Professional Documents
Culture Documents
José A. Sanchidrián
- This is not different from many other situations in engineering, physics, chemistry,
natural sciences… We need models borne in mathematical formulae for our
engineers’ work.
- We know material models are not exact but we are all the time describing nature by
means of mathematical expressions. Even the perfect gas law does not describe gases
exactly… and, under certain conditions, it does not describe gases at all! We know
that quite well when we look for equations of state in detonation physics!
- So we are perfectly well in using formulae to describe the nature or the behavior of
materials but we must be aware of their limits of use. Away from these, the reality
may be different (or very different) than what our formulae tell.
We want to describe
the size distribution of
rock fragments from
blasting by means of
an analytical
distribution function
x n OLS fits
FWRR 1 exp , 0 x
xc Rosin-Rammler
RRW
x x x n OLS fits
FS WRR 1 exp max c
, 0 x xmax
xc xmax x Rosin-Rammler
Scaled-Rosin-Rammler
S-RRW RRW
x x x
x x
OLS fits
FS GRA 1 1
max g
exp x max g
, 0 x xmax
x g xmax x
x g xmax x
Rosin-Rammler
Scaled-Rosin-Rammler
Scaled-Grady
S-GRA
S-RRW RRW
1 OLS fits
FSWE b
, 0 x xmax
log( xmax / x) Rosin-Rammler
1
log( xmax / x50 )
Scaled-Rosin-Rammler
Scaled-Grady
Swebrec
SWE
S-GRA
S-RRW RRW
x n1
x n2
OLS fits
FBiWRR (1 f )1 exp
f 1 exp
, 0 x
xc1
xc2
Rosin-Rammler
Scaled-Rosin-Rammler
Scaled-Grady
Swebrec
Bimodal Rosin-Rammler
SWE
Bi-RRW S-GRA
S-RRW RRW
1
1
2
2
OLS fits
FBiGRA (1 f )1 1
x exp x
f 1 1
x exp x
, 0 x
xg x g1 xg x g2 Rosin-Rammler
1 2
Scaled-Rosin-Rammler
Scaled-Grady
Swebrec
Bimodal Rosin-Rammler
Bimodal Grady
SWE
Bi-RRW S-GRA
Bi-GRA
S-RRW RRW
1 OLS fits
FExSWE b c
, 0 x xmax
log( xmax / x) ( xmax / x) 1 Rosin-Rammler
1 a (1 a)
log( xmax / x50 ) ( xmax / x50 ) 1
Scaled-Rosin-Rammler
Scaled-Grady
Swebrec
Bimodal Rosin-Rammler
Bimodal Grady
SWE
Extended Swebrec
2
min wi F ( xi , k ) pi
i
1
wi s
pi
Rosin-Rammler
x n
FWRR 1 exp , 0 x
xc
OLS
Rosin-Rammler
x n
FWRR 1 exp , 0 x
xc
1/P0.5
OLS
Rosin-Rammler
x n
FWRR 1 exp , 0 x
xc
1/P
1/P0.5
OLS
Rosin-Rammler
x n
FWRR 1 exp , 0 x
xc
1/P2
1/P
1/P0.5
OLS
Scaled Rosin-Rammler
x x x n
FS WRR 1 exp max c
, 0 x xmax
xc xmax x
OLS
Scaled Rosin-Rammler
x x x n
FS WRR 1 exp max c
, 0 x xmax
xc xmax x
1/P0.5
OLS
Scaled Rosin-Rammler
x x x n
FS WRR 1 exp max c
, 0 x xmax
xc xmax x
1/P
1/P0.5
OLS
Scaled Rosin-Rammler
x x x n
FS WRR 1 exp max c
, 0 x xmax
xc xmax x
1/P2
1/P
1/P0.5
OLS
Bimodal Rosin-Rammler
OLS x n1
x n2
FBiWRR (1 f )1 exp
f 1 exp
, 0 x
xc1
xc2
Bimodal Rosin-Rammler
1/P0.5
OLS x n1
x n2
FBiWRR (1 f )1 exp
f 1 exp
, 0 x
xc1
xc2
Bimodal Rosin-Rammler
1/P
1/P0.5
OLS x n1
x n2
FBiWRR (1 f )1 exp
f 1 exp
, 0 x
xc1
xc2
Bimodal Rosin-Rammler
1/P2
1/P
1/P0.5
OLS x n1
x n2
FBiWRR (1 f )1 exp
f 1 exp
, 0 x
xc1
xc2
The data:
136 data sets, bench blasting, muckpile sived
Fits of:
RRW, S-RRW, S-GRA, SWE,
Bi-RRW, Bi-GRA, ExSWE
Weights:
1, 1/P1/2, 1/P, 1/P2
Log error in size:
eL log( x*p / x p )
xp x*p
Example of errors
Scaled Rosin-Rammler
1/p1/2-weighed fits
3.5
Error points
Maximum error Median
3 95 percentile
Expected error:
2.5
𝑚𝑒𝑑( 𝑒𝐿 )
Absolute log error
2
Expected error
Maximum error:
1.5 𝑞95 ( 𝑒𝐿 )
0.5
0 -1 0 1 2
10 10 10 10
Passing (%)
Expected error
OLS fits
0.8
RRW
100 % S-RRW
0.7 S-GRA
SWE
0.6 Bi-RRW
Absolute log error, median
Bi-GRA
Ex-SWE
0.5
50 %
0.4
x*p x p x*p
0.3 eR 1
xp xp
25 %
0.2 eL log( x*p / x p ) log(eR 1)
10 %
0.1
0 -1 0 1 2
10 10 10 10
Passing (%)
Expected error
1/p1/2 fits
0.8
RRW
S-RRW
0.7 100 %
S-GRA
SWE
0.6 Bi-RRW
Absolute log error, median
Bi-GRA
Ex-SWE
0.5
50 %
0.4
x*p x p x*p
0.3 eR 1
xp xp
25 %
0.2 eL log( x*p / x p ) log(eR 1)
10 %
0.1
0 -1 0 1 2
10 10 10 10
Passing (%)
Expected error
1/p fits
0.8
RRW
S-RRW
0.7 100 % S-GRA
SWE
0.6 Bi-RRW
Absolute log error, median
Bi-GRA
Ex-SWE
0.5
50 %
0.4
x*p x p x*p
0.3 eR 1
xp xp
25 %
0.2 eL log( x*p / x p ) log(eR 1)
0.1 10 %
0 -1 0 1 2
10 10 10 10
Passing (%)
Expected error
1/p2 fits
0.8
RRW
S-RRW
0.7 100 % S-GRA
SWE
0.6 Bi-RRW
Absolute log error, median
Bi-GRA
ExSWE
0.5
50 %
0.4
x*p x p x*p
0.3 eR 1
xp xp
25 %
0.2 eL log( x*p / x p ) log(eR 1)
0.1 10 %
0 -1 0 1 2
10 10 10 10
Passing (%)
Maximum error
OLS fits
1.6
RRW
S-RRW
1.4 300 % S-GRA
Absolute log error, 95 percentile SWE
1.2 Bi-WRR
200 % Bi-GRA
Ex-SWE
1
0.8
100 %
x*p x p x*p
0.6 eR 1
xp xp
50 %
0.4 eL log( x*p / x p ) log(eR 1)
25 %
0.2
10 %
0 -1 0 1 2
10 10 10 10
Passing (%)
Maximum error
1/p1/2 fits
1.6
RRW
300 % S-RRW
1.4 S-GRA
Absolute log error, 95 percentile SWE
1.2 Bi-RRW
200 % Bi-GRA
Ex-SWE
1
0.8
100 %
x*p x p x*p
0.6 eR 1
xp xp
50 %
0.4 eL log( x*p / x p ) log(eR 1)
20 %
0.2
10 %
0 -1 0 1 2
10 10 10 10
Passing (%)
Maximum error
1/p fits
1.6
RWW
300 % S-RRW
1.4 S-GRA
Absolute log error, 95 percentile SWE
1.2 Bi-WRR
200 % Bi-GRA
Ex-SWE
1
0.8
100 %
x*p x p x*p
0.6 eR 1
xp xp
50 %
0.4 eL log( x*p / x p ) log(eR 1)
25 %
0.2
10 %
0 -1 0 1 2
10 10 10 10
Passing (%)
Maximum error
1/p2 fits
1.6
RRW
300 % S-RRW
1.4 S-GRA
Absolute log error, 95 percentile SWE
1.2 Bi-RRW
200 % Bi-GRA
Ex-SWE
1
0.8
100 %
x*p x p x*p
0.6 eR 1
xp xp
50 %
0.4 eL log( x*p / x p ) log(eR 1)
25 %
0.2
10 %
0 -1 0 1 2
10 10 10 10
Passing (%)
Expected error (%) Maximum error (%)
Size-scaled Bi-components Size-scaled Bi-components
Passing Weights Passing Weights
(%) exponent RRW S-RRW S-GRA SWE Bi-RRW Bi-GRA Ex-SWE (%) exponent RRW S-RRW S-GRA SWE Bi-RRW Bi-GRA Ex-SWE
99 0 23 7 7 10 1 2 5 99 0 109 33 33 33 50 65 33
0.5 26 8 8 11 2 3 6 0.5 164 58 57 40 63 77 40
1 26 8 10 14 5 5 6 1 293 58 74 47 62 96 48
2 61 8 11 31 9 11 7 2 1798 83 95 626 178 460 95
95 0 15 5 6 7 1 2 4 95 0 86 31 28 30 15 19 29
0.5 19 7 7 8 2 2 4 0.5 148 44 56 39 42 29 36
1 19 7 8 9 4 4 4 1 288 55 69 43 51 81 43
2 42 8 8 18 5 6 6 2 1505 81 94 407 138 404 91
90 0 8 5 6 5 1 1 2 90 0 31 28 24 25 9 14 13
0.5 12 5 6 6 1 1 2 0.5 31 31 34 29 12 14 14
1 14 7 8 5 2 2 2 1 48 47 35 38 18 16 14
2 29 7 8 12 3 4 5 2 1322 78 96 101 38 28 22
80 0 5 2 3 4 1 1 1 80 0 20 20 19 20 6 7 10
0.5 5 3 3 4 1 1 1 0.5 25 21 20 20 7 9 10
1 7 4 4 4 1 1 2 1 37 23 23 20 8 12 11
2 11 6 6 7 2 2 3 2 940 47 42 30 19 20 15
50 0 6 4 5 4 1 1 3 50 0 21 20 18 14 10 11 9
0.5 5 5 5 5 2 2 3 0.5 22 21 18 14 10 10 10
1 6 5 5 5 2 2 3 1 22 21 20 15 9 9 12
2 12 7 8 6 2 2 3 2 349 62 81 21 12 12 17
< 5 % 20 < 25 %
0 14 8 7 10 4 4 5 20 0 34 34 35 30 22 22 26
5 – 10 % 0.5 8 6 5 9 4 3 4 25 – 50 % 0.5 26 31 34 25 20 23 22
1 6 5 5 8 4 4 3 1 38 31 34 22 24 33 20
10 – 25 % 2 6 4 4 5 4 4 3 50 – 100 % 2 46 46 66 17 24 32 20
10 0 39 13 14 13 6 7 8 10 0 162 78 94 66 38 39 58
> 25 % > 100 %
0.5 22 9 9 11 6 7 5 0.5 121 62 73 47 27 32 42
1 16 9 11 11 6 6 3 1 95 60 64 41 22 24 34
2 11 11 12 8 4 5 4 2 60 42 50 29 17 15 28
5 0 59 23 34 27 8 7 8 5 0 458 207 290 178 87 79 104
0.5 35 11 18 15 6 5 7 0.5 235 90 108 98 32 42 69
1 15 13 12 15 6 4 8 1 101 49 55 79 19 20 48
2 11 12 12 10 3 4 7 2 61 53 56 34 32 33 35
2 0 272 149 193 62 16 20 22 2 0 2555 769 1207 282 200 284 143
0.5 180 76 99 32 9 15 13 0.5 1561 512 702 173 118 148 103
1 73 29 39 19 10 10 12 1 650 343 471 128 61 47 87
2 25 28 30 11 9 11 12 2 128 116 110 142 37 35 44
1 0 1224 690 1123 195 44 70 54 1 0 8701 2098 3504 1535 422 661 637
0.5 725 399 572 165 31 33 53 0.5 4075 1049 1574 2193 359 419 368
1 342 195 285 97 21 17 35 1 1360 670 1048 1128 122 155 218
2 44 35 41 15 11 13 14 2 160 155 145 220 51 52 79
0.5 0 2808 1061 1898 346 82 154 60 0.5 0 10725 3689 5105 4421 573 810 2165
0.5 1299 529 863 262 74 89 61 0.5 5223 1732 2329 3332 405 573 691
1 580 239 376 199 35 49 42 1 2046 961 1245 1815 226 282 488
2 53 38 43 15 10 13 13 2 249 224 217 383 67 57 150
- Muckpile fragmentation data (sieving) will probably have an error of about
10 – 20 %.
- You don’t want to add much more to that when you fit a function.
- If you are using a distribution function upfront (e.g. when using a
prediction formula – Kuz-Ram type) you want it to be able to represent
fragmentation with an error comparable to experimental.
Swebrec 3 100 – 3 4 – 14 13 – 72
Bi-component Rosin-Rammler 5 99 – 2 1 – 10 8 – 61
4 – 17 4 – 17 4 – 14 1 – 10 3 – 14
17 - 96 14 - 83 13 - 72 8 - 61 11 - 95
98 - 10
4 – 25
15 - 161
100 - 2
Function Err. Exp. Err. Max. 100 - 1 4 – 19
(%) (%) 4 – 35 13 - 124
R-R 4 – 17 17 – 96 15 - 136 99 – 0.5
4 – 25 15 – 161
2 – 11 100 – 0.5
S-R-R 4 – 17 14 – 83
4 – 35 14 – 136 9 - 178 3 – 14
Swebrec 4 – 14 13 – 72 9 - 150
4 – 19 13 – 124
Bi R-R 1 – 10 8 – 61
2 – 11 9 – 178
Ex-Swebrec 3 – 14 11 – 95
3 – 14 11 – 150
No. of constants 2 3 3 5 5
4 – 17 4 – 17 4 – 14 1 – 10 3 – 14
17 - 96 14 - 83 13 - 72 8 - 61 11 - 95
98 - 10
4 – 25
15 - 161
100 - 2
Function Err. Exp. Err. Max. 100 - 1 4 – 19
(%) (%) 4 – 35 13 - 124
R-R 4 – 17 17 – 96 15 - 136 99 – 0.5
4 – 25 15 – 161
2 – 11 100 – 0.5
S-R-R 4 – 17 14 – 83
4 – 35 14 – 136 9 - 178 3 – 14
Swebrec 4 – 14 13 – 72 9 - 150
4 – 19 13 – 124
Bi R-R 1 – 10 8 – 61
2 – 11 9 – 178
Ex-Swebrec 3 – 14 11 – 95
3 – 14 11 – 150
Thank you !
https://sites.google.com/site/explosivesmadridupm/
ja.sanchidrian@upm.es
Rosin-Rammler
96 %
96 – 13 %
4 @ 71-79 / 17 @ 96
17 @ 69-74 / 96 @ 96
13 %
Rosin-Rammler
98 %
96 – 13 %
4 @ 71-79 / 17 @ 96
17 @ 69-74 / 96 @ 96
98 – 10 %
4 @ 68-78 / 25 @ 98 10 %
15 @ 71 / 161 @ 98
Scaled-Rosin-Rammler
100 %
100 – 3 %
4 @ 19-20 / 17 @ 3
14 @ 71 / 83 @ 99
3%
Scaled-Rosin-Rammler
100 %
100 – 3 %
4 @ 19-20 / 17 @ 3
14 @ 71 / 83 @ 99
100 – 1 %
4 @ 19-20 / 35 @ 1
14 @ 71 / 136 @ 1
1%
SWEBREC
100 %
100 – 3 %
4 @ 56-89 / 14 @ 99
13 @ 66-69 / 72 @ 3
3%
SWEBREC
100 %
100 – 3 %
4 @ 56-89 / 14 @ 99
13 @ 66-69 / 72 @ 3
100 – 2 %
4 @ 56-89 / 19 @ 2
13 @ 66-69 / 124 @ 2 2%
Bimodal Rosin-Rammler
99 %
99 – 2 %
1 @ 70-85/ 10 @ 2
8 @ 68-82 / 61 @ 2, 99
2%
Bimodal Rosin-Rammler
99 %
99 – 2 %
1 @ 70-85/ 10 @ 2
8 @ 68-82 / 61 @ 2, 99
99 – 0.5 %
2 @ 28-88/ 11 @ 0.5
9 @ 62-70 / 178 @ 99
0.5 %
Extended Swebrec
100 %
100 – 1 %
3 @ 44-86 / 14 @ 1
11 @ 67 / 95 @ 99
0.5 %
Extended Swebrec
100 %
100 – 1 %
3 @ 44-86 / 14 @ 1
11 @ 67 / 95 @ 99
100 – 0.5 %
3 @ 44-86 / 14 @ 0.5
11 @ 67 / 150 @ 0.5 0.5 %