14
Dynamic Reduced-order Model for the Air Temperature Field Inside a Data Center G.W. Woodruff School of Mechanical Engineering Georgia Institute of Technology Atlanta, GA 30332-0405 Rajat Ghosh and Yogendra Joshi

Dynamic Reduced-order Model for the Air Temperature Field Inside a Data Center G.W. Woodruff School of Mechanical Engineering Georgia Institute of Technology

Embed Size (px)

Citation preview

Page 1: Dynamic Reduced-order Model for the Air Temperature Field Inside a Data Center G.W. Woodruff School of Mechanical Engineering Georgia Institute of Technology

Dynamic Reduced-order Model for the Air Temperature Field Inside a Data Center

G.W. Woodruff School of Mechanical EngineeringGeorgia Institute of Technology

Atlanta, GA 30332-0405

Rajat Ghosh and Yogendra Joshi

Page 2: Dynamic Reduced-order Model for the Air Temperature Field Inside a Data Center G.W. Woodruff School of Mechanical Engineering Georgia Institute of Technology

2

Project Objective

• Development of experimentally validated reduced order modeling framework for dynamic energy usage optimization of data centers and telecomsDynamic reduced order modeling framework development Experimental validation of dynamic reduced order

modeling framework • Implementation and generalization of modeling

approach in data centers and telecom test sitesAssessment and refinement of approach at a selected

facilityDevelopment of data center thermal design software

Page 3: Dynamic Reduced-order Model for the Air Temperature Field Inside a Data Center G.W. Woodruff School of Mechanical Engineering Georgia Institute of Technology

3

Accomplishments

• Developed a CFD/HT model for predicting transient temperature field

• Developed an experimental setup for measuring transient temperature field

• Utilizing a reduced-order model to generate new temperature data from an existing temperature ensemble obtained from CFD/HT simulations or experiments

Page 4: Dynamic Reduced-order Model for the Air Temperature Field Inside a Data Center G.W. Woodruff School of Mechanical Engineering Georgia Institute of Technology

Modeling Algorithm

4

POD coefficient calculation

.1 ;,

max 2

2

2

T

)().,,(

),,(

1tbzyx

zyxT) T(x,y,z;t

i

nk

ii

Interpolation

Ensemble generation

POD mode calculation

Number of principal components determination

nmT

CFD/HT simulation

),,( zyxi

)(tbi

k

99%

1

1

n

ii

k

ii

Reduced-order temperature computation

Error estimation

n

kiiO

1)(~Error

Page 5: Dynamic Reduced-order Model for the Air Temperature Field Inside a Data Center G.W. Woodruff School of Mechanical Engineering Georgia Institute of Technology

5

Case Study for CFD/HT Simulation

900

1016

609

5082

CR

AC

CR

AC

A1

A2

A3

A4

B4

B3

B2

B1

Cold aisle

Hot aisle

Row

XY

3000

455830

0020

003860

X

Z

Row B

Row A

121

8

Plenum

CR

AC

Adiabatic Symmetry plane

Insulated room wall

•Initial condition- T(x, y, z; t=0)=150C- V(x, y, z; t=0)=0

•Heat load/ rack= 5 KW• Air flow rate from CRAC= 5500 CFM

•Grid Size- 182,000- Adaptive meshingWith hexagonal grid-cells

Page 6: Dynamic Reduced-order Model for the Air Temperature Field Inside a Data Center G.W. Woodruff School of Mechanical Engineering Georgia Institute of Technology

6

Row Inlet at a Known Time (t=30s)

2.5 3 3.5 4 4.5

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

22

24

26

28

30

32

34

36

38

2.5 3 3.5 4 4.5

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

-0.12

-0.1

-0.08

-0.06

-0.04

-0.02

0

0.02

0.04

0.06

X

Z

2.5 3 3.5 4 4.5

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

22

24

26

28

30

32

34

36

38

2.5 3 3.5 4 4.5

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

POD model can reproduce CFD/HT data accurately

2.5 3 3.5 4 4.5

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1.1

2.5 3 3.5 4 4.5

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

18

20

22

24

26

28

30

32

34

36

38

2.5 3 3.5 4 4.5

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

18

20

22

24

26

28

30

32

34

36

38

2.5 3 3.5 4 4.5

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

-0.025

-0.02

-0.015

-0.01

-0.005

0

0.005

0.01

0.015

0.02

0.025

Row A inlet

POD temp. Field CFD temp. field Deviation~1%Velocity field

Row B inlet

Page 7: Dynamic Reduced-order Model for the Air Temperature Field Inside a Data Center G.W. Woodruff School of Mechanical Engineering Georgia Institute of Technology

7

Temperature at an Intermediate Instant (t=15 s)

2.5 3 3.5 4 4.5

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

-0.12

-0.1

-0.08

-0.06

-0.04

-0.02

0

0.02

0.04

0.06

2.5 3 3.5 4 4.5

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

17

18

19

20

21

22

23

24

25

26

2.5 3 3.5 4 4.5

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

17

18

19

20

21

22

23

24

25

26

2.5 3 3.5 4 4.5

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

16

18

20

22

24

26

28

2.5 3 3.5 4 4.5

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

16

18

20

22

24

26

28

2.5 3 3.5 4 4.5

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

-0.025

-0.02

-0.015

-0.01

-0.005

0

0.005

0.01

0.015

0.02

0.025

• POD based model can efficiently generate temperature data at t=15 s from existing CFD/HT temperature ensemble, obviating need for independent

simulation

X

Z

POD temp. field~4 s

CFD temp. field~8 min Deviation~1%

Row A inlet

Row B inlet

Page 8: Dynamic Reduced-order Model for the Air Temperature Field Inside a Data Center G.W. Woodruff School of Mechanical Engineering Georgia Institute of Technology

8

Experimental Validation

12700 CFM CRAC unit

14 kW racks

• Parameters -Eight 14 kW racks arranged symmetrically about cold aisle

-CFM from CRAC unit=12700

•Transient ConditionSudden shutdown of CRAC unit for 2 min

-Observe following transient temperature field at cold aisle for 200 s at 10 s interval

Page 9: Dynamic Reduced-order Model for the Air Temperature Field Inside a Data Center G.W. Woodruff School of Mechanical Engineering Georgia Institute of Technology

9

0 100 200 300 400 500 600

400

600

800

1000

1200

1400

1600

1800

13

14

15

16

17

18

19

20

21

22

0 100 200 300 400 500 600

400

600

800

1000

1200

1400

1600

1800

13

14

15

16

17

18

19

20

21

22

0 100 200 300 400 500 600

400

600

800

1000

1200

1400

1600

1800

14

15

16

17

18

19

20

21

22

0 100 200 300 400 500 600

400

600

800

1000

1200

1400

1600

1800

14

15

16

17

18

19

20

21

22

23

Temperature Measurement at Rack A Inlet

t=0 s t=30 s

t=60 s t=90 s

X

Z

Page 10: Dynamic Reduced-order Model for the Air Temperature Field Inside a Data Center G.W. Woodruff School of Mechanical Engineering Georgia Institute of Technology

10

0 100 200 300 400 500 600

400

600

800

1000

1200

1400

1600

1800

-1

-0.5

0

0.5

1

Validation of POD based Interpolation

0 100 200 300 400 500 600

400

600

800

1000

1200

1400

1600

1800

14

15

16

17

18

19

20

21

22

0 100 200 300 400 500 600

400

600

800

1000

1200

1400

1600

1800

13

14

15

16

17

18

19

20

21

22

POD temperature field ~4s Experimental temperature field ~ 3 min

Error between POD and Experimental temperature

field~1%

•Temperature data at t=45 s are not included in original temperature ensemble generated by experiments•POD based model can generate temperature data at t=45 s from existing temperature ensemble generated by experiments, , obviating need for independent experiment•POD based model is significantly faster than experiments without compromising accuracy

X

Z

Page 11: Dynamic Reduced-order Model for the Air Temperature Field Inside a Data Center G.W. Woodruff School of Mechanical Engineering Georgia Institute of Technology

0 100 200 300 400 500 600

400

600

800

1000

1200

1400

1600

1800

14

15

16

17

18

19

20

21

22

Validation of POD-based Extrapolation

0 100 200 300 400 500 600

400

600

800

1000

1200

1400

1600

1800

13

14

15

16

17

18

19

20

21

22

X

Z

0 100 200 300 400 500 600

400

600

800

1000

1200

1400

1600

1800

-1.5

-1

-0.5

0

0.5

1

POD temperature field ~4s Experimental temperature field ~ 6 min

Error between POD and Experimental temperature

field~1%

•Temperature data at t=205 s are outside the temperature range t=0-200 s•POD based model can generate temperature data at t=205 s from existing experimental observations, obviating need for independent experiment•POD based model is significantly faster than experiments without compromising accuracy

Page 12: Dynamic Reduced-order Model for the Air Temperature Field Inside a Data Center G.W. Woodruff School of Mechanical Engineering Georgia Institute of Technology

12

Publication/ Presentation

• Conference Proceedings

Ghosh, R., and Joshi, Y., 2011,”Dynamic Reduced Order Thermal Modeling of Data Center Air Temperature”, ASME InterPack 2011 Conference

• Poster Presentation

Ghosh, R., and Joshi, Y., 2010 " Dynamic Reduced Order Modeling of Convective transports in Data Centers" at NSF I/UCRC meeting

Page 13: Dynamic Reduced-order Model for the Air Temperature Field Inside a Data Center G.W. Woodruff School of Mechanical Engineering Georgia Institute of Technology

13

Plan for Next Quarter

• Refining POD based model – Designing more representative experiments with

distributed temperature measuring facility• Capable of measuring instantaneous room level

temperature field

• Developing thermal design software for data centers

Page 14: Dynamic Reduced-order Model for the Air Temperature Field Inside a Data Center G.W. Woodruff School of Mechanical Engineering Georgia Institute of Technology

14

Acknowledgement

We acknowledge support for this work from IBM Corporation as a sub-contract on

Department of Energy funds