237
![Page 38: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/38.jpg)
PPM for flow simulation
38
TS waves in duct flow & Flow past a single cylinderLong-time tail ofthe velocity auto-correlation function (VAF)
![Page 39: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/39.jpg)
collsii
l s n=∑B
collsobs
1 1ij iPV Nk T
dim tδ= + ⋅∑r p
( ) ( ) 21 lim 02 i it
D tdim →∞
= −r r
2
m12gW
Uν =
Obtained in equilibrium molecular dynamics simulation
Mean free path
Pressure
Self-diffusion coefficient
Shearing viscosity
Periodic boundaries
1 2 3 ... Nc
Boundaries
W
Hg
Properties of the pseudo-particles
Obtained in non-equilibriummolecular dynamics simulation
39Chen & Ge 2010 Particuology 8:332-342
![Page 40: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/40.jpg)
Hard sheremodel
( )21 21ηχη
−=
− ( )31 21ηχη
−=
−1/ 2
* EE
0
12 8
DDv
πσ ηχ
= =
1/ 2* EE
0
1.018963 16
DDv
πσ ηχ
= =
0.01 0.1 11E-5
1E-4
1E-3
0.01
0.1
1
10
Simulation value Dsim/(σv0) Enskog value DE/(σv0)
Dimen
sionle
ss dif
fusion
coeff
icient
D/(σv
0)
Effective packing fraction ηηt≈0.70 1E-3 0.01 0.1 1
1E-6
1E-5
1E-4
1E-3
0.01
0.1
1
10
ηt≈0.55
Simulation value Dsim/(σv0) Enskog value DE/(σv0)
Dimen
sionle
ss di
ffusio
n coe
fficien
t D/(σ
v 0)
Effective packing graction η
Effective self-diffusion coefficient 2D 3D
41
![Page 41: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/41.jpg)
Hard sphere model ( )2
1 21ηχη
−=
− ( )31 21ηχη
−=
−1/ 2
* 2E
1 1 82 18 2
πν η χηη χ π
= + + + ( )
1/ 2 22*
E5 8 7681
96 3 5 25πν χη χη
χη π = + +
0.0 0.1 0.2 0.3 0.4 0.50
1
2
3
4
5
Simulation value νsim/(σv0) Enskog value νE/(σv0)
Dimen
sionle
ss sh
ear v
iscos
ity ν/(σv
0)
Effective packing fraction η0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
0
1
2
3
4
Simulation value νsim/(σv0) Enskog value νE/(σv0)
Dimen
sionle
ss vis
cosity
ν/(σv
0)
Effective packing fraction η
2D 3D
Effective shearing viscosity
42
![Page 42: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/42.jpg)
PPM HS Buffer
Simulation of dilute gas:Hard sphere + pseudo-particle modeling
43Zhang, Shen & Ge et al. 2016 Molecular Simulation 42: 1171-1182
![Page 43: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/43.jpg)
Miller et al., 2003. Computational Physics 193, 206-316
dash line: Available HS
dash-dot line: Ideal HSparallelized perfectly
dash-dot line: HS-PPMmultiple processes with same scale on each process
Demonstration of scalability
44Zhang, Shen & Ge et al. 2016 Molecular Simulation 42: 1171-1182
![Page 44: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/44.jpg)
46
Simulation of flow past a single sphere
Flow field across the shockwaveComparison of the drag coefficients from simulation with experiments
Ma=3.865Re=104
![Page 45: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/45.jpg)
47
Simulation of flow past a single sphere
Flow distribution at two planes across the shockwave
![Page 46: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/46.jpg)
48
The diffusion between two chambers
The number of particles (NB) in chamber B during the diffusion processZhang, Shen & Ge et al. 2016 Molecular Simulation 42: 1171-1182;
Zhang et al. 2016. Comp. & Appl. Chem. 33(11):1135-1144 (In Chinese)
![Page 47: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/47.jpg)
Effect of coke on gas diffusion in zeolites
ZSM-5 zeolite (MFI) : 8×8×8 (u.c)Gas properties similar to methane at 723K & 1tamGas loading: 4 molecules/u.c.Coke deposition at T12 sites.
49
Li, Zhang, Li, Liu & Ge, Chemical Engineering Journal, 2017, 320: 458-467
Self-diffusion coefficients vs. coke amount for the two coke formation mechanisms
Coke Model I Coke Model II
![Page 48: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/48.jpg)
1027/m3
Flow-diffusion-reaction in a porous particle
Solid material structuresimilar to ZSM-5
Gas propertiessimilar to methaneat 723K & 1tam
50
Zhang et al. 2016 IPE Internal ReportThe number density distribution of products
![Page 49: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/49.jpg)
Virtual Process Engineering
from reactions to reactors
![Page 50: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/50.jpg)
The VPE platform at CAS-IPE
Experiment - Measurement - Computation - Visualization - Control integrated
Processing
Simulation
Control
Visualization
ControllerComputer
instructionoperation
SignalMeasurement
Signal & data
Display
Ge et al. 2011 Chem. Eng. Sci. 66:4426-4458; Liu et al. 2012 Chem. Eng. Proc. 108:28-33 52
![Page 51: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/51.jpg)
53
Offline Online interactive simulation
陈ppt-第24页
real solids: ~ 10 billionCG solids: ~ 0.3 milliondp: 200 microns~ 1/30-1/50 realtime speed@ 0.1ms time step
Ge et al. 2015 Chem. Eng. & Tech. 38:575-584;
Lu et al. 2016 Chem. Eng. Sci. 155:314-337
Xu et al. 2017 Internal Report
![Page 52: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/52.jpg)
Physical system decomposition scheme
Simplifiedtwo-fluid model for
developing period
CG-DPM for steady state
dynamics
coupling
CGratio 3
Particle size distribution
10-4m(1.23 1.85 2.46)
%mass(5 90 5)
Nparticle 81.5 Million
Ncell 94,694
K80 54
CPU 108 cores
dtsolid 2.0e-6
dtgas 4.0e-5
Speed ≈40s/dayVelocity
Discrete simulation of a CFB at lab-scale
54Xu et al. 2017 Internal Report
![Page 53: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/53.jpg)
Very long time simulation at industrial scale
22x(CPU+GPU), 640s/day, 6800s totally5555Lu et al. 2016 Chem. Eng. Sci. in revision; TFM data from Lu et al. 2015 Internal Report
![Page 54: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/54.jpg)
-1.0 -0.5 0.0 0.5 1.00.94
0.95
0.96
0.97
0.98
0.99
1.00
1.01
1.02
Kn=0.185 Kn=0.154 Kn=0.132 Kn=0.116 Kn=0.103 Kn=0.092
T(x)
/Tav
g
x/WShen & Ge, Phys. Rev. E, 2010
Reaction-diffusion-flow coupling
56
Xu et al. 2015 Chem. Eng. Sci. 121:200-216
![Page 55: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/55.jpg)
57
Inflatable reentry vehicle experiment (IRVE)
57
![Page 56: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/56.jpg)
58
The simulation shape of IRVE
Simulations on IRVE
The flow field and
temperature distribution of the simulation
of IRVE
The temperature profiles of two crossing lines
Zhang et al. 2016. Comp. & Appl. Chem. 33(11):1135-1144 (In Chinese)
![Page 57: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/57.jpg)
Full system test on Haihu Light (125PF)
dp>60 nm, length>55mm10.27 million cores (160K CPEs), 8.5-17.6Pflops sustainable performance
Fu et al. 2016 Sci. China: Info. Sci. 59(7):072001The Sunway TaihuLight supercomputer system and applications.
59
![Page 58: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/58.jpg)
solid particle(silicon powder): GPU
gas (silane): CPU
particle scale reactor scalemolecular scale
from Reactions to ReactorsRcp= ratio of
computing time to real time
Rcp~50
Rcp~2000
60
solid particle: CPUgas: GPU
Rcp~1012
![Page 59: Virtual Process Engineering - Energy](https://reader030.vdocuments.us/reader030/viewer/2022012616/619edb695b34ba5256648ca7/html5/thumbnails/59.jpg)
Acknowledgements:NSFC MOF MOST CASSinoPec PetroChina BaoSteel YeTech SynFuelsAlstom BASF BHPb BP EDF GEIntel nVidia Shell Total Unilever
www.emms.cnemms.mpcs.cnwww.multiscalesci.org