Baikos Weibull Estimation Spreadsheet by TaylorRandle

VIEWS: 48 PAGES: 11

									                                   6a8cb18c-08a4-4010-bafb-67ee85779552.xls


       1      2       3          4         5      6       7    8     9  10 11 12   13
 1   MAXIMUM LIKELIHOOD ESTIMATION OF WEIBULL PARAMETERS WITH A WORKSHEET
 2                       Larry George
 3               Oct. 23, 1987 revised July 3, 1989
 4              Tables 1, 2, and 3 installed in May 1990.
 5                 Table 6 installed Sept. 25, 1991
 6       This worksheet computes the maximum likelihood estimator (mle)
 7   of the two parameter Weibull distribution function. The Weibull
 8   distribution function F(y) is defined by
 9
10           1 - F(y) = P[Life > y] = exp[-(y/a)^b].
11
12   The two parameters to be estimated are a and b. Data may be
13   multiply censored; i.e. some data may be observed lives of unfailed
14   items and some may be the lives of failed items. Reference [1] gives
15   the mle computation method. It requires a 1-dimension search for the
16   mle of b. The worksheet automatically computes the mle of a and the
17   mode from the value of b.
18       Study my data and formulas before modifying the worksheet to
19   compute estimators from your data. (Columns A and B contain Terry
20   Baiko's rolling reliability data); column A indicates whether the
21   observed life in column B was terminated by failure or not. At the
22   bottom of column A is the total number of failures. Column C
23   contains the lives from column B raised to the power b. At the bottom
24   of column C are the sums of the column entries and the mle of a.
25   Column D contains the natural logarithms of the lives. Column E
26   contains terms used to compute the mle of b. At the bottom of column
27   D are the mle of b and my guess for b. Column F uses the mle of a and
28   the mle of b to compute an estimate of the Weibull 1-F(y) at the data
29   points.
30       Table 6. computes the regression estimators of the parameters.
31   You may want to use the regression estimate of b as a guess to start
32   the maximum likelihood estimation.
33       The worksheet also constructs the Kaplan-Meier estimate of the
34   complement of the cumulative distribution function, reference [2].
35   Column G contains the Kaplan-Meier estimate of 1 - F(y).
36       Terry's data had a lot of survivors after the last failure. The
37   number is imbedded in the formula for the last row, column C.
38       The mle of the mode is below column G.
39       Once you see how I set up the worksheet, use it for your data.
40   Copy your life data in column B in place of mine. Insert or delete
41   rows in the middle of my data if you have more or fewer samples.
42   Indicate a failure with a 1 and a survivor with a 0 in column A.
43       To compute the mle of b, input a positive starting value for
44   the parameter b in the cell to the right of the label "b =", under
45   the cell to the right of the label "b(mle)=". Hit calc (splat-= in
46   Excel) and the spreadsheet will compute its estimate of b under
47   column E. Change the value of b in the upper left hand corner toward
48   the value under column E. Hit calc again. Repeat until the two
49   b-values converge. Read the mle of a under column C. If the
50   computation doesn't converge, try different starting values for b.
51       You may also have to revise the Kaplan-Meier formulas if you


                                                       Page 1
                                           6a8cb18c-08a4-4010-bafb-67ee85779552.xls


        1     2        3         4         5      6     7                       8          9         10           11          12          13
 52   have different failures than Terry had. The
 53   Kaplan-Meier estimate changes only at failure values.
 54
 55                       REFERENCES
 56   [1] Wayne Nelson, "Applied Life Data Analysis", Wiley, New York,
 57   1982
 58   [2] E. L. Kaplan and P. Meier, "Nonparametric Estimation from
 59   Incomplete Observations", J. Am. Stat. Assoc., Vol. 53, pp 457-481,
 60   1958
 61
 62   Table 1. Estimate Weibull cdf of time to failure of computer.
 63                                                       y        Weibull 1-F(y) -->                                      Failure rates
 64   fail?=1 time hrs a           sumlnxi         b                                   logic boards
                                                          t, hours all analog boards HDA                       Kaplan-MeierWeibull Empirical
 65        1       1 1.0E+00                0           0        1 0.999 0.9999 0.9997 0.9998                    0.9985 0.0005 0.0015
 66        1       1 1.0E+00                0           0        1 0.999 0.9999 0.9997 0.9998                    0.9985 0.0005                 0
 67        1       1 1.0E+00                0           0        1 0.999 0.9999 0.9997 0.9998                    0.9985 0.0005                 0
 68        1       2 1.4E+00 0.69315 -0.0216                     2 0.999 0.9998 0.9996 0.9997                     0.997 0.0003 0.0015
 69        1       2 1.4E+00 0.69315 -0.0216                     2 0.999 0.9998 0.9996 0.9997                     0.997 0.0003                 0
 70        1       2 1.4E+00 0.69315 -0.0216                     2 0.999 0.9998 0.9996 0.9997                     0.997 0.0003                 0
 71        1     10 2.9E+00 2.30259 -0.0717                    10 0.997 0.9995 0.9991 0.9994                     0.9965 0.0001 6E-05
 72        1     13 3.3E+00 2.56495 -0.0799                    13 0.997 0.9994 0.999 0.9993                      0.9961 0.0001 0.0002
 73        1     24 4.4E+00 3.17805 -0.0989                    24 0.995 0.9992 0.9986 0.9991                     0.9946 9E-05 0.0001
 74        1     24 4.4E+00 3.17805 -0.0989                    24 0.995 0.9992 0.9986 0.9991                     0.9946 9E-05                  0
 75        1     24 4.4E+00 3.17805 -0.0989                    24 0.995 0.9992 0.9986 0.9991                     0.9946 9E-05                  0
 76        1     28 4.8E+00 3.3322 -0.1036                     28 0.995 0.9991 0.9985 0.9991                     0.9941 8E-05 0.0001
 77        1     34 5.2E+00 3.52636 -0.1096                    34 0.995 0.999 0.9984 0.999                       0.9936 8E-05 8E-05
 78        1     65 7.0E+00 4.17439 -0.1295                    65 0.993 0.9984 0.9977 0.9986                     0.9931 5E-05 2E-05
 79        1     70 7.3E+00 4.2485 -0.1318                     70 0.992 0.9983 0.9976 0.9986                     0.9926 5E-05 1E-04
 80        1     72 7.4E+00 4.27667 -0.1326                    72 0.992 0.9983 0.9976 0.9985                     0.9916 5E-05 0.0005
 81        1     72 7.4E+00 4.27667 -0.1326                    72 0.992 0.9983 0.9976 0.9985                     0.9916 5E-05                  0
 82        1     84 7.9E+00 4.43082 -0.1373                    84 0.992 0.9981 0.9974 0.9984                     0.9911 5E-05 4E-05
 83        1     88 8.1E+00 4.47734 -0.1387                    88 0.992 0.9981 0.9973 0.9984                     0.9906 5E-05 0.0001
 84        1     90 8.2E+00 4.49981 -0.1394                    90 0.991 0.9981 0.9973 0.9984                     0.9901 4E-05 0.0002
 85        1 113 9.1E+00 4.72739 -0.1463                      113 0.99 0.9978 0.9969 0.9982                      0.9896 4E-05 2E-05
 86        1 140 1.0E+01 4.94164 -0.1528                      140 0.989 0.9974 0.9966 0.998                      0.9891 4E-05 2E-05
 87        1 148 1.0E+01 4.99721 -0.1545                      148 0.989 0.9974 0.9965 0.998                      0.9886 3E-05 6E-05
 88        1 160 1.1E+01 5.07517 -0.1568                      160 0.989 0.9972 0.9963 0.9979                     0.9877 3E-05 8E-05
 89        1 160 1.1E+01 5.07517 -0.1568                      160 0.989 0.9972 0.9963 0.9979                     0.9877 3E-05                  0
 90        1 169 1.1E+01 5.1299 -0.1585                       169 0.989 0.9971 0.9962 0.9978                     0.9872 3E-05 6E-05
 91        1 192 1.2E+01 5.2575 -0.1623                       192 0.988 0.9969 0.9959 0.9977                     0.9867 3E-05 2E-05
 92        1 256 1.3E+01 5.54518 -0.1709                      256 0.986 0.9963 0.9953 0.9974                     0.9862 3E-05 8E-06
 93        1 264 1.4E+01 5.57595 -0.1718                      264 0.986 0.9962 0.9952 0.9974                     0.9857 3E-05 6E-05
 94        1 330 1.5E+01 5.79909 -0.1784                      330 0.984 0.9956 0.9946 0.9971                     0.9847 2E-05 2E-05
 95        1 332 1.5E+01 5.80513 -0.1785                      332 0.984 0.9956 0.9946 0.9971                     0.9842 2E-05 0.0003
 96        1 336 1.5E+01 5.81711 -0.1789                      336 0.984 0.9956 0.9946 0.997                      0.9837 2E-05 0.0001
 97        0 336 3.0E+04 5.81711 5.77211                      336 0.984 0.9956 0.9946 0.997                      0.9837 2E-05                  0
 98   ------- ------- ----------- ----------- ----------- -------- -------- ---------- ---------- ----------   ------------------------------------------------
 99      32 2033 30531.2 b(mle)= 0.46782 mode =                           0
100           a = 2339627             b=       0.46782 Median##### hours
101   Weibull mean is b*Gamma(1/b+1) --> MTBF #####
102


                                                                   Page 2
                                      6a8cb18c-08a4-4010-bafb-67ee85779552.xls


        1      2       3         4         5      6      7        8       9         10     11             12   13
103   Table 2. Estimate Weibull cdf of time to failure of computer analog boards.
104                                                                   n(t)       d(t)    t
105   fail?=1time hrs a       sumlnxi      b          weiccdfKaplan-Meier2033         13       0
106   1      0      2.2E-06 -20.723 2.30259 1E-09            1
107        1     2 1.5E+00 0.69315 -0.077           2        1 0.9995
108        1    24 7.4E+00 3.17805 -0.3528         24 0.999 0.999        2020          1       2
109        1    70 1.5E+01 4.2485 -0.4713          70 0.998 0.9985       2023          1      24
110        1    72 1.5E+01 4.27667 -0.4744         72 0.998 0.9975       2020          1      70
111        1    72 1.5E+01 4.27667 -0.4744         72 0.998 0.9975       2019          2      72
112        1    90 1.7E+01 4.49981 -0.499          90 0.998 0.997        2018          1      90
113        1 192 2.7E+01 5.2575 -0.5823           192 0.997 0.9965       2015          1    192
114        1 330 3.9E+01 5.79909 -0.6415          330 0.996 0.996        2014          1    330
115        1 336 3.9E+01 5.81711 -0.6435          336 0.996 0.9951       2013          2    336
116        0 336 7.9E+04 5.81711 5.80424          336 0.996 0.9951
117   Ignore failure at time zero     -----------     --------        ----------
118        9 2034     79074 b(mle)= 0.62971 mode =           0
119          a = 1831571        b=     0.62971 Median##### hours
120   Weibull mean is b*Gamma(1/b+1) --> MTBF #####
121
122   Table 3. Estimate Weibull cdf of time to failure of computer logic boards.
123                                                                          n(t)       d(t)    t
124   fail?=1 time hrs a           sumlnxi         b                            2033
                                                              weiccdfKaplan-Meier            13      0
125        1       2 1.4E+00 0.69315 -0.063                 2        1 0.999
126        1       2 1.4E+00 0.69315 -0.063                 2        1 0.999    2020          2       2
127        1     13 3.8E+00 2.56495 -0.2329                13 0.999 0.9985      2019          1      13
128        1     24 5.2E+00 3.17805 -0.2885                24 0.999 0.998       2018          1      24
129        1     28 5.6E+00 3.3322 -0.3025                 28 0.999 0.9975      2017          1      28
130        1 140 1.3E+01 4.94164 -0.4477                  140 0.997 0.997       2016          1     140
131        1 160 1.4E+01 5.07517 -0.4597                  160 0.996 0.9965      2015          1     160
132        1 169 1.4E+01 5.1299 -0.4646                   169 0.996 0.996       2014          1     169
133        1 264 1.8E+01 5.57595 -0.5045                  264 0.995 0.9955      2013          1     264
134        1 330 2.0E+01 5.79909 -0.5244                  330 0.995 0.9951      2012          1     330
135        1 332 2.0E+01 5.80513 -0.5249                  332 0.995 0.9946      2011          1     332
136        0 336 4.1E+04 5.81711 5.80065                  336 0.995 0.9941      2010          1     336
137   ------- ------- ----------- ----------- -----------     --------       ----------
138      11 2132 41603.4 b(mle)= 0.51946 mode =                      0
139           a = 7715375             b=       0.51946 Median##### hours
140   Weibull mean is b*Gamma(1/b+1) --> MTBF #####
141
142   Table 4. Estimate Weibull cdf of time to failure of computer HDA, all brands.
143                                                                           n(t)       d(t)   t
144   fail?=1 time hrs a           sumlnxi         b          weiccdfKaplan-Meier2033         5      0
145        1       1 1.0E+00                0           0   1        1 0.9995
146        1     10 2.9E+00 2.30259 -0.2092                10 0.999 0.999        2028         1       1
147        1     65 6.9E+00 4.17439 -0.3788                65 0.999 0.9985       2027         1      10
148        1 148 1.0E+01 4.99721 -0.4531                  148 0.998 0.998        2026         1      65
149        1 160 1.0E+01 5.07517 -0.4601                  160 0.998 0.9975       2025         1     148
150        1 256 1.3E+01 5.54518 -0.5024                  256 0.997 0.997        2024         1     160
151        0 336 3.0E+04 5.81711 4.16717                  336 0.997 0.997        2023         1     256
152   ------- ------- ----------- ----------- -----------     --------        ----------
153        6 2032 29847.5 b(mle)= 0.46218 mode =                     0


                                                            Page 3
                                6a8cb18c-08a4-4010-bafb-67ee85779552.xls


       1      2      3      4        5      6     7      8       9     10   11   12   13
154        a=     1E+08    b=     0.46218 Median##### hours
155   Weibull mean is b*Gamma(1/b+1) --> MTBF #####
156
157   Table 5. Constants for an approximation to the gamma function
158   From Press et al "Numerical Recipes…"
159   gamma 5
160   cee0      1
161   cee1 76.2
162   cee2 -87
163   cee3    24
164   cee4 -1.2
165   cee5      0
166   cee6     -0
167   stp   2.51
168
169   Table 6. Regression estimates of Weibull parameters
170                                          Regression
171   fail?=1      R(t)
             time hrs      ln(t)  ln -lnR(t) b     a
172        1     1 0.99852       0 -6.5145 0.378 -6.37 <-coeffs
173        1     1 0.99852       0 -6.5145 0.012 0.048 <-stderr
174        1     1 0.99852       0 -6.5145 0.972 0.122 <-R^2,Sy
175        1     2 0.99704 0.69315 -5.8206 1044        30 <-F,DofF
176        1     2 0.99704 0.69315 -5.8206 15.45 0.444 <-RSS,SSE
177        1     2 0.99704 0.69315 -5.8206 beta alpha
178        1    10 0.99654 2.30259 -5.6662 0.378 #####
179        1    13 0.99605 2.56495 -5.5324
180        1    24 0.99457 3.17805 -5.2132
181        1    24 0.99457 3.17805 -5.2132
182        1    24 0.99457 3.17805 -5.2132
183        1    28 0.99408 3.3322 -5.1259
184        1    34 0.99358 3.52636 -5.0457
185        1    65 0.99309 4.17439 -4.9713
186        1    70 0.9926 4.2485 -4.9021
187        1    72 0.99161 4.27667 -4.7764
188        1    72 0.99161 4.27667 -4.7764
189        1    84 0.99112 4.43082 -4.719
190        1    88 0.99062 4.47734 -4.6647
191        1    90 0.99013 4.49981 -4.6131
192        1 113 0.98963 4.72739 -4.5641
193        1 140 0.98914 4.94164 -4.5173
194        1 148 0.98865 4.99721 -4.4726
195        1 160 0.98766 5.07517 -4.3887
196        1 160 0.98766 5.07517 -4.3887
197        1 169 0.98717 5.1299 -4.3493
198        1 192 0.98667 5.2575 -4.3113
199        1 256 0.98618 5.54518 -4.2747
200        1 264 0.98569 5.57595 -4.2393
201        1 330 0.9847 5.79909 -4.1721
202        1 332 0.98421 5.80513 -4.1401
203        1 336 0.98371 5.81711 -4.1091




                                                 Page 4
                    6a8cb18c-08a4-4010-bafb-67ee85779552.xls


     14   15   16         17        18       19        20      21   22   23
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51


                                    Page 5
                                    6a8cb18c-08a4-4010-bafb-67ee85779552.xls


          14            15       16           17                  18                         19          20         21          22         23
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63 n(t)           d(t)       t            P[DoA]
 64         2039           13          0   0.006376
                                                                                        1
 65                                        Kaplan-Meier
 66         2026            3          1   0.998519
 67         2023            3          2   0.997038                                  0.998
 68         2020            1        10    0.996545
 69         2019            1        13    0.996051                                  0.996
 70         2018            3        24    0.994571
 71         2015            1        28    0.994077
                                                          Reliability, P[Life > t]



                                                                                     0.994
 72         2014            1        34    0.993583
 73         2013            1        65      0.99309                                                                                 all
 74         2012            1        70    0.992596                                  0.992
                                                                                                                                     analog boards
 75         2011            2        72    0.991609
                                                                                                                                     logic boards
 76         2009            1        84    0.991115                                   0.99
 77         2008            1        88    0.990622                                                                                  HDA
 78         2007            1        90    0.990128                                  0.988
 79         2006            1       113    0.989635
 80         2005            1       140    0.989141
 81         2004            1       148    0.988648                                  0.986
 82         2003            2       160      0.98766
 83         2001            1       169    0.987167                                  0.984
 84         2000            1       192    0.986673
 85         1999            1       256      0.98618                                 0.982
 86         1998            1       264    0.985686
                                                                                             0            200             400
 87         1997            2       330    0.984699
 88         1995            1       332    0.984205                                               Age-at-failure, hours
 89         1994            1       336    0.983712
 90
 91                                                                                               Failure rates
 92
 93                             0.0015
 94
 95
 96
 97
 98 --------------               0.001
 99
                                                                                                                                                Empirical
100
101                                                                                                                                             Weibull

102
                               0.0005



                                                                           Page 6
                                        6a8cb18c-08a4-4010-bafb-67ee85779552.xls


         14           15           16           17      18          19           20     21         22         23
                                  0.0005
103
104                P[DoA]
105                0.006394
106
107   ln(1-F(t))   Kaplan-Meier         0
108    0.000495    0.999505                 0    50    100    150        200     250   300   350        400
109     0.00099    0.999011                                         Age, hours
110    0.001485    0.998516
111    0.002476    0.997527
112    0.002972    0.997033
113    0.003468    0.996538
114    0.003965    0.996043
115    0.004959    0.995054
116
117
118
119
120
121
122
123                P[DoA]
124                0.006394
125   ln(1-F(t))   Kaplan-Meier
126    0.000991      0.99901
127    0.001486    0.998515
128    0.001982      0.99802
129    0.002478    0.997525
130    0.002974    0.997031
131     0.00347    0.996536
132    0.003967    0.996041
133    0.004464    0.995546
134    0.004961    0.995051
135    0.005458    0.994557
136    0.005956    0.994062
137
138
139
140
141
142
143                P[DoA]
144                0.002459
145   ln(1-F(t))   Kaplan-Meier
146    0.000493    0.999507
147    0.000987    0.999014
148     0.00148    0.998521
149    0.001974    0.998028
150    0.002469    0.997535
151    0.002963    0.997041
152
153


                                                        Page 7
                     6a8cb18c-08a4-4010-bafb-67ee85779552.xls


      14   15   16         17        18       19        20      21   22   23
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203




                                     Page 8
          6a8cb18c-08a4-4010-bafb-67ee85779552.xls


     24
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51


                          Page 9
                         6a8cb18c-08a4-4010-bafb-67ee85779552.xls


                    24
       52
       53
       54
       55
       56
       57
       58
       59
       60
       61
       62
       63
       64
       65
       66
       67
       68
       69
       70
       71
       72
       73
       74
analog boards
       75
logic boards
       76
       77
       78
       79
       80
       81
       82
       83
       84
       85
       86
       87
       88
       89
       90
       91
       92
       93
       94
       95
       96
       97
       98
       99
        Empirical
      100
        Weibull
      101
      102


                                         Page 10
           6a8cb18c-08a4-4010-bafb-67ee85779552.xls


      24
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153


                           Page 11

								
To top