Professional Documents
Culture Documents
01234567 8392
4
A3B/57.C90D349
(<*734541426'(52727
E2
766
6462
12735345267813
6
2534957
613466!24273537
5
"2
2#72
$134672342!%
&
342!536'61
3267
37
5
"24234
31
342!3%
(65722
27245!331
35723223
35)246
2
127353452%
*17
2347
4
43!22
57 1!7353552
6!65357 54
1%
813 *72274(
22424F657
( 2
+,-./,01.23-/.4567.7,8,97:/
;657<232
=424639
123456732457!147
"72
6169%
;657'6'22 G6 57
012456789
41
212128
28
1""C!0
!!""#!
$5%12&
886
'78622
2'
(85)(4(589
'7
11*6+8,
9
%
'
-! ./!0
12"3456 7839.:
15"5 !;5
< "!=# > ?@!"# !#
=215423223 <657622 K657
01234567 87323) *55"12345673 =242
1!545673
01234567 8392
46
B0C:D/0679E4/,068
8257F7523546G2514
67322
1421775745266
254!"13#2$554%
6
257663542!2634227
672425432341
3677242
46
61754324!7
23423&24273567642!6436
572457
66571357!"13#2$554465
42412'5742324!(421742
6
2544923617
3466242()4644526342(*62722
61
2+++
8!"13 H67242 *422 ?275G2566?4253 I57542J227487353
*2
,-./0-12/34.0/5678/8-9-:8;0
<657=232>424639"123456732457147
7261
69(
<657622 K657
8 87323@)A
=2M?
752N
46
7F7523546*52727
O2766
54579542
5457227457342417
223217757452+7
42!244253
2422335M233422314
6237427'!
2246761414+17
3267662576
246P7
52$2P2611632+QA
66
19
H542
8
61375>6
46
8257F7523546G2514
4225345235746572545312
444245235746532"1467
42$5722743532"146!14576431254453412(4135612
7
6127436312345674622'232327
542(479616612&A
H542 =2627
4567
8
61375>6 R46
8257F7523546G2514
012346789
32
3
03
1
3
3
2
01
1
01
702
2
23
71
73673
6
070
30231
307302
3
3
30
0
070
73!"0
#$12
070
60186780
73
07
2
01
1
301
#201
01
070
30
0%3
0
01206782
23
#
023
3
6781
#0123467822
3&'
(
0
3
1
2 )201+7,)-$.
/
40103
06
67823
3
2
22
3
(
0
0
#75
$'3
2
372
320
01
20
2
20
3
#
3
30
61
301
72
320
3
2
07
4
3067
2782
!020
#71
,01
20
202
320
2820
301
820
732
20
5
261
11601
2
207
2##
07
3
2
01
72
320
81
230
01
20
2
20
3
3
0
802
320
3
0
06
1
37
,
73
78780
7323
80
)'322
3720
#
3
3020
2
90
3
26
21
072326
27
3
0320820
#
7
20
#
3
36
1
373
91
231
27#
1
0178122
3
:';22
80
73
%1278
2#22
2##
20
73232383
317013;23
312
!
7601
2
7
(
0
)<
7
320
73
"
6
=22>21
:-01+7,)-$.
13
223?720
73203
06723
@78230706#
72
3&0
2
32322
3
27,
1
12
#70230
073830
180
6788
62
30,01
80
370
6
21
(
0
$<
7
320
73
%121783
/17 $0
,)-$.
%
2303
0679
80
7426
217
2178022
3
31
28
221
33
321278%3
012346782&'
(
0
$<
7
320
73
"236@23
48 $.01
,)-$A
1
34
#7
7#
0
#2320
737B212%112
23
100#&CC
37
8
737C17C*!0
#!07!1
#!21278!83!20
(
0
"
6
=22>21
$.01
,)-$A
13
223?720
73203
06723
8
2
30
2322
3
(
0
01234567 8392
34
2574262432
55763342 3
JF<@K;LMDF<@NOFK9;PEQF<FR@
82
ST75722571 $!
!6"7#322
135 14567"548$%13&
!5
7961'2 157
'7266127
45 2(
#'24"6%12345673)
#"742
46976"6"465723242 2 66439*+,8+53-$*142'27#324
42565446557439 7262%1234 62 2 6"275"7446344426$
42642
6174
6237646 6247.$*!6"7#57232425 54&/
$24535 5453"4426$ 0*722
37
4225376722
46652*
/7426427
#'24"66233632.*.1!2*14461# 1775776"
4"635 1456733423267
35 1456713231$6,8+4675653132
*#3422
7"4657232453&
#"742
46976"6" 14222 27442243421745 2*7620 26
6
2# 13574242
22 27437
54322 346492672*
7961
8$%13 757542T2 274+6
257
U2
89:;<9=>;?@:<;ABCD;D9E9FDG<
H657,2321424639%123456732457147
'7261
"69*
H657622 657
165
5(23(6)16*13+5*5913
23
,)16-5+24.
$9+1
+14
)*56
3/94*456"8&5'3
54
44
1+1623
"*13
4
84489
6
56956
9
96
!6539
"
(36+5*&
4+
541&+5-.9
596
9191"1"6.(
061&5&4**
.5359
5444+
9967486'36++16*4454
1&269
95674.
(194*9:
3
.$
659*5*1965
;(
861
354<'3516261
16
;$9
5
965
93+&61)261
16
8&5'3
3
23
,=1656
6'36.
(#16-5+24)165+14$5+3
9"656544+9
59
+
15419"6.(
29
19591)5434519*135619"69+24
+5*&31
5)6191669"89-24-+5*&31
59"
+5446+9
19
59
639"+9+3++96+9.
0 21++9519
>5?&65+@5$
)559
8*6?9"969"A+&B
59*136*+36.!6.C
$574595
65&136DA&EF15
65
19$9
29
193#
.
$54
15522459"
99.749$591
3&+.41&5
G
.....
HIHJKLMNMOPQMRO
STUVNW
+216664
1
C
....
45
49$231
16$549
.$12
1665*.89*5*5
$
+919$54
16'3
+16++16*9
3&+
16
.
5169"1
49
2G<<.7-&*199.1+<=<<661966
65/9*66"646261
65
6
+<X.Y?8-Z&$6%
B159$911+59*16
12+5413
;
6"56
0
56956 9
9
96
!6539
"
8#8$%5&5'3
139
541"54261
165
5(23(6)16*13+5*5913
23
,)16-5+24.
013456347895
6
8858169985598589954348584
9855638
6883489815864815
68 !84338176415
'
7(
3)7805
1 ""6
#8$%"&
87*(11881+3,
6
1483
0
8$-.949485545
818049738954685/"&
384381485
06398381615/.
086
888168458
5886
83898586
8$7613856
8
5888168176
846
856
8994687!1
0741461465866864141853964149989348539641523
348558545+
68
7888)18654 $1741$%"6
!551783478185
9
89989)6415
8978817816418438686
8418*738154198-96
71349836854134566416
:-;841546"%5865198949
48554
481746$:%586&:949485545
58854<86="#4983>
84641134794766
8538638$#17?#4983518
5855899989)87
68 $!84338176415
177
@90593 " 6
1$%".
098180156668498
1494
,'
7(
3)7805
1
598589549865144
841714633586616458876
6
8
4361484
8
68
7985
7
856
"$6
49$%".
05644
5156
"484192
"484:48
418744585617
2
68
,553914
3387
9*B69*5698
19C1856 &6
A4$%".
01124567891
67676686678
7698158578
627
986
!
2"9716#6
$5
92 8
%6&
97&6'
9(6798)8#5
7&69
'79
8686
66
879
&1
76456
8
9
&67686
561*6
66
876
&697
7
6886815766+6
66
87,*&157698979
6(98
69*86616897811&1
6+
229
8186769*156
66
87645289&86
9897
17597688988-67
1
6
896*115795
891
.
126766
66
8/015&
576777&
9
81766259*
156&1
25&89
795
891
1
45716+
9&98
718579
1895
561*
'786&15
9&891
1(6629
&6767969*155
561*'7
%
-15
986
317515259 28!1(
06
8'
9(6798
46
1567279
7897112&89&681
719(686715&61*869
*1891
8815(61(9262786
67
6886788395
61
6
676*1*666576*5
*16(61
6
315&61*86112&89&6788879119
*1795
61
&11
11889788186
845785
8
*786
986 26&16
2891
7
:392 8$
%6+7;$'
9(6798
49$89
#6-6
8157(626&676286
561*6
66
879
126
625&9
8679<61*
616886795
891
64596716618
869661689816
561*6
66
87
6
66
879
18126
76
9
6
4(615&16&1778971
6=
986
$89
#6-6 8$
%6&
97&6'
9(6798)8#5
7&69
!1
2789
-88979875979
015>
27169
*1891
1
615769
012456789
41
212128
28
0""D!/
!!""#!
$71%&%785
%
'1926'85(0)*(51+6,%121,
'691%
52
-! .!/
01"234562727
5!8!
9!!!:;0<!/4 "=>< ?@A!":>
>" !@B C"C4##!9D!:"#"/4!<>EFG
/ "<C!"/4!H!#!:>#I1 "=6C6C6
C64"/#@J@9K!"!4!/4!<4/!:!"
:!:>C!""</!4">>!# /:44<!L>
!:CEFG@
m/
M5N12OPQ%7
85
2RS*,TQN8
U8
%5%
521%1%5%195(V
2%81%
52
5
2%
259
WN52%
266'X898'82%
-! 2.!/
01"2Y40C627?A
Z4"[D"@
[;D!
94; "!0</4:">!@84!9!4 "60<\/4
4]
^_\#;"!>`!:!C !! "I# !!J!@^
[ /9#!L\4a9C !! 4bcd_e!@
B4!1
m/
W
01234567809
7 2398188
!"#$!
59'5
9030
07519929
91./
"49594451929
91472
8949
920799229929
91451022989051059
"1
52#209441."/04
9929
52991
9245719007575492901
57
54744515494919
994
022497089051540(9199010+50220
54
0259
9920
944
9
0519229307903890554549149
919
9451409207992294
4
012345789
41255
545
124547
54154512
JKLMNOPMLQKNRKSTNUKLNKVWXOYWKYNZ[OL\L]NR\^KL]ZWK_`
abcdefgehijkliefmielgnokfbfpgqbrefpniepqebestuevpnkrbfpgqewelbiqjxgirdlm3?3
+.H&")&"D.I!7!.+
y+?A91?22,
19
,3#19,?,9234
21-#2A
219,94#3191,
3
9FA#9
49#99
43
21F913#42
3y
8'08H
z{{3?3
|'('+/'8&})&"D.I&&70+
I
?2
21?21234
43A91?39
#
8'08H
2I.8I./8@4.I~@8|08'22(@+
p
01234567809
7
2398188
!""#$"# %&% %'#(%$"%)*#(%+
178 7
7865
6 68 6 61
58 6 5286
$%&'()*+,*-%.)()/0*1)23*4)/2%0%)2
56*7%&%4*894:*;9/20(<%/0,
= 77 5 67
3
6 0123456178 612 16736 5
618 6
03460 8 736 7 6 0123456178 10
>818
11 710 18 6 274 0
11 7 >8 0
61780 7 6 274
?31
58 >8 6
12 67 5
11 7 67
3
6 0123456178 612
5
1
86 50 67 >8
1
!""62852
152
"#$%&"$&"%'"294
628(2228
((942(
49648
")
*12998514
9
29+26
228724916
2
292
48
,2-
268
212.
/04
666816
22
234567854936:3;<5=43=>?345?3@=A93463>?B7C?345?
C6DE74=4F6:=G34FD?3F:3=3HIJ3KFD7G=4F6:LM
NOPQRSTPUVWSXYORPRZ[
6
!
\+1*,%]^#$%_\0%$!%&+
`\abcd,e`f9
122646
484866
176422
\
49648
48\g\hie
6268546
9484119
6964842416
2642514
215
4812
4964896486425455226
26168912
1+,j|
kPZl[mnZYOkR
6
!
\+1*,%]^#$%_\0%%!o%+
f4d
61pq651641225\1272
41999228526
1+,j|
XURWrWsStRPmVPUnVsu
6
!
viao^#$%_\0]!$w+
`68*16
225264
8x94
628228
h942e49648
yz*68259212
22288972
"2592bd6
2
1+,j|
e0\10\g,dq\0{d1b+1ee`d
|
123456789
57
2765448
6
7622862842678!48"#$"%&'()%*)"%+
9283678)
;<==> ?@AB CDEFB
+6746,7
8262)7464743765$48828
6-7$768637827.287.8///263$578
6
874.745526/062345278228
578
6874
767616,6874.786)27.,83726
4.2742
822,8564,28
6,,4,7664556238
257
37552,4736268616268,6.6786
6567234527866/
&/9787463766563688
86,6
8676328,7,782428
7648828
6626
843676
6676673$.2.2,6386
667
66735628
73768.2,6578
2345278/
2345278.25665636876366.25546
6/
973362,7563$7855,8623456/02
74564,62345278236.256,7.7/:26
82452278727828422725678245266
,73566,784,$.256234528
7855/38
123362,7,,8646$2.25564,62345278
2366868376$.256255557.28
82452278766826
,784,/
38736,66
67368625728
66
123362,$4464685728
6287/861356
01345675118709
30
10653075
50815
0
45345307513
134578184
151345
0713
01356
5108
35810
01537055531
7345
0153705
5
5557
35
45
13
350345
0153705
56053345
5103401
667
41
51108
30305
!"187534517
34
3
0310
871135#1$%&
'
0
57551
75555151650
675
300
34517
34
3
345
101
57
65355(56305745
83
57555
3(556
17
34057015130
35450
3
45(
050345
3
05857)&"&*$+*&,&-"
.*)/45%
081551755734
345
83
$%&
0
50738
5705815101(16
501815
1
030
57/70303
75
0
701(75080751
030
30534817580657
7
5
075
10
108
30305
#3
10555533
78087108
3016550
%
081."6003
%56003
703401030
13
54530507553
81581334575
7551
58435187513
003/45
1
57305075530181575075531
755553
108
353451
5
83
33
305
345108
303
(51
57
4513
53050755301
15555351031095
130
5114513
530507553017585034
1
57
5103
1095
15
1034
757130
511
%
081."6003345
0
13
53050755301
8
356575553451
513
85
5553101
345815
7345
1014015
134
30
5553183
54
5
13
530507553
253
345
135553
4
1
13
530507553
254
30507553
254
0581545108
3003
(5255305157853
34011055553034
1
5713
5305075536
75
815
340103010673
33(40455531
75
003034530507553
0
3450
6110545
#570
514#3034551485
40404355531
034
13
53050755351134
1650
05
85401
456153570045754
51
4
51306
3/4
3
012045246782526049
614712
252609202854629014949
9
8400260012594
2
6494
2850124
690479
1
6460125291901
0012942012922
25260946782
929
4
8272945
046045229274
401282
82
2
2
252609401
57195
2890
2045246782526001
6012
8290
9297
468
9490
047
6
92946
9!"
4740
#6!"
4740
6
929945
04604527
628272401200
045202945
02498272801290
2045246782526049
46782
92$8
9490
047
6
92901200
0452945
02
7
69520452928272
46012
9028%1497
6
6026246246402
&126012
49
420
90
6
5472270990
80
46
8
2
60129
0464971
62
'1269468
0222626082804290129291
92
97
2#6971
7
92405
22
9428046782
9201290
2
0452467825260
%1290
204524678252607
6246782
9246782
946012
5
99%1497
62625
6
927446
82826940
97
6
9
05
047
46782
9226940046782
92012
90
20452467825260
65
(2012945
04686
9028%14949
7
25
9997
46#049
4
24601290224088!"
4740
9029628012)5
9997
46)0
*27
925
9949
2
462804
87294
25442+
9126012
49
42
90287
8291
20
(2601
0462804
227096)046
2672
0129
046,012(462047262891
295
-92
8
2
4
046
.62012590227042
90827286646045249092
8
2
8729946#6
28270
8
2
446945
046012
945
0460452494420126528782992#6
8
7047201494962280127
9227
92400
(299522800
4420128
25
6490840240284282607829*046
9527
9297
620287
92
/
8
2
4
0467
62940712685012)!400)4
"
46012)/
8
2
4
046)0
'1268(4685012755
6
462
012046&791240120126528782902
92
#049447
046045994
202028546212
945
046
4
8
2
42322801282
82952
708901
0
82(66
0
45400129944
40429
8
2
4
046971
9&
45402
56028229822545.$6%122812
8
012133435617589
5
24317543
542
49
74
942
8
5
13342
4
14
449
449
19
41034
424
29959
89
824
824
1
17133
3842
749
29959
89
34
824
1
234
8
7
7424
83
4
17
3417
042
824
!
89717"89721597
598359
1
3124
47589
8
74
843
#
13
8
74
843
5
5934
59
1
5934
89721597
717
19$7
4
554
842
824
749
74
843
533
987
134
0280423
49
59
824
719
%
824
&9959
89
375034
824
24'524
824
78(49
)8
19
8
944
19
4
133174
59
82
94
78(49
1331782
78(49
124
9444
82
29959
89
1
5934
824
*0
78
+
824
41
15758913
824
24'524
19
15758913
78(49
82
1
042473
1359
5317589
75
419
717
859
28
78
,
78(49
74
5317589
754
19
4
134
82
1'
74
4
8
042724159
5
987
248494
414
57
84
987
044
0
74
5317589
534
824
78(49
124
4
-
14
89
74
80742
74
1513
942
8
028482
717
5
24189134
5
74
942
8
0
513
824
987
74
942
8
52713
824
)
042724159
19
4
7294
8
28
74
.#/
#
498
78(49
19
80742
24824
124
024497
57
19
1(4
494
78
29
74
527
0127
8
19
1913
5
89
542497
942
8
824
78
44
8
433
57
134
19
4742594
19
8075
942
8
824
,
01
177497589
78
89721597
149
5787
355759
012133435617589
244
08597
3
89721597
19
14
1
941754
447
89
5317589
754
59
1'"103557
82
89721597
1
-#1
4*05#67
804
19
46614-&/
19
503557
100281
5
4
59
1'"103557
-5
19
24173
244
29959
044
4045133
5
19
984
124
59834
4$4
449
19
41034
8
1
5317589
4859
824
719
9
754
1742
24859
19
804
19
1959
57
5978
89717
57
1
255
214:
;
&43123
247127
82
74
#
14
98754
717
89
25983
82(717589
29959
754
19
4
244
49
24712759
74
74
59
47449
29959
5
1913
4
#$4
449
1
29959
754
8
187
<
19
8
187
%
82
82
74
14
02834
29959
89
133
9,
824
59
87
14
-8
74
247127
1
987
4
74
893
14
57
84
44
78
430
#9
198742
14
4
219
19
1913
5
59
187
,
012345679564
945456
645079
2456
05
2
616
12
2
452144
9320123456324
61095661574595
56655045
2554
2610
9564
9412456
64500
95
1295
65932212655
2409551454412
39
36932
6459
42539 651
1646
4
!
16"
36745267#$73956
%&'()*(+(,&
-./0.10107-"02"-13$
4155
964155
969164125
269
3936536462416614563
3
5
65
91
7
2563951
36163612
1943
0509
9
6103
412924
764595
2
651
2139
412717634
6541943
0509
958
057154500
29
654412345
45
61
0509
$
366103
412646
4599364:129505246
5
3651263115240509
7
244217
2
6545950
36121943
0509
2612
25
#5
4;<=>?@>A>B<
CD+&EF&'G)HII&'J)KL&MG+
-./0.101071"0"113$
41$12
7
N1943
0509
16365
340
41
7145
61
0509
162O452316610
05
264
4156
95
914452416569
419549
163659416595
03644595955523166
5124569
4
19549
94904167
0121030
0324#$1625557416
16152
6O01210300509
953195O1245
4157
012356789
99
8912
09898
8
801212
11832
1308
9339
3898
5
01!"#$%&&"'$()*#
+,--./+0,1
234534646278679:;7
<
1
=
31
8
01
>
?1939
2
98999
91982
81191
8
8
1982
818@1123
3>
@229@19966A=99
130891
12
2=
4B3
3C283130
88
8930
8
?29420DDD=
38
19119
8E29831F6A11301:6A
21@
12
012
9
12
G
5
01HI##J$HIK
L0MNOPNQ-/RSSNQT/UVNW-M
234:34646947F676F;7
<8X
=@Y81Z
898
1
0
89
88
81381;3
9
@1@93@
8
18889929G=012291E
11E
8991318138
9@1@
290[8
13206C291388123
9
1\8
1
01231
9339
3898
+N,/ON-
F3]346469479^749;7
E012302@
130123
2
199
5
01HI$#
_,/̀a,MP0,
0123124245372879
5
325 !!
"#$#%
&'()*+,-*./0*
1234564789:;;47<9=>4?83
01941242452743708
#&@5
ABCC##DE'
(' !#$5! ('$
(FG'A'AH5AAA5AD
CC( $#'A$A
'#!$( G'A#C#
$D
I#C#5
"#
&'(J0.KL/KMN,OPPKMQ,RSKTN.
UV2WXXWY9ZV>58[4W7
918124257\27\]^
"#5
E_$#( #$'D
B'#!C5(!` A@!_
AC_(0 A'
!` $5!C0A' A#D
" ( '! %
&'(ab0*cc*d,ebSLNfK*M
1234564789:;;47<9=>4?83
91]124257079
#5
B#ACg'#!B_
5! '#'D
I#C#5
"#
012346789
!"#"#!$#%#&%#'(
)*11+67,*7--./6)06.11)*)121040611
3142)3111*60)*3/267206751201+$)-)330)123/61
31)*67371+77*7316151/1
01234678999:;
<=>?@>ABB>CDE>F=
!!#"#!$!%"G%H"I(
J*72302227362/K1L.6620%---M0)12317*M771 0)123)/N
0222736O4727/*6)*+/3)*,676.1./3+-/313142)3111*60M
012346789
PQRB
H!"#"#!$G%H&%H&I(
S.1*)2011/0007/3)*K673TU23)7)60)123/6)7*M$-./6)
./V1677.17,6777*67316.131023/60W
0123467X9;YZ99::;
<=>?@>ABB>CDE>F=
H!!#"#!$#%H[%HHI(
\)]/41+$
^.)0+121*+07*-.16.13472/3120)*KTU23)7)6673/*
)*.131*634+4*/1)7/*/340)0$7367307116.)*K6./6)0
)*61*+1+675106/6)7M_*6.16)3067/01$6.17./*K1)*1/00
0.723+51V13401/3367*76)*6321*716.151./V)73M_*6.1
0177*+7/01$6.1/*/340)00.723+06)3351./V142/0)N06/6)7
/66131/0007/3)*K$076.1,)*16)71*13K40.723+5101/33M
`10631K/3+0$
a.3)06)*1
012346789
bcdRA
[#e#"#!$!%!%"I(
012235678
9
53737
12265
173216
3123761
3123356175313736635
13
312
153621663
131253621
637
6
655165315515576
153
6537
5
17661211033362
7619
31
123!"#$%&
'()*+,*-./011*-2/34*5.)
678978:8;;8<;=<>:?@
@31226
1536257667
17
137215
32
712711377613116373
366353711766735656332
1
3571531515367122
3736635
6
153623211276665
33
75
13126656576537
A11
6523<
5<77635621337237551B3126B
6
B5362BB5771B61763B6BC5B7B;37
371631D21
123EFGHIJHKL#&MMHKN#OPHQLG
R.5S/TS-U.+,S-*
V7;W78:8;6<8;<W=?@
06
9
53737
12235
7631D11
35763711312
@7315537
5WV=X37
1?
331
6226538::X63756
13223617737<
YZC56213361376115Y
113
13776136D66
1
3223631<
51556351[215\
1712@137]66^W>:::::::_
03117562216
15117737
`3232151
1216
7321
1234567
89
!"#$$%"&'(!
)*+,*-.-+7+-/-./.+01
2374546
73889533
9954:36;
34
236
<=987
3456;>>96
23
;?4
3?
;<@=3A;66;
<
BC96
<9>3<567D2<36
349
89
EFGHIIJKLG
M!%N!$O "P!O
Q*R*-.-+7S/R)/.+01
23D2<36
3497
12345;T<;<;69T;29
345?883
894
B423U9
3A;96
34B
4
<395
638;3
9
:53663833<83
9<3361+3451-7
19
36934T<3?
346
3<:9534>79
2983
9<33=96
99
345@;8;43;8B34
29638;3
34<96;
;
;
29
<96;
T
29
98
9<3
;<9:3=9+++Q595959><996
D9?3;6B2:9U9<7
2989
34>
98
9<3
;<9T@363=;
,..59>BD9?3;6345T<3?
346
3<:9534>36
29:9534>
T63833<34553663833<83
9<3363
635V6
3
962:
?34
23683
?2:3
2
29<933
36@?344
6
33=U9
,..59><B?9?3;6W
2345;
89
JXIGFYF
!"#$$%"&'(!
Q*R*-.-+7S/Z,/R,01
[38;3
3485963<9436>536
2934
;
:9
;
34
298B4T:954
;
3489
34>=923U3;<34
298597
294
29
98
9<3
;<9:3599
<3634>B
4T
29<96;
5964
83
?2<933
7
294
3?349<
8<9366;8
3463<94
359A;3
97<
29T343
999894
89
25364
3
3953
<
<33
99B
89
EFGHIIJKLG
\]^O$O"_$`]N%O
Q*S*-.-+7)/.S/.+01
01345167189
18
0195
5918595974959748
61897995759
1699876798
97995759
16998799987669
899769 !"796711794#158
$575989#95587796%
&'67$18164
749617188(781$74161685$569
6756974181616618)
65851718749691896*
+6,---18695,$56,&.
6785/59,6#95187
0
+6,---18695,$56,&.
!874957418(8765949$$971#916679
697718414166$6*
127989,279
&89,3
9-75718,3
18,&''''''86,3
139"995759
169987599,889
6796779
'''.&&9
&''''.
12718"9481+979,245!0!"46
13875659697
138756597956,7191859987718
%'''%'''%'''%'''%'''%'''
5917984985#19
7!
097789:;<;=><?8@AB;CD
EFGHIJHKLMNOOHKPMQRHSLG
TUVU.'.&&*.W*XX5
"48$5749198768(7741886
8741869$874161747699187499"495959
869674779896911$74959598
999878U5749591641488
1895786
998874918$749 !"9546749591669
711718774959
097YBZD[:DC\=]<<DCA=^_D̀\Z
001
2345789
!"#$%"#&#&
'(") *+,-"-.,/0.-1
2"3(44"-"(0".'(-"56-."'63".6"!/,".'"7.!*".("86!"*0*49-(-"(0") *+,-
-68.3*!1":8"3"3*0.".6"5!86!"-(,4*.(60"(0") *+,-"-68.3*!"(0"/90*(
754((."6/"6!"/90*(".5!*.,!"/(-54*0."754((."6/"(."*0
-6.(-".*;"*"460<".(".6"654.".'"-(,4*.(60"*0*49-(-"!,01"='(-
-6.(-"/50/-"60".'"5!6--6!",-/"(0".'"65,.!1"=!9".'
864463(0<".'6/-".6"!/,".'"7.!*".(".',-".*;01
>?"='"!(<(/"5*!."*0"",-/"(0".'"*0*49-(-1
#?"@*.!(*4"5!65!.(-",0(.-",-.""60-(-.0.1
A?"B!9"-'6!."/,!*.(60"*0"",-/"(0".'"-.51
$?"=!9"C7/".("D,-!0/C0/".("(0!0.?"(0"-.5"(0!0.*.(601
!"!
#$"!$73353
783
%
&'1"!$#$"43!3(3#$34)!
#*3#$(3+3
51
43,3%
-.!$73351435"3(3!/1$1004#"#1$#$#$34!"#1$%
1240!"3140!"3"1$!""!$7335333$#)3
833"3
0143$34!
"1$!"%
423
3"(3"1443"/3(8
3%
5673(343,#435$/73410/3(3
3/3$%
5573/
#
3
41"314#$
!4!
3
#*!#1$%
9872
0123435789
134
3
1473
01213343464789
1863
7
9
!"#$%&'( <=>?@=ABCDEFGH
)$ *+ ,$$-#.*"+* $/ ;".+*
)$ *+ ,$$- .*"+)$ ) & 0 * IJ.$
* $/
')$ ,$$$$-)' ) +-*"+'$$$ 3;-K)$
$)+- ,$$/
$0"+$-)**,*'$$.-)1"$)'0 L .M"'
-)1"$))*)*'$ *'* / )
$0"+"-$-)**,*'$ )*'$$$.
231"$))*)*'$ '*'* 4
5-+6.'$)$$$)(
307.''$$.
8*10'9
:'10'9
2$.''+*
;'
NODAPG@HQ!"#$%&'!"#$%L!&
307.''$$.
9 !"#$%&'$*'* )66,$()*0 )'045-)*10*-)"7$
-*)''.)*$ )$$. $-)*'$$45-)*10*-)$-)+45-
'10*-)'9$610'$-$' $"+*$+-*' $45-* R)R
) '' $ +-*M,+-610'9)$4
9 "07)",$'+,.-*"+'$$$"$)4
STEP@BUF>=BVH?W=Q
& -+6.' -** )(
?X?YPHZDX[\]^_`abcGEPH[`
I*'R-+6.'6 -M"d"*e 6'$$$(
nonpqrstuovownxs
yzqrv{
fX?YPHghfiBVH?W=Q
|"*)(M")(N?O?>>=>F}?@FDT(.. VH=B~P>@FE>=BEODG=HHDOHR )$'+,-*"+'$$$R`
8*10'9
5-)*10*-)$'$-*)*"+'.)*$45-$)*$+)$'
)*$)$.$--*+*-)*$$$)6-- ,$$45-)*$)$")0 ,* .
-0"'$$$45- ,$$$- ))' ) ,-)*4j60R+* *$"
'$$)"6 -)*$-* "$-)*$$-** " )$4k-23 )-)1
"$)'9 *)$$'')6--)*10*-)4
8.9 R-)*10*-))0)$-*)$--$.)*$7''*,-$*
* +*' ' $45-)" $)+)$-+-*' ' $+)*$
-*$' $0'$$-+)*$-$' $0'$$4I$$-".$. + )
*" R-"$-)" .$)#l$*" m"$*" )0'$0
*l), **" mR-), *)" .- #*,"'')l$!"#$%J )) )
!"#$%&' m48, *)" .$"$) 01)*'$ (-$$$*"+)*$
-$*'+-*"+'$$$48.- R!"#$%&'6.,*$-
*' ' $ ))$"-)*$0-'$$$$$**9-)*" 45-
+6.+ ,$$'')6-), *)" .(
J0$",.lJ0$",.m
L1$* l!"1$* m
)+)+)$$.$$+l)+)I)$m
5-++ ,+-), *)" .$-*)' )$ -)*" - -'"*R -).
+0)*'$ R ) -++ ,+--)642$*" )'"*$6$'*'+*
*'0* 6- -*"+)*$$6+*$-*"+'$$$45-++ ,*,"
$. + ,)) $,$*$6-$6 R$-$K."&-$$4k$$$")6-
$))R$-$6-*)+$R $-)1**,$,$*4
!'' $*$7," ++*), *)" .'"*$6-$ .,*1)' ) )%
$',0,.*' )4&*'$*)$ .".$R6- 1*'0,$-.-,
9) )*)' ) / )'):. . 1& '"*$R6- $00,+6$-*
$*0$-.-*',$'4
01232457484982
25
7227528
927893747487424
252847
02
25
7247328 !
"22#
5289374432
$4%527471&
!
27752
5752'()*+ ,-./012748
2125282
15
'()*+ ,-./3452
12
722752822"112942
5752748942
2125282
15
12
92792
924734%94624594
5972485292
728212528
9:;<=>?@AB>CDEFBG
045252
9119"4%459452933748142H
EIEJ<DKLIMNOPQ#RSTU;<DM#;EVEAABAMWLXE@:DMSWY:EX@UZALEWZIEAE:U@:F
79287312!52
9119"4%45"11452[9\273]945"992
9
"55289374^12621771121_7594
32598H
01{3|}0~04|
{3~|5545~|3~
289374^12621771121_7594325987471
92
25452246943245
12
4%52246943245
12
773252
;EVEAABAM`abcde748WLXE@:D
fIEJ<Dghfi>CDEFBG
939812H[92859HEVEAABA@uE=@L:H59%%1294CDB>X<A=@;AB>;VLUBDDLVD
748
2
&524329
92
9
!#p
<XIBV>Lk>WLXE@:DGSp
59%%1294fU=@tE=B>WY:EX@U>ALEW>IEAE:U@:FpEVEAABA@uE=@L:>XB=qLWH
LXE@:
j9747567528&473197871744%"24524329
89374
7315129
524329
92
9
hL:D@D=B:UY>Lk>VBD<A=D
27471&
2
15
724822482459
524329
92
9
28
9527471&
l9"262!522
15
898224894
524329
77112189374
2884%528937482939
5940825
97
2
4"52
4%12^74831512^
8937439821
728
22458259
2752
57572495&257671712"53151277112189374
'8
28219"2!
52
28
2242
91825%%228941&&
4522
942
25
79287312!529829
5249871
927
231&
37&8224894524329
77112189374
!"742
1548
2242
45714%8%5
45293528
92
m932&
71
&
523
72%1&
24
56259
3711257594
!
9754&8
2242452
9271284942423245
742
15449527128
2242
42
15
4
2245423245
m317594
469164%n14%748952
7594
5248592
24
56259
3711257594
9957494
52457471&
2
15
934594!524329
89374
284528937482939
594
9182
94
5745o427
4%524329
89374
427
2
529357594719
5
1%51&p522
92!412
8&4731978
71744%
24%7128!5
293324828575524329
89374
2
252715952378334329
92
9
28
97471&
282594
9953712
93742o
&989495
2
&524329
89374
!52432
82
715
59524329
92
9
?BE=<VBD>=qE=>WL>:L=>EAALr>WLXE@:sABtBA>;EVEAABA@uE=@L:
2
29
5289374^12621771121_759432598
4957119"28"552
9119"4%
2752
H
0852326712955
m5278&^
57528252594
o
52
2
2752
7241828!742932
7%2"112
28
?BE=<VBD>=qE=>UE::L=>IB>D;A@=>EUVLDD>WLXE@:D
2574
2752
744952
1579
89374
28937482939
59471%953759375711&57n2
5
4597945
748
92
52
2
2752
592945742824521&"5494289374o
2"289374
57422
52892
9
72
27528!7
0815
2
742932
2
15
37&2
%4
7451&8
2245
935759
7
2714"54977112189374
m25524329
77112189374
59v!748
"55952199^12621771121_759432598
5
2769
482
712m22$2
44%w078756232
89374
47
0815
982571
945757744952
1579
77112189374
!5
27752945757
724952
5528592452
732
77112189374945757575
2
52n423759457571%95322
5757119
524982
7
97528"552
4691628
72
2"547
4%12771121893747484952
728"574&95277112189374
945757575
2
5224715&9457571%95322
575527
975284982
2759
7
4%1277112189374!552
24982
37&71
92759
95277112189374
471&
2
4"717%222457%29
4982
72469162849457537&495
712"211
945757
72
28!2
2711&"5n4237524
9232459
9457594
5745
x2427194575892
495
135528937482939
59494872
y982
46916284n4237594
5745
'95z4237594
5745
2!"5522825949
72^7
28
211^59^
918
94
5745
!"112"547
4%1277112189374p74852&"114952
728"57495277112189374l9"262!5"9
n4237594
5745
57589495
724982
7421728"548
224577112189374
0134567836396856
65613838
6814658964
7614638568
8
6
45813861341
4
96853
4367616
4583346314741766
478414696851464
4
968538418
16
83661
16
8469685367419684
4
54361496853
!"#$%"&'"(')*!"(&')+'!,!'"
-4.7811
.61766458167454341936714164
54
6
641388
6166161
6745436148
.36
/36761.596
48
8
6
458134
936718816
65613634.
967413
8164638568
8
4581
06718458167454341
64178196.36.
618
478
64146546
3745.8418
16136
16
4
581768135896876696116
478
6418381166161458167454341
6412
6
693
9.1745.84146
478
641385418
6
6561318.36
/367694668
46546
831417
.6181.36
/36764581
67454341
641374136
6836
498
6418138
346745436146.36
/36761.596
48
8
45813-4.781367868767454341
641781966745436.318
67.
36744
186936741789
8
4
54
8
88
4118
4
58515:6361.596
438
6146367898
4
53668.
4
8
6
458134
635.
8413
6
6
361.596
4
478
641361683166161458167454341
6413<
363.8
4=816
656138
61417
.61
478
641361683166161458167454341
6413<46
362 3><
366745436874
4
458167454341
641<
366745436874
4
6
498
458167454341
641<81
36.36
/36761.596
4458136
458167454341
641366?4581/
66
8
8
6
:841
068
86458167454341
6413589663
624
6685
6219
/3
6546
36
674187/58787
3
478
:68156661618174.
65.
6
81/@8
8181
49
653
478
:6886563
616561
6
66
656138
6
6168146745.8418
74366685
696
4A.
6=343836
78
4B67
6
58718
8
8688
.
6546
2.38
6
476331A.
6D01
36685
6 81 <6
64
62
E!F%GHI!F!"*$(')*!"(&')+'!,!'"H
012345789
115
15
11
8
E:*#A
012455674282492
156
44242457428244556
1252
1
2
265425
95
5
12596
545
012
6226
5
45567428246
4424
592
442
11
21282
159742824
6
4424
50126226
5
448
2625
122
242
12425
442
2
42
6
442445569642
2
1222
45
45
1929
5
012455674282492
15942
655
45
95
2
15
6
525
2625
12494
6426
442459
1
192
1544
2
26
442462
5
921225
259922012455674282492
155
665
25
125
64
5
9
4
24
5
192
1555
2625
1292
56
525
2
!"#$%&'()*&+,-.*/
2
1254456
5
12599420
-1-2$,34156789:;<=>#$,5:#-?-))*)5@AAB
01245567428246
4424
592
154522
1228
592
42
1228
592
42
6
92
2
#-?-))*)5CDDE
F1-2$,GHFI&+,-.*/
U595420V52
5
0W-?-))*)(X-%(4"0
5425+,*&Y$)%(#)*&#?4>*,,4?,
62
1292
56
525
:ZW-?-))*)(X-%(4"&Y*%[4\/&]44#
J*,%-?%
012
2
25
2
552
2
1
242
2622
2
4124556742824
6
4424
5K5
62
5
92
25
1292
56
525
212
2
9
2
5
5
1292
5
6
525
2
125
4401295
59952542
2
1
129292
56
525
2
125
44LM64
462
5
145567428246
4424
5
21L2
5
2
2
25
5
2
1597428246
4424
5
21L2
T
6
012
2
255
6
2
5
Michigan Technological University
Digital Commons @ Michigan Tech
2017
Recommended Citation
Roehm, Paul M., "Minimizing Run Time of Finite Element Analyses: Applications in Conformable CNG
Tank Modeling", Open Access Master's Report, Michigan Technological University, 2017.
https://doi.org/10.37099/mtu.dc.etdr/380
By
Paul M. Roehm
A REPORT
MASTER OF SCIENCE
In Mechanical Engineering
2017
1. Introduction…………………………………………………………………..1
1.1 Project Overview………………………………………………………1
1.2 Background…………………………………………………………….1
2. Problem Analysis……………………………………………………………….7
2.1 Project Goal…………………………………………………………….7
4. Mesh Selection……………………………………………………………...….14
iii
4.1 Shell v. Solid Elements (First Order)………….…………………….14
iv
8. Dynamic Modeling – Runtime Reduction Testing………………………….34
References………………...………………………………………………………46
v
Abstract
REL Inc. has proposed a CNG tank that deviates from typical cylindrical storage methods. The
goal of REL working with Michigan Tech is to minimize mass and meet NGV2 safety standards
for pressure and drop testing for this tank.
The model has undulated outer surfaces and Schwarz P-surface internal geometry. To accurately
mesh this, a small element size is necessary; this creates a model with millions of elements. In
explicit analyses, this requires a large amount of computational resources to run.
This report focuses on methods to reduce model run time without reducing accuracy. Methods
covered include creating symmetric building blocks, gradient meshing, and hybrid (shell to solid)
meshing. Results indicate that building block and gradient methods reduce run time by 84% and
78% respectively for an accuracy cost below 5%. Hybrid meshing shows a potential of 50%
additional reduction but element formulation must be changed to reduce accuracy cost.
vi
1. Introduction
1.1. Project Overview
REL, Inc. needs to optimize a conformable compressed natural gas (CNG) tank to
minimize mass while meeting national safety standards. Instead of creating a large
number of physical models and going through an excess of tooling and molding
equipment, it is cost effective to simulate these standards using finite element analysis
(FEA). While many FEA models for optimization have a short runtime due to low
number of equations (few nodes) or have a large timestep in dynamic simulations, the
CNG tank requires a large number of elements with a small timestep. Additionally, future
CNG tank model CAD is predicted to increase in size by a factor of 3.5. Therefore it is
critical to explore methods to reduce the runtime of CNG tank models while maintaining
model accuracy. This report is focused on exploring these methods and recommending a
course of action for future modeling attempts.
1.2. Background
1.2.1. Alternative Fuels
Alternative fuels are a rising interest worldwide – as such, vehicle technology has started
to shift focus toward fuels that replace traditional gasoline. This is especially prevalent in
the public transportation sector, where it is reported that, as of 2013, over 35% of U.S
public buses are using “alternative fuels or hybrid technology” to reduce emissions. [1]
This growth is also being incentivize by the US government. Congress has continued to
extend the Protecting Americans from Tax Hikes (PATH) Act which provides tax
benefits for CNG and liquefied natural gas (LNG). [2]
Alternative fuels are defined by the lack of gasoline and therefore include a broad set of
fuel types. In the transportation sector leading fuel types include biodiesel, ethanol,
electric, natural gas (CNG or LNG), and hydrogen. Figure 1 shows the number of
vehicles made for different fuel types between 2004 and 2015.
1
Figure 1: Growth of Alternative Fuels Vehicles in the Transportation Sector from 2004
to 2015 [3]
Ethanol 85 (E85) and Hybrid Electric Vehicles (HEVs) both utilize petroleum gas.
Without these two components, the leading alternative fuel vehicles as of 2015 are
biodiesel, followed by CNG and Plug-in Electric Vehicles (PEVs).
Focusing on natural gas, this fuel is highly available and is mass produced in the US. As
such it is easily accessible to the public. In addition to this, it is a clean-burning fuel and
provides similar “power, acceleration and cruising speed” compared to gasoline. The
current downside of using natural gas is that the total mileage a vehicle can travel before
refilling is reduced. [4]
2
Figure 2: Type 3 Cylindrical CNG Tank [5]
There are currently 4 types of CNG tank. As the type number increases the typical weight
decreases. Type 1 tanks are a formed steel or aluminum which typically have a protective
coating around them. Type 2 tanks use a steel liner around the volume which is
surrounded by a partial composite material ‘wrap’ in the middle of the tank. Type 3 tanks
are typically made with an aluminum liner and a full composite wrap (ex. carbon fiber).
Type 4 tanks have no metal, instead using plastic with a rubber liner and a composite
wrap around the natural gas. [6] [7]
The cylindrical tank geometry removes stress concentrations that would occur at sharp
corners. However, when mounting the cylinders, the geometry makes it difficult to place
the tanks on vehicles without a fixture that accounts for the circular cross section.
Therefore the tanks are typically placed in the back of truck beds, in trunks of cars, or on
top of buses. Of these options, only buses have an option that does not remove available
vehicle space. Additionally, because the fixtures must account for the circular cross
section, the true packaging space for the tank is a rectangular cross section along its
length. Therefore there is a motivation to find a tank design that uses this space and can
be mounted more conveniently on vehicles.
3
Figure 3: REL Modified Type 1 Conformable CNG Tank showing Additional
Volumetric Storage (left) and Internal Structure (right) [8]
The tank offers an increase in CNG storage and a more convenient mounting shape by
using radiused corners for the external surfaces and Schwarz P-surfaces for the internal
geometry. These tanks are called ‘conformable tanks’ because they are not limited to a
cube, but can instead form shapes around key components on vehicles or mounting
brackets.
4
standards must be met, the focus of this research is the high-stress, non-fracture testing
scenarios. These tests include burst and drop testing.
Because drop testing should cause the highest stress concentrations and requires dynamic
simulation, this test method is the primary focus for this report. Pressure testing
conditions are only used to determine a starting point for drop test simulations.
It should be noted that physical v. simulated NGV2 test standards are not the focus of this
work. This report is not directly concerned with a model that exactly matches the
expected strains within these tank simulation. Instead, the report focuses on assessing a
‘generally accurate’ model to discern runtime reduction methods. As such, NGV2
standards are only referenced for completeness and an understanding of where the load
and boundary conditions originate. Further discussion on defining the ‘general accuracy’
of the simulations is discussed in the Physical Validation section of the dynamic
modeling results.
Figure 5: NGV2 Drop Test Configurations for Horizontal and 45° Impact
6
Note that Figure 5 shows an exaggerated size of the current iteration of CNG tanks. The
decrease in height to initial impact in the 45° rotation configuration is not this large.
However, in the case that the height decrease is significant, the tank must be rotated such
that there is a minimum height of 24” and the CG is still at 72” above the ground. The
internal pressure of the tank for these tests should be 0 psi (empty) and dropped at
ambient temperature. [9]
2. Problem Analysis
2.1. Project Goal
The goal of the project is to recommend a method or set of methods that reduce current
model run time while maintaining model accuracy. The model run time is monitored by
CPU time taken by the simulation, while the model accuracy is measured by convergence
compared to a baseline simulation. Note that convergence has multiple measures that
must be considered. Acceptable percentile differences are further discussed in the
associated analysis sections.
• Determine a valid starting mesh size via static pressure analysis (implicit)
• Use converged static mesh size as a basis for dynamic impact convergence study
(explicit)
• Test timestep reduction methods and compare convergence to the baseline. Use
the converged dynamic impact model as this baseline.
These objectives were chosen to form a general method for replication and model
formation if and/or when new CNG tank designs are created. The final objective is the
critical piece of this report that directly answers the project goal. Note that validating a
converged mesh size should, in future, also include physical strain comparisons on a
model to further prove convergence. Since no strain data is known for drop test scenarios
at this time, this step was not available.
2.3. Limitations
Limitations of this project are runtime based. For the scope of this paper, all runs are
performed on Michigan Technological University’s SUPERIOR cluster with 4
processors. Even so, runs take up to ~17 hours. Therefore, a general upper limit of 12
hours was set to determine a lower bound of element size for baseline models in impact.
Considering phase 2 tanks are projected to increase the number of elements by a factor of
3.5, this would increase the runtime to ~42 hours per run. This time could increase
depending on quality of elements and any additional contact between parts in future
models. Running iterative mass optimization or full vehicle impact on models this large
would be too time consuming to make the process worth it.
7
This limitation constrains both the timestep and element size (aka number of equations in
the solver). Since the contact conditions given in the dynamic impact test case are simple,
the timestep is directly related to the element size, and therefore directly constrains the
minimum element size. If these results are replicated without use of SUPERIOR, element
sizes that require a large runtime may need to be removed from the analyses.
Figure 6: Imported Entire Tank (left) and Internal Geometry (right) of a Phase 1 Tank
While the tank itself is formed symmetrically, the geometric lines seen in the left image
of Figure 6 do not allow for symmetry of the elements without a large amount of ‘line
toggling’. In this process the user manually removes geometric lines from the model to
re-impose symmetry and improve mesh quality.
Another critical point in the geometry is the ‘3-radius interface’, where 3 different radii
come together in the geometry. This location occurs at every connection between the
external and internal geometry as shown in Figure 7.
8
Figure 7: 3-Radius Interface Location in CNG Tank
This location has significance in element type selection and is further discussed in the
Shell v. Solid Elements (First Order) section.
Finally, it is important to note that this geometry is not the only type that the methods
described in this paper have been used for during research. Additional models include
prototype phase 2 models, fork truck tank models, and flat external surface models.
9
Figure 8: Concrete Slab Geometry with Front (bottom left), Top (top left) and Isometric
(right) Views
For consistency with the modeling units used in this research (kg, mm, ms), the model
dimensions in mm are 152.4 x 1524 x 1524 mm.
While concrete dimensions may vary between test locations in physical testing, the
general dimensions are only necessary to have a realistic concrete impact surface. As
long as the bending moment placed upon the concrete slab is relatively small (aka impose
small displacement of concrete), user defined dimensions should be appropriate for the
impact model. Note that there is a threshold at which small concrete plates without
fracture mechanisms will create a much higher effective stiffness of the concrete.
Similarly, using a larger concrete floor may reduce simulation noise (good for
convergence) at the cost of model realism.
10
3.3. Aluminum Material Model
The material used for these CNG tanks is A206 aluminum. The heat treatment options
considered during this research include hot isostatic pressing (HIP), solution heat treated
and natural aging (T4), and solution heat treated and artificial aging (T6). This report
focuses on a T6 aging condition. The tensile curve and general material parameters are
given in Figure 9 and Table 1 respectively.
A206-T6 Aluminum
11
Within FEA solvers, the plastic curve portion of this elastoplastic material data can be
created using multiple methods. The method used in this report is the Johnson-Cook
curve fit method. This follows the following equation format:
Where σ is the stress, A is the yield stress, B is the strain hardening coefficient, n is the
strain hardening exponent, C is the strain-rate coefficient, 𝜀𝜀̇𝑝𝑝 ∗ is the normalized plastic
strain rate defined in equation 2, 𝑇𝑇 ∗ is the normalized temperature defined in equation 3,
and m is the temperature softening exponent.
𝜀𝜀𝑝𝑝̇
𝜀𝜀̇𝑝𝑝 ∗ ≡ Eqn. 2
𝜀𝜀𝑝𝑝0̇
Where 𝜀𝜀𝑝𝑝̇ is the input strain rate (strain rate at any given time), and 𝜀𝜀𝑝𝑝0̇ is the reference
strain rate of the stress-strain data.
𝑇𝑇 − 𝑇𝑇0
𝑇𝑇 ∗ ≡ Eqn. 3
𝑇𝑇𝑚𝑚 − 𝑇𝑇0
Where T is the current temperature, T0 is the reference temperature, and Tm is the melting
temperature of the material.
The Johnson-Cook curve is a version of von Mises plasticity which assumes isotropic
hardening and can account for both strain rate and temperature effects on the material.
[11] Note that the isotropic hardening assumption is critical in this model and in future
modeling of the CNG tank.
Isotropic hardening assumes that the plastic properties of a material are the same in
tension and compression even after plastic deformation occurs. Therefore any work
hardening done in a tensile element will increase its compression properties equally. In
models, this hardening approach is considered valid when there is no cyclic loading that
alternates compression and tension.
For completeness, and to help potential cyclic loading models, the alternative hardening
approach is called kinematic hardening. This model assumes the yield surface shifts as
the material yields in compression or tension. Therefore, yielding in compression will
cause a decrease in required stress to yield in tension and vice versa.
The Johnson-Cook parameters for the A206 aluminum can be found in Table 2. The units
for these parameters match the units used in the simulations (kg, mm, ms).
12
Table 2: Johnson-Cook Material Parameters Fitted to Tensile Test Data
An elastic material model was selected to model this concrete – this assumes the concrete
does not fail by plastic deformation or fracture. Since concrete is typically brittle, the ‘no
plastic deformation’ assumption should be valid. The ‘no fracture’ assumption is only
valid if the concrete is not reaching stresses high above its ultimate strength. By assuming
an elastic model without fracture for the concrete, the concrete no longer has an energy
release mechanism; the stored energy must therefore stiffen the structure and increase the
reaction forces on the tank. In effect, the elastic concrete will stiffen where actual
concrete would release energy via fracture therefore creating a ‘worst-case’ scenario for
the tank. The associated concrete properties are given in Table 3.
13
Table 3: Concrete Elastic Properties for Portland cement concrete [12]
4. Mesh Selection
Of the 3 general options for elements (1d, 2d shell and 3d solid), only 2d and 3d element
types are considered for this report. While 1d elements have potential to approximate
internal geometry, the error is likely to be too large to make it worth the run-time
reduction. As such, they are considered a potential area for future model reduction.
The internal geometry of the tank fits the requirements for a 2d mesh; there are no
thickness gradients or complex interfaces. However, the 3-radius interface shown in
Figure 7 and external geometry limit potential uses for 2d elements. A 2d mesh of the 3-
radius interface misses the connectivity between surfaces caused by the radii and the
external geometry has thickness gradients close to these interfaces. Therefore, a 2d mesh
is not recommended for the entire tank.
3d elements are the default element for 3d structures. The two primary elements are hex
(6 side cuboid), tetra (4 side triangular prism) and the connector pyramid element (5 side
pyramid). Hex elements are preferable as they shear lock less than tetra and pyramid
elements. Pyramid elements are only used when a transition between hex and tetra
elements is needed.
While hex elements are preferable, the tank geometry does not allow for parallel quad
meshes to create a valid set of hex elements. Therefore, a tetra mesh is the recommended
14
3d element style to model the tank if only 1 element type is used. These elements are
used for all initial convergence testing and defining baseline model results.
Figure 11: 3600 psi Pressure Loading Conditions on Phase 1 Quarter Tank
While this pressure could have been set to 8100psi, the convergence study was performed
at 3600psi to compare with and validate convergence study data from previous students.
However, since the load conditions would not change beyond a simple scalar multiplier
the convergence of a 3600psi load should match that of an 8100psi load.
16
associated with element stress gradients in using maximum non-averaged stresses as the
convergence criterion should be minimized.
In an effort to further prove convergence with standard deviations, 3 simulations were run
at each element size. In each simulation the model was created using a building block
mesh. This mesh formed the entire tank by creating a 2-by-2 set of cores and extending it
either in the x, y or z direction. Each of the 3 simulations had a different primary
extension direction. This method forces all element quality criteria to be identical within
an element size setting and therefore removes noise associated with remeshed element
quality. The only variable tested by this standard deviation is what element stacking
orientation effects are on the model. Therefore, the standard deviation for these models
can be considered a true standard deviation used to determine whether models are
equivalent or not. A supporting visual for this method is shown in Figure 12.
17
Figure 13: 3mm z-extension Static Test Results at 3600 psi: Simple Average Mises
Stress
Note that the stress results are given in GPa due to the simulation units used. The
maximum stress in this simulation is therefore 205.7 MPa. These images show the
location of maximum 75% averaged von Mises stress in the model; this location is
consistent between models and is a key defining feature of the static analysis for all
pressures.
Another defining feature of the model is that the stress is not completely symmetrical
along the 2x2 section of the model (shown in the bottom left image). This demonstrates
the rationale for measuring the standard deviation of the models that occurs from mesh
stack-up error. If there are small differences using identical elements, the impact of
orientation should be considered.
18
Convergence results for the CNG tank including standard deviation are shown in Table 4
and Figure 14 respectively.
Table 4: von Mises Stress and Standard Deviation at Various Element Sizes with 3600
psi Loading
Figure 14: Average Maximum von Mises Stress v. Element Size at 3600 psi with Error
Bars
19
The stress results show a decrease in accuracy below 3.5mm mesh size and a steep drop
off in accuracy below 4mm mesh size. Similarly, at 2.5mm the there is an increase in
maximum stress implying that there is still room to reduce the mesh size to increase
accuracy. However, when considering the standard deviation of the model, the 3mm
mesh size should be considered acceptable compared to the 2.5mm mesh size. Therefore,
the 3mm mesh size is chosen as the starting element size for dynamic impact analysis.
The standard deviation of these convergence results demonstrate the difficulties with
convergence criteria. While convergence can still be estimated, there is up to 2.8% error
in noise associated with mesh orientation in this model. When compounded with other
noise factors such as remeshed elements (varying element quality), and numerical error in
the solver this variation may reach 5% or higher. For this report the von Mises acceptable
deviation is considered to be below ~5%. This assumes variation will increase due to
remeshed elements and considers a statistical 5% error interval – ‘less than 5% error is
‘insignificant’.
Within these element sizes a modified edge drop test condition was the primary focus.
The edge (45° rotation) drop test will have larger stress concentrations than the horizontal
drop test because the contact surface is smaller. In order to ensure this is the case for any
CNG tank, the drop height was changed to be 72” from the lowest point of the tank. This
test condition matches that of the horizontal drop test. Therefore, the converged element
size from these results can be used for any test condition.
In addition to an edge drop test, convergence was analyzed for a corner drop test (45° /
45° rotation). This test condition is a worst-case scenario for stresses applied to the tank.
Therefore, convergence results will provide a second estimate of the minimum element
size required for convergence.
20
6.1. Dynamic Model Overview – Boundary Conditions
6.1.1. Dynamic Displacement Constraints
Displacement conditions for the drop test are set on the concrete block on all outer
surfaces except for the impact surface as shown in the sketch and actual model views of
Figure 15.
Figure 15: Sketch of Constrained Concrete Slab (top) and Model of Constrained Slab
(bottom)
This configuration assumes that the sides and bottom surface of the concrete are not
allowed to move during impact. This ‘no bending’ assumption was chosen as a worst
case scenario; minimizing bending in an elastic model increases the concrete contact
forces resisting deformation during the impact.
The alternate assumption is to leave the bottom of the concrete unconstrained (similar to
conditions where it may bend slightly, such as compressing dirt below the concrete).
Since dynamic simulations show minimal deformation of the concrete, the difference
between the two assumptions should be minimal. The worst case scenario with no
bending is preferable for determining convergence.
21
6.1.2. Dynamic Loading Conditions
Loading conditions are applied to the tank such that the kinetic energy from a 72” drop is
converted to an initial velocity. The kinetic energy of each run is calculated to be
approximately 742.3 J. This initial velocity is derived from the mass of the tank (~41.24
kg) and the transfer of 100% potential energy of the tank at 72” to 100% kinetic energy
of the tank at 0”. Using this mass, the initial velocity is approximately 6 m/s for the
baseline 3mm phase 1 tank as shown in Figure 16 via sketch and model images.
Figure 16: Sketch of Initial Velocity Condition (top) and Modeled Velocity Condition
(bottom)
22
It is critical to note that the mass of the tank will have small fluctuations depending on
the mesh size. This occurs due to geometric approximation of elements adding or
subtracting small amounts of mass. Therefore the tank velocity is recalculated for each
model using 742.3 J as the target initial total energy.
The kinetic contact method uses a predictor/corrector algorithm. It is designed for precise
enforcement of contact conditions and is recommended for contact with elastic materials
(such as the concrete slab). Since the driver for loading in the simulation is contact, this
style of contact control is preferable. Limitations of this contact method include: inability
to model rigid-rigid contact, cases with ‘chatter’ contact (vibrations) and has difficulties
when interacting with other constraints (typically one constraint is removed if
interference occurs).
For completeness, the alternate method is the penalty contact method (via penalty
algorithms). This method is a considered the ‘general’ contact method and can account
for more types of contact (such as rigid-rigid contact) and interacts well with other
constraints, but will introduce minor additional stiffness as the contact is “less stringent”
in enforcing the constraint (penalizing any penetration so the solver can correct it via
reaction force). The primary limitation of the penalty method contact analysis is any large
deformation at the contact surface with high reaction forces. These scenarios may be
caused by displacement-controlled loading, coarse mesh, or elastic responses where large
deformation occurs. Additionally, Abaqus estimates this method to decrease the timestep
by up to ~4%. [13] Since many of these conditions are met and the timestep decrease is
not ideal, this method is not chosen in this simulation.
Internal energy, in reference to this simulation, is defined as the sum of elastic and plastic
strain energy of the tank. As such it is more sensitive to stress gradients within elements
and can vary accordingly. Recoverable strain energy is only the elastic strain energy of
the tank; while this criteria is less inclusive than internal energy, it is more robust as a
23
convergence criteria in elastoplastic models as high stress gradients will not contribute to
the recoverable strain energy. The simple average signed von Mises stress is also
measured because it is the primary output considered in the models and has a clear
physical connection to the results. As such this measure should still be included when
assessing a converged model. In physical validation this criteria should also include
strain.
For completeness it must be noted that signed von Mises stress (as opposed to regular
von Mises stress) is the desired metric to track for failure criteria in the drop test
simulations. Signed von Mises stress looks at the sign of the first principal stress in an
element and gives the same sign to the corresponding von Mises stress. This is critical for
this model because the edge/corner is put into bending. This bending applies a larger
compressive stress than tensile stress to the material. Since aluminum fails at different
compressive and tensile stresses and the compressive strength is known to be much
higher than the tensile strength due to plastic flow, we focus on tensile stresses in the
model.
Figure 17: von Mises Stress Results for 2.25mm Element Size with Maximum Stress
Region (bottom)
24
Note that the stress results are given in GPa due to the simulation units used. The
maximum stress in this simulation is therefore 390.7 MPa. These images also show the
location of maximum 75% averaged von Mises stress in the model; this location is
consistent between models.
While the model shows minimal deformation, it is important to note there is not
symmetry of the stresses along the tank in these runs. This is because the tank is not
located along the midplane of the concrete slab. This does not affect the convergence
results but must be recognized in future modeling efforts. The reason for this effect is
because the concrete can deform less near the boundary conditions. Therefore, the contact
forces closer to the edge of the slab are slightly larger.
Another key feature to note is that the maximum stress exceeds the tested ultimate
strength of the material. The Johnson-Cook material model does not have an input
maximum stress. This allows the stress to continue along the estimated stress-strain curve
which avoids any truncation past the ultimate tensile strength. Additionally, as mentioned
in the Dynamic Convergence Criteria section, the signed von Mises stress is a better
measure of failure in this model. The maximum stress regions shown are compressive
regions, while the stresses on the inside of the tank are in tension. Because of the
geometry, the internal stresses are much lower than the external stresses. The von Mises
stress results shown in Figure 17 are shown in Figure 18 in signed von Mises stress for
comparison.
Figure 18: Signed von Mises Stress Results for 2.25mm Element Size with Maximum
Compressive Stress Region (middle) and Maximum Tensile Region (bottom)
25
The previous maximum von Mises stresses are shown to be compressive stresses while
the new maximum tensile stress is located on the inside of the tank as shown in the
bottom image of Figure 18. However, by using signed von Mises stress in regions where
the stresses alternate between tensile and compressive, the stress gradients appear to be
larger than that of regular von Mises stress. This occurs due to averaging.
Convergence results for the CNG corner tank impact are shown in Table 5 and Figure 19
respectively. Similarly, convergence results for edge impact are shown in Table 6 and
Figure 20 respectively.
Table 5: Convergence Criteria and Run Time at Various Element Sizes in Drop Test
Figure 19: Graphed Convergence Criteria and Run Time v. Element Size for Corner
Drop Test at Constant Initial Kinetic Energy (742.3 J)
26
Table 6: Convergence Criteria and Run Time at Various Element Sizes in Edge Drop
Figure 20: Graphed Convergence Criteria and Run Time v. Element Size for Edge Drop
Test at Constant Initial Kinetic Energy (742.3 J)
Assuming an acceptable convergence of 5% or less, the stress and energy results indicate
a minimum element size of 3mm for the corner impact. The edge impact results imply an
element size of 2.5mm. This discrepancy is likely due to the ‘simple’ nature of the corner
impact where there is only 1 impact location. In the edge case, 4 different contact points
impact the concrete with varying concrete stiffness at each point due to the boundary
conditions. This results in more simulation noise. The edge impact case can therefore be
27
considered a more complex impact scenario and the driver for convergence if a non-
infinite concrete block is used.
From the stress results, it can be concluded that the models are either converged or
beginning to converge at 2.5mm. The corner and edge impact percent difference between
stresses from 2.5 to 2.25mm are both below 1%.
Contrary to this, the energy results seem to be continuing to increase / decrease instead of
completely tapering off at 2.25mm element size. This may indicate that the model is still
not 100% converged at 2.25mm. The energy results therefore support the static analysis
results that the convergence can be improved further. However, because the model
already requires 10.5 hours to run at a 2.25mm element size and because runtime displays
an exponential growth as element size decreases, this element size is considered the
minimum allowable size for the scope of this paper (and is recommended to be
considered a minimum for optimization purposes).
With this assumption, the 2.5mm element size can be considered a valid converged
element size as there is minimal difference in output stress and only minor error in energy
results between these element sizes. While this assumption will introduce error, it should
not affect the results of applying time reduction methods to the model. As long as the
time reduction methods are applied at a constant element size, noise associated with
convergence will be minimized.
For energy convergence criteria, it is critical to note that the difference between runs is on
the order of 1’s in Joules. Considering this is very small compared to the actual value
(less than 5% error in all cases of internal energy), it is difficult to decide upon a useful
percentage error that may be acceptable. This is reinforced by the fact that strain energy
(and strain energy density) are calculated over a specific area. In this case the entire tank
was used, but in cases that only take a small area of the tank, percentage error may
fluctuate. Numerical error within the solver may also be the source of this gap as
truncation can easily cause error on the order of 1’s. Therefore, numerical error should be
kept in mind for all convergence results. Additionally it is recommended to include stress
in future convergence criteria to maintain an understanding of the physical meaning of
error between runs. In physical testing this may also include strain convergence.
Physical validation for drop testing has not been fully carried out at this point. As such,
the data that we currently have for impact scenarios is limited to photos as shown in
Figure 21.
28
Figure 21: 2.25mm Edge Impact Plastic Maximum Principal Strain Results
Both test scenarios show minimal deformation. A set of comparative views from
simulations are shown in Figures 22 and 23.
Figure 22: 2.25mm Edge Impact Plastic Maximum Principal Strain Results
Figure 23: 2.25mm Corner Impact Plastic Maximum Principal Strain Results
29
Note that the deformation of these simulations is ~1.6mm for the edge impact model and
~3.9mm for the corner impact model. Comparing the images, the edge impact case
should be considered reasonable for impact conditions. The maximum principal strain is
around the maximum seen in the material model and therefore requires minimal
extrapolation. Therefore, any errors that occur in measuring deformation are due to gaps
in the tensile v. compressive properties of the material.
In the corner impact case, the deformations are larger than depicted in the image. This
can be explained by the maximum principal strain which reaches 18% (~2x the maximum
tensile strain). This large strain forces the solver to extrapolate the stress-strain tensile
curve. Not only does this material curve need to be compressive for a better measure of
strain, the approximating tensile curve is significantly extrapolated which is likely the
root cause of this error.
Therefore, when further physical tests are performed, strain measurement is highly
recommended to improve calibration of the material model. Further compressive material
properties may also be necessary to better understand changes in tensile v. compressive
properties in the material curve. The simulations should use these results to edit the
material model and account for high compressive stress regions. This should improve
accuracy of corner impact deformation estimation on the outer surface.
Run time reduction is determined by increasing timestep (function of only element size in
these simulations) and decreasing the number of equations to solve (number of nodes).
Therefore, alterations to A206 material properties is considered a negligible factor in run
time reduction method assessment.
There are two additional methods that were considered but not used for run-time
reduction in this report. Either the user can opt to use second-order elements and reduce
the total number of elements or the concrete floor can be replaced with a rigid wall.
These methods are also described with reasoning as to why they were not used.
Figure 24: Phase 1 and 2 Model – Unique Parts for Conformable Model Creation
Note that the top line shows the bulk piece within the model, and the bottom line shows
the piece the user needs to mesh. These small pieces represent the entire geometry of a
Phase 1 or Phase 2 tank (assuming only 90° corners and edges occur). In all cases, the
smaller blocks can be merged and meshed to form a larger component to be reflected like
the one shown in Figure 25.
Figure 25: 4mm Phase 1 Base Part – Colored Segments Indicate Different Parts from
Figure 20
31
The CNG tank model can be formed from this part simply by reflecting segments of the
base part. Each reflection follows geometric lines created by boundaries between parts.
This method is ideal to enforce a defined element size and element quality in the model
and ensures that stress concentrations do not occur in local regions due to differing mesh
quality. Additionally, once the parts are created, various tank geometries can be quickly
formed by rotating and reflecting these components. Therefore this method also reflects a
decrease in model creation time for all models after the building blocks are created.
This method should be used for any model that has areas with minimal impact. However,
it is important to note the limitations to this gradient. Because there are elements that
transition from small to large element sizes, the transition should maintain the element
quality required of all other elements (typically the aspect ratio and jacobian are worse in
transition regions). To this point, the user must ensure that the gradient does not
significantly change the results of the simulation. In the case that they are affected, the
transition mesh may need to transition slower or begin farther from the critical location.
Determining if the results are significantly changed requires a full model to be tested at
the selected element size. While this may not be possible for very small element sizes
(ex. 1mm in the dynamic cases), it is the easiest method to prove results are not
significantly changed when gradient meshing.
32
Figure 26: Hybrid Element Mesh for 4mm Hybrid Model – Transition Elements in Light
Blue
Note that there are shell elements (light blue) covering the surface of the solid elements at
the 3-radius interface (red). These elements help transfer bending rotation of the 3d
elements to the shell elements on the internal P-surface (orange). Without these elements
any rotation of that surface about the point of contact between internal and external
surface would not be transferred to the internal surface. Despite this, the hybrid transition
will still not properly transfer shear stresses. Therefore, it is recommended to place this
transition away from critical regions similar to a gradient mesh.
This method does not require simulation to understand – in a symmetrical model the run
time will be reduced by approximately half for each plane of symmetry (nodes on the
plane of symmetry are not halved). Planes of symmetry only change the number of
elements (not the timestep) and therefore are linearly related to the run time.
33
these elements will converge at larger element sizes which may allow for lower run
times.
The rationale for not using second order elements in this analysis is due to the thickness
of the tank. Since the tank is thin, only 1 to 3 elements can fit within the thickness in the
range of first order elements tested. Therefore it is difficult to reduce the element size far
enough to make the computational time lower than the first order response. Additionally,
even if the element size can become large enough, the second order elements will still not
match the actual geometry. The element analysis tensors (deformation, strain, stress etc.)
may be second order but the element geometry remains first order. Any error associated
with geometry mismatch will not be fixed.
Future iterations of the CNG tanks will need to make this decision based upon accuracy
v. run time. Using a rigid will reduce time due to removing elements. Additionally, it will
reduce noise associated with where the tank is dropped on the cement slab due to a
constant stiffness (infinite) at all points.
If the user prioritizes run accuracy, a rigid wall is not recommended. The amount of
elastic energy absorbed by the concrete in edge and corner impact was approximately 140
J and 60 J respectively. This increase in energy ranges from 10% to 30% of the energy
absorbed by the tank in concrete impact simulations. While drop location on the concrete
slab will affect maximum stress, a consistent drop location can fix this issue; the increase
in energy however cannot be corrected in the model.
Additionally, because the rigid wall has an infinite stiffness, the relative stiffness of the
tank to the rigid wall is very small. This large gap in stiffness will cause truncation error
in the analysis responses and potentially add more noise to the simulation results (in
addition to error in energy absorption / dissipation).
Timestep of an explicit run is generally defined by the amount of time taken for a wave at
the speed of sound to propagate through the defining length of an element. From this
definition, the smallest element’s size is the critical limitation to increasing the timestep
34
without artificial means. Therefore within these runtime reduction tests, the minimum
element size was kept constant for each model.
For completeness it should be noted that in an explicit analysis the timestep may change
during the run depending on contact or constraint interactions. Additionally, the timestep
may be adjusted using mass-scaling or direct control (user-defined minimum). While
these methods are applicable, these are considered outside the scope of this report
because direct timestep adjustment will vary between load rate and boundary conditions
even within a model. These methods can be used in tandem with the analyzed methods.
Note that care must be taken in manually adjusting timestep; a large timestep will cause
excessive error.
Figure 27: Non-Symmetric Model (left) and Symmetrically Built Model (right)
35
Table 7: Convergence Criteria and Run Time Comparing Non-Symmetric (with ports)
and Symmetric Models
Average Max
Recoverable Internal CPU Run
Model Signed von
Strain Energy, Time,
Style Mises Stress,
Energy, J J hr:min
MPa
Non-
Symmetric 269.2 152.1 472.2 17:47:07
Edge
Symmetric
259.9 152.5 467.5 3:03:24
Edge
This conclusion is further supported by the maximum compressive stresses in the models
which are 371.2 MPa and 370.9 MPa for the non-symmetric and symmetric edge runs
respectively. Because the maximum compressive stress is a combination of bending and
the impact itself, the compressive stresses are closer together than the tensile stresses. To
fix this gap in maximum stress, it is recommended that the next set of building blocks
include the ports. While this will take extra work, it will increase the accuracy of the
results.
Looking at strain energy and internal energy, the values are within 1% and imply that the
models are close enough to approximate each other. This gap with stress reemphasizes
that stress should be tracked in convergence analyses.
As for run time, there is an 83% decrease in runtime by using the symmetric model. This
decrease occurs due to the geometric fixes and mesh quality improvement prior to
reflection. If the non-symmetric edge impact were to undergo the same improvements the
run should take a similar amount of time. However, the time required to perform these
fixes will increase as the tank size increases. In the building block method, only 1 piece
needs to be fixed for each unique geometry while fixing the entire tank requires the same
fixes to be applied to each piece.
36
This method is recommended for all CNG tank models, and any model with systematic
fixes in geometry. The demonstrated runtime reduction for this model was roughly 83%.
However, the primary gain from using this method is consistency in mesh quality and
style.
Figure 28: Gradient Mesh Transition; 2.5mm (Pink), 2.5 to 5mm (Green), and 5mm
(Light Blue)
The rationale for where to begin the transition mesh is important. In this case the
maximum stresses are all found within the first ¼ core (pink). Therefore, the lower
stresses outside of this region do not require a fine mesh. Using this information, the
transition zone (green) was placed outside of this critical region. Without this distinction
the method is not applicable.
37
Convergence results for the CNG tank edge impact are shown in Table 8 and Figure 29
respectively.
Table 8: Convergence Criteria and Run Time at Various Transition Element Sizes
2.5mm
293.6 (0) 146.5 (0) 467.4 (0) 6:34 (0)
(Baseline)
Figure 29: Graphed Convergence Criteria and Run Time v. Transition Element Size at
Constant Initial Kinetic Energy (742.3 J)
38
While the graphs initially seem to indicate a large decrease in accuracy between runs it is
important to note the scale of these points from Table 8. The stress results vary by less
than 0.1% between tests. Similarly, all values of internal energy are within 0.15% of each
other. The recoverable strain energy shows the largest difference between models which
goes up to ~3 J difference between the baseline and final model and is an ~2% error. The
closeness of these results implies that the gradient method is equivalent at all tested
values of transition element size.
It is critical to note that this does not mean that any element size can be used as a
transition size. The element sizes were limited to 5mm by measuring the aspect ratio of
the transition elements. At 5.5mm element size, the transition elements exceeded an
aspect ratio of 5. Therefore, the limitation on transition element size should be set by the
user’s quality criteria. However, this may allow for the user to continue the transition
from 5mm to a larger element size assuming the mesh quality is still met.
Overall, the gradient mesh method is recommended for every model with non-critical
stress locations. Using this method the runtime of edge impact models are shown to
decrease by up to 78% on top of the gains displayed by the Symmetric Building Block
method. This value should increase in corner impact tests because the entire edge does
not need to be at the converged element size. Again, it must be noted that these tests are
only considered valid after showing the gradient meshes do not deviate from the constant
element size mesh. Therefore, this method is recommended for iterative runs where the
runtime reduction can be utilized at least 1 subsequent model.
An additional model was created using a combination of the gradient mesh and hybrid
mesh styles. This model follows the ‘1 cell in’ method but uses a gradient mesh from
2.5mm to 4mm to test the combined effect of the two methods. This ‘1 cell in gradient’
model is shown in Figure 33.
39
Figure 30: ‘Full 2d’ Hybrid Mesh Model – Impact Edge and Surrounding Cores
Figure 31: ‘1 Cell In’ Hybrid Mesh Model – Impact Edge and Surrounding Cores
40
Figure 32: ‘Half 2d’ Hybrid Mesh Model – Impact Edge (lower left) and Mesh
Transition
41
Figure 33: ‘1 Cell In Gradient’ Hybrid Mesh Model – Impact Edge and Surrounding
Cores
The ‘full 2d’, ‘1 cell in’, and ‘half 2d’ models are designed to find where the transition
region needs to be to create an accurate model. The ‘1 cell in gradient’ model is designed
to understand the theoretical minimum time the model can run using the proposed
methods. Convergence and runtime results for all 4 models and the baseline are given in
Table 9.
Table 9: Convergence Criteria and Run Time for Hybrid Element Models
Baseline
293.6 (0) 146.5 (0) 467.4 (0) 6:34 (0)
(full 3d)
1 Cell In
346.6 (18) 149.37 (2.0) 461.6 (1.2) 2:08* (65)
Gradient
42
In these cases the max signed von Mises stress shows that the hybrid mesh method used
is not applicable with the current connections between 2d and 3d elements. It should be
noted that the unsigned von Mises stress does not vary from the max von Mises stress of
the baseline model. This further proves that signed von Mises stress should be considered
in the convergence criteria.
The reason for this discrepancy in maximum stress is attributed to the transition regions
in the hybrid models. Because the connection elements cannot match the external
geometry (via thickness property), the bending stiffness of these elements is lower than
the actual bending stiffness in the tank.
The energy results all show below 3% difference in internal and recoverable strain
energy. These results are both small compared to the typical 5% statistical standard for
significant difference. In fact, these are large relative to the changes in energy seen in the
gradient models and in convergence testing. This supports the conclusion that the current
method of hybrid meshing does not properly replace the full 3d model. Additionally it
supports the conclusion that energy measures should not be the only consideration for
convergence of an FEA model.
The CPU time results show a minor decrease in run time when using a constant element
size and a slight increase in run time as the number of tetra elements increases in the
model. The reason for this is the timestep itself. In the hybrid models, Abaqus uses the
minimum timestep from all elements (in this case 9.54e-5 ms). However, in the tetra
models, the timestep is almost twice this value (1.78e-4 ms). Because the tetra models
can be run at a larger timestep it can be inferred that the hybrid mesh can also be run at
this timestep, reducing the run time.
With a user-defined timestep equal to the tetra mesh timestep, the hybrid models would
significantly reduce the run time of the model. This is especially true of the ‘1 cell in
gradient’ model which would reduce the run time to approximately 1 hour instead of 2.
The ‘1 cell in gradient’ model shows minimal difference from the ‘1 cell in’ model (less
than 0.5 MPa and 0.5 J) and reduces the run time by more than a factor of 2 (50%). This
is promising for combining the two methods if a valid hybrid mesh can be created.
Overall, the hybrid mesh must be fixed before further conclusions can be made on
potential time reductions.
43
Additionally, it is recommended that convergence criteria includes signed or unsigned
von Mises stress along with strain energy criteria. This allows the user to better
understand the physical implications of the convergence results.
The recommended methods to use for reducing model run time is the Symmetric
Building Block method and gradient meshing. These methods should be combined both
to improve consistency in geometry fixes and ensure that a valid transition region is
selected.
The Symmetric Building Block method should be performed by taking the primary parts
required to build the CNG tank, fixing their geometry and combining them into the
proper parts manually. This method is primarily based upon reflection and therefore does
not require a large amount of time. Results from the Symmetric Building Block Analysis
section indicate that, while ports were initially ignored, they should be included in the
building blocks to avoid small errors in maximum signed von Mises stress. This meshing
method reduced model runtime by 84% in the tested models. However, this reduction
will fluctuate depending on the amount of time the user spends attempting to fix the
entire tank if the method is not used.
The gradient mesh method should be performed after an initial (uniform element size)
model of the tank has be run to determine a valid element transition starting point and
baseline model. The mesh must maintain the user’s minimum element quality at all times
during the transition. For the edge impact condition, this transition was valid outside of
the ¼ core where maximum stresses occurred. It is recommended that the user measure
convergence criteria for all gradient models to ensure the model does not vary much from
the baseline model. Gradient meshing showed a runtime reduction of up to 78%. This
gain should be larger for corner impact.
Hybrid meshing in the style used in this report shows promise for reducing runtime by an
additional 50% on top of that indicated by gradient meshing with user-defined timestep
control (though no reduction occurs if timestep is not set to the value in non-hybrid
models). However, the meshing style used was not valid for maximum signed von Mises
stress convergence criteria and must be fixed before usage in any optimization runs.
Hybrid meshing work may require user-defined elements or some other element type that
can fully transfer stresses from 3d to 2d elements. If a hybrid meshing method could be
used, there is a high potential to further reduce runtime.
Mass-scaling or user-defined timesteps should be worked into impact analyses rather than
studied directly. These methods typically are model and element dependent and will
44
likely shift between element sizes and geometries. This is especially true in the case of
potential thickness optimizations.
If further work must be done in increasing accuracy for high deformation loading
conditions, the material model in compression should be used to predict deformation.
Methods such as adaptive remeshing may need to be included if elements deform out of
acceptable tolerance.
45
References
[1] More than 35% of U.S. Public Transit Buses Use Alternative Fuels or Hybrid
Technology [Online]. Available: http://www.apta.com/mediacenter/pressreleases/2013
/Pages/130422_Earth-Day.aspx
[2] Congress Passes Continuing Resolution into New Administration [Online]. Available:
http://www.apta.com/gap/legupdatealert/2016/Pages/Congress-Passes-Continuing-
Resolution-into-New-Administration.aspx
[3] Alternative Fuels Data Center: Maps and Data - Clean Cities Alternative Fuel and
Advanced Vehicle Inventory [Online]. Available: http://www.afdc.energy.gov/data/
10581
[9] L. Gambone, “Overview of NGV Fuel Tank Testing Requirements” CSA Group,
Nov 18, 2013. [Online]. Available: https://www.arpa-e.energy.gov/sites/default/files/
documents/files/6_SessionA_Gambone_0.pdf
46