Professional Documents
Culture Documents
Structural Damage Detection Employing Best Achievable Eigenvectors in Bayesian Framework
Structural Damage Detection Employing Best Achievable Eigenvectors in Bayesian Framework
2 Bayesian Framework
4 ABSTRACT
5 Efficiency of a Bayesian model updating algorithm is greatly affected by the choice of vari-
6 ance of prediction error models of different data points (evidence) used for model updating. A
7 sensitivity based novel approach is proposed in this work to find these variances without increas-
8 ing the dimensionality of the model updating problem. Well-established relations of modal data
9 sensitivity towards structural parameters are incorporated in Bayesian framework to evaluate the
10 prediction error variances. A high rise shear building is considered for the numerical illustration
11 of the approach. Markov chain Monte Carlo (MCMC) simulation technique is employed using
12 Metropolis-Hastings algorithm to simulate the samples from the posterior distribution. Results are
13 presented as a comparison of unknown parameters obtained using the present approach and an
14 approach in which all prediction error variances are assumed to be equal. The study shows that the
15 proposed approach is highly efficient in extracting information from the data appropriately. It is
16 also illustrated that damage locations also play an important role in selection of variances of pre-
17 diction error model . Each data point in the evidence can be very effective in estimation of model
1
PhD Candidate, Department of Civil Engineering, IIT Kanpur, Kanpur, UP-208016, India
2
Associate Professor, Dept. of Civil Eng., Indian Institute of Technology, Kanpur, U.P. - 208016, India.
1
20 THEORETICAL FORMULATION
22 For an n − DOF structural dynamic system the equation of motion can be given as:
M ẍ + Kx = f (t) (1)
23 where K and M are the n × n stiffness and mass matrix of the structure, respectively, and x
24 and f (t) represent the displacement and applied force vectors respectively. The associated eigen-
Kφ = M φ∧ (2)
27 of eigenvalues of the system. If Ki and Mi represent the element stiffness and mass matrix
28 respectively for the ith element. Then, for a system of p number of total elements the system
p p
X X
K= Ki and M= Mi (3)
i=1 i=1
30 Now, for a damaged system with damage confined to the loss of stiffness only (i.e. system
31 mass matrix is intact) the modified stiffness matrix can be written as:
p
X
Kd = Ku + ai K i (4)
i=1
2
32 here, subscripts d and u are used to represent the damaged and undamaged states of the
33 structure respectively. It is clear from Equation 4 that the system stiffness in damaged state is
34 the sum of system stiffness in undamaged state and the reduced stiffness due to the damage. The
35 reduced stiffness of the structure is calculated by imposing a stiffness reduction factor on each
36 of the structural element. The value of stiffness reduction factor varies between 0 to −1 for no
p
X
ai Ki φt = M φt ∧t −Ku φt (5)
i=1
39 where, subscript t denotes the measured data. For a j th measured mode Equation 5 can be
40 rewritten as:
p
X
ai Ej−1 Ki φtj = φtj (6)
i=1
41 where,
2
Ej = (ωtj M − Ku ) (7)
2
42 In Equation 6 φtj represents the eigenvector and ωtj in Equation 1 represents the eigenvalue
2
43 of the damaged structure. It is to be noted that for ωtj ≈ ωj2 (frequency of the intact structure)
44 matrix Ej may appear as ill-conditioned and application of the algorithm may become limited.
3
p
X
ai Aij φtj = φtj (8)
i=1
46 being
Pp {1} Pp {2} Pp {n}
Sj = i=1 ai {Aij } i=1 ai {Aij } ... i=1 ai {Aij }
(10)
48 here, superscript of{Aij } represent the corresponding column of matrix Aij . Equation 8 now
50 Equation 11 shows that φtj is a linear combination of columns of matrix Sj that means φtj lies
51 in the subspace defined by the columns of matrix Sj . Further, for similar reasons it can be observed
52 from Equation 10 that each k th column of matrix Sj , k = 1...n spans in a subspace which includes
53 the k th column of each matrix Aij . Therefore, each independent column out of these p number
54 of columns will serve as the basis for that subspace. Further, it can be concluded that φtj lies in
55 a subspace which is formed by the union of all the subspaces in which each column of matrix Sj
56 spans. It is to be noted that this united subspace is independent of the value of stiffness reduction
57 factors ai i = 1...p. Therefore, if out of p number of elements, let us assume, for example, that 2th
58 5th and 9th elements are damaged, p >= 9, then the subspace to which φtj belongs is formed by
4
Xj = A2j |A5j |...|A9j (12)
60 It is to be noted further that owing to the rank deficiency of matrix Aij (being the element
61 stiffness matrix), each column of matrix Xj is not an independent column. This may result in
62 a repeating subspace for one or more columns of matrix Sj . However, the resulting union of all
64 The above discussion draws to the conclusion that if damage in a structure is caused by loss of
65 the stiffness in some of the elements and if each of these stiffness loss is reflected in the measured
66 mode φtj , then φtj will lie exactly in the subspace defined by the columns of augmented matrix
67 Xj . This augmented matrix Xj consists of all Aij , where subscript i belongs to only damaged
68 elements of the structure. The best achievable eigenvector of a subspace occupied by columns of
71 elements, while forming the matrix X matches exactly with the actual damaged elements in the
72 structure, φatj will be identical to φtj . However, if this is not the case, then the achieved φatj for the
73 considered subspace will not match with the measured φtj . This fact makes the foundation of dam-
74 age localization algorithm discussed in this work. The following subsection describes the Bayesian
75 framework for the best achievable eigenvector based damage localization algorithm adopted in this
76 work.
5
77 Bayesian statistical framework
78 Bayesian statistical framework allows to update the probability of a belief or hypothesis about
79 a physical system based on some observation from the considered system itself employing Bayes’
80 theorem. For this purpose, first the physical system is mathematically modelled with some pa-
81 rameters, each unique combination of these parameters defines a unique model for the system. If
82 there is some specific reason to give priority to a particular model over others it can be achieved
83 by assigning a higher prior probability to that model over others. To know which of these models
84 represents the physical system at the best, a response collected from the system is used to compare
86 Mathematically, let D is the available data from the system and each model M θ is defined by
87 a parameter vector θ. Then, the probability of each belief of M θ is given using Bayes’ theorem
88 as:
θ p D|M θ p(M θ )
p M |D = (14)
p(D)
89 Here, expression p(M θ ) represents the belief of M θ without knowing the evidence and
90 known as prior distribution of M θ . Expression p D|M θ represents the probability of data D
91 when a belief of M θ is taken as true and called as the likelihood of data D for that belief. It is
92 to be noted that the total probability of data D is a constant, which is nothing but the sum of the
93 likelihood of data D for each and every belief of M θ and represented by p(D) (evidence). The left
95 the shape of the posterior distribution is independent of p(D). The posterior distribution gives the
6
97 a belief in posterior distribution represents its ability to produce the response at a comparatively
98 less deviation from the data D. The next subsection describes how to form likelihood function of
99 Bayesian framework for best achievable eigenvectors based damage localization algorithm.
100 Likelihood function for best achievable eigenvector based damage localization algorithm
101 As discussed in section 1 the concept of best achievable eigenvector of a space or subspace
102 can be used to determine the location of damage in a structure. This needs at first the modal data
103 (frequencies and mode shapes) of a structure for which the damage identification (localization) is
104 required. This can be achieved easily owing to the recent advancements in extracting modal data
105 of a system from its dynamic response and then applying any of the available modal identification
106 algorithms. Now, let us assume that for a model M θ , where parameter vector θ differs in terms of
Mθ
107 number and location of damaged elements for each model, the best achievable eigenvector is φtj a
108 corresponding to the measured eigenvector φtj of the system for j th mode. Then, the relationship
Mθ
109 between the the best achievable eigenvector φtj a for model M θ and the measured eigenvector φtj
Mθ
φtj = φtj a + e(M θ ) (15)
111 Here, e(M θ ) is the deviation of model output from the system output and often called as
112 prediction error. The likelihood of Y is expected to attain maximum for a realization of M θ when
113 e(θ) is minimum. Also, considering the fact that e(θ) can be spread on both positive and negative
114 sides of the number scale, a normal distribution with zero mean (µ) vector and a covariance matrix
115 (V ∈ Rn×n ) gives the maximum entropy for the choice of prediction error model [? ]. Therefore,
116 the likelihood of a measured mode φtj for a realization of model M θ can be written as:
7
θ
1 1 Maθ T −1 Maθ
p φtj |M , V φtj = 1/2
exp − (φtj − φtj ) V φtj (φtj − φtj ) (16)
(2π)n/2 V φtj 2
117 V φi represents the covariance matrix of the prediction errors in the ith mode shape vector
118 components. Now, if the available data D from the system contains m measured modes and s sets
s Y
Y m
p(D/θ) = p(φqr |θ, Vφqr ) (17)
q=1 r=1
120 In writing Equation 17, it is assumed that the different data sets are statically independent, in-
121 formatively. Further, it is assumed that various modes are also statically independent informatively
122 to each other. Now, to evaluate the covariance matrix V φi , mode shape components are taken as
123 uncorrelated to each other, resulting V φi to be a diagonal matrix. It is shown in Section 2 that how
124 matrix V φi has been formed in this study for different modes.
126 The term p(M θ ) in Equation 14 is known as the prior distribution for unknown models M θ .
127 The choice of prior distribution for unknown parameters or models facilitates to incorporate all
128 the available information about unknown parameters in Bayesian updating algorithm. This helps
129 to achieve a less scattered and more concentrated posterior distribution near the true value of un-
130 known parameter. In this work prior distribution of unknown model is also evaluated using the best
131 achievable eigenvector concept. In BAE based damage localization approach comparing models
132 with different number of damaged elements simultaneously can mislead the damage localization
133 in structure. This is because the BAE approach works on finding a subspace which has its best
8
134 achievable eigenvector same as the measured eigenvector of the structure. However, it is to be
135 noted that once the correct damaged elements has been identified then the BAE of a subspace
136 formed by the damaged elements and any extra element other than the damaged elements will
137 also match the measured eigenvector of the structure. This issue has been resolved in subsequent
138 sections to identify the correct location of damage. However, the damage algorithm adopted in
139 this work considers the same number of damaged elements model at a time to find out the best
140 model out of these. Therefore, let us assume that out of p number of elements, a model class is
141 considered with only r number of damaged elements. The total number of unique combinations of
p!
Crp = (18)
r! (p − r)!
143 Now, for these Crp number of unique combinations, all possible unique damage locations are
144 considered. It is to be noted that in each of these combination the number of damaged elements is
145 same and these combinations are unique in regard of the damage locations. Each of these unique
146 combination forms a unique model whose posterior probability is to be determined based on the
147 available modal data from the structure. To get the prior probability of each model, firstly, the
148 likelihood of available data for each damage location is found out separately using Equation 17.
149 Then, the prior probability of each unique model is evaluated by multiplying the likelihood of
150 available data for each damage location in a combination and then normalizing it. For example, if
151 out of p number of elements, let us assume, that 2th 5th and 9th elements are damaged, p >= 9.
152 Then the joint likelihood of available data for model M θ , where θ ∈ {2, 5, 9} is evaluated as:
9
p(D/M θ ) = p(D/M {2} ) × p(D/M {5} ) × p(D/M {9} ) (19)
153 While evaluating Equation 19, statistical independence is assumed between different terms
154 on the left hand side. To form the prior distribution it is assumed that prior probability of a model
155 M θ is proportional to the joint likelihood of the available data for the model M θ as evaluated in
p(D/M θ )
p(M θ ) = (20)
Crp
P
p(D/M θi )
i=1
158 This section describes the algorithm adopted in this work for damage localization using best
161 • Evaluate total number of combinations for considered number of damaged elements in the
163 • Form all the unique combinations for the considered number of damaged elements then
164 assign a unique number to each unique combination for identification purpose of this com-
165 bination
167 • For each combination evaluate the best achievable eigenvector using Equation ?? corre-
169 • Evaluate the likelihood of the available data for each unique combination using Equation ??
10
170 • Get the posterior probability of each combination according to Equation ??
171 • Repeat point 2to6 for each number of damaged elements i.e. for i=1...p.
172 • Collect the most probable combination from posterior distribution of each number of dam-
174 • Plot the likelihood of each most probable combination against the number of damaged
176 • A sharp change in the above plot shows the exact number of damaged elements in the
177 structure and the associated combination shows the location of damage.
179 To study the efficiency of the proposed approach in damage localization, a 12-storey shear
180 building frame adopted from ? ] is considered. Figure ?? provides a schematic diagram of the
181 adopted shear building. For the intact structure the value of mass and stiffness parameters are
182 adopted as 1.0 × 105 kg and 2.0 × 108 N/mm, respectively [? ] for each of the element. In
183 damaged model it is assumed that the mass parameters associated with each degree-of-freedom
184 of the structure mi (with i = 1, 2, ...12) is known and invariable. However, stiffness of some of
185 the elements of the intact structure are degraded using a stiffness reduction factor as discussed in
186 Section 1 (Equation 4). These damaged elements are then found using the BAE approach discussed
187 in this work under Bayesian inference. For this purpose, modal data of the damaged structure is
188 found out using Eigenvalue analysis to simulate the evidence in the Bayesian framework. To
189 represent a realistic scenario, the data are contaminated with different noise levels. For example
190 to represent 5% noise level, the data are contaminated by a noise of coefficient of variation 5%. A
191 total of 15 such data sets are taken from this noisy data to localize damage in the structure.
11
192 Three different noise levels namely 0%, 2% and 5% for two levels of damage in the structure
193 are considered. The two levels of damage in the structure are defined as 20% and 40% reduction in
194 the stiffness of structural elements. The example presented here considers damage in four elements
195 of a 12-storey shear building frame. The damage is considered in the second, third, fourth and sixth
196 storey of 12-storey shear building frame for illustration purpose. Only fundamental mode shape
197 data is used to localize damage in the structure. At first, 0% noise case is considered to see the
198 working of present algorithm in localizing damage under idealistic conditions. Since, both the
199 number of damaged elements in the structure and its location is unknown, so, it is necessary first
200 to evaluate the exact number of damaged elements and then finding the most probable damage
201 combination for this number. Figure ?? shows the normalized likelihood (normalized with respect
202 to maximum likelihood) versus number of elements plot for 0% noise case for five most probable
203 damage locations combinations of each considered number of damaged elements. The considered
204 number of damaged elements ranges from 1 to p − 1 (p being the number of elements). Although,
205 it can be easily extended up to p but there is no gain to apply the present algorithm when all
206 elements are damaged in a structure so this case has been excluded here. Figure ??(a) refers to the
207 case of 20% damage in the elements, whereas, Figure ??(b) shows the case of 40% damage in the
208 elements. To evaluate the exact number of damaged elements it is assessed that for which least
209 number of damaged elements the normalized likelihood becomes one for most probable damage
213 CONCLUSIONS
12
TABLE 1. The unique number associated to the unique combination of 4 damaged
elements
No. Comb. No. Comb. No. Comb. No. Comb.
1 9 10 11 12 41 2 4 11 12 81 1 2 10 12 121 4 5 8 12
2 8 10 11 12 42 1 4 11 12 82 7 8 9 12 122 3 5 8 12
3 7 10 11 12 43 2 3 11 12 83 6 8 9 12 123 2 5 8 12
4 6 10 11 12 44 1 3 11 12 84 5 8 9 12 124 1 5 8 12
5 5 10 11 12 45 1 2 11 12 85 4 8 9 12 125 3 4 8 12
6 4 10 11 12 46 8 9 10 12 86 3 8 9 12 126 2 4 8 12
7 3 10 11 12 47 7 9 10 12 87 2 8 9 12 127 1 4 8 12
8 2 10 11 12 48 6 9 10 12 88 1 8 9 12 128 2 3 8 12
9 1 10 11 12 49 5 9 10 12 89 6 7 9 12 129 1 3 8 12
10 8 9 11 12 50 4 9 10 12 90 5 7 9 12 130 1 2 8 12
11 7 9 11 12 51 3 9 10 12 91 4 7 9 12 131 5 6 7 12
12 6 9 11 12 52 2 9 10 12 92 3 7 9 12 132 4 6 7 12
13 5 9 11 12 53 1 9 10 12 93 2 7 9 12 133 3 6 7 12
14 4 9 11 12 54 7 8 10 12 94 1 7 9 12 134 2 6 7 12
15 3 9 11 12 55 6 8 10 12 95 5 6 9 12 135 1 6 7 12
16 2 9 11 12 56 5 8 10 12 96 4 6 9 12 136 4 5 7 12
17 1 9 11 12 57 4 8 10 12 97 3 6 9 12 137 3 5 7 12
18 7 8 11 12 58 3 8 10 12 98 2 6 9 12 138 2 5 7 12
19 6 8 11 12 59 2 8 10 12 99 1 6 9 12 139 1 5 7 12
20 5 8 11 12 60 1 8 10 12 100 4 5 9 12 140 3 4 7 12
21 4 8 11 12 61 6 7 10 12 101 3 5 9 12 141 2 4 7 12
22 3 8 11 12 62 5 7 10 12 102 2 5 9 12 142 1 4 7 12
23 2 8 11 12 63 4 7 10 12 103 1 5 9 12 143 2 3 7 12
24 1 8 11 12 64 3 7 10 12 104 3 4 9 12 144 1 3 7 12
25 6 7 11 12 65 2 7 10 12 105 2 4 9 12 145 1 2 7 12
26 5 7 11 12 66 1 7 10 12 106 1 4 9 12 146 4 5 6 12
27 4 7 11 12 67 5 6 10 12 107 2 3 9 12 147 3 5 6 12
28 3 7 11 12 68 4 6 10 12 108 1 3 9 12 148 2 5 6 12
29 2 7 11 12 69 3 6 10 12 109 1 2 9 12 149 1 5 6 12
30 1 7 11 12 70 2 6 10 12 110 6 7 8 12 150 3 4 6 12
31 5 6 11 12 71 1 6 10 12 111 5 7 8 12 151 2 4 6 12
32 4 6 11 12 72 4 5 10 12 112 4 7 8 12 152 1 4 6 12
33 3 6 11 12 73 3 5 10 12 113 3 7 8 12 153 2 3 6 12
34 2 6 11 12 74 2 5 10 12 114 2 7 8 12 154 1 3 6 12
35 1 6 11 12 75 1 5 10 12 115 1 7 8 12 155 1 2 6 12
36 4 5 11 12 76 3 4 10 12 116 5 6 8 12 156 3 4 5 12
37 3 5 11 12 77 2 4 10 12 117 4 6 8 12 157 2 4 5 12
38 2 5 11 12 78 1 4 10 12 118 3 6 8 12 158 1 4 5 12
39 1 5 11 12 79 2 3 10 12 119 2 6 8 12 159 2 3 5 12
40 3 4 11 12 80 1 3 10 12 120 1 6 8 12 160 1 3 5 12
13
TABLE 2. The unique number associated to the unique combination of 4 damaged
elements
No. Comb. No. Comb. No. Comb. No. Comb.
161 1 2 5 12 201 1 2 10 11 241 4 5 8 11 281 1 2 5 11
162 2 3 4 12 202 7 8 9 11 242 3 5 8 11 282 2 3 4 11
163 1 3 4 12 203 6 8 9 11 243 2 5 8 11 283 1 3 4 11
164 1 2 4 12 204 5 8 9 11 244 1 5 8 11 284 1 2 4 11
165 1 2 3 12 205 4 8 9 11 245 3 4 8 11 285 1 2 3 11
166 8 9 10 11 206 3 8 9 11 246 2 4 8 11 286 7 8 9 10
167 7 9 10 11 207 2 8 9 11 247 1 4 8 11 287 6 8 9 10
168 6 9 10 11 208 1 8 9 11 248 2 3 8 11 288 5 8 9 10
169 5 9 10 11 209 6 7 9 11 249 1 3 8 11 289 4 8 9 10
170 4 9 10 11 210 5 7 9 11 250 1 2 8 11 290 3 8 9 10
171 3 9 10 11 211 4 7 9 11 251 5 6 7 11 291 2 8 9 10
172 2 9 10 11 212 3 7 9 11 252 4 6 7 11 292 1 8 9 10
173 1 9 10 11 213 2 7 9 11 253 3 6 7 11 293 6 7 9 10
174 7 8 10 11 214 1 7 9 11 254 2 6 7 11 294 5 7 9 10
175 6 8 10 11 215 5 6 9 11 255 1 6 7 11 295 4 7 9 10
176 5 8 10 11 216 4 6 9 11 256 4 5 7 11 296 3 7 9 10
177 4 8 10 11 217 3 6 9 11 257 3 5 7 11 297 2 7 9 10
178 3 8 10 11 218 2 6 9 11 258 2 5 7 11 298 1 7 9 10
179 2 8 10 11 219 1 6 9 11 259 1 5 7 11 299 5 6 9 10
180 1 8 10 11 220 4 5 9 11 260 3 4 7 11 300 4 6 9 10
181 6 7 10 11 221 3 5 9 11 261 2 4 7 11 301 3 6 9 10
182 5 7 10 11 222 2 5 9 11 262 1 4 7 11 302 2 6 9 10
183 4 7 10 11 223 1 5 9 11 263 2 3 7 11 303 1 6 9 10
184 3 7 10 11 224 3 4 9 11 264 1 3 7 11 304 4 5 9 10
185 2 7 10 11 225 2 4 9 11 265 1 2 7 11 305 3 5 9 10
186 1 7 10 11 226 1 4 9 11 266 4 5 6 11 306 2 5 9 10
187 5 6 10 11 227 2 3 9 11 267 3 5 6 11 307 1 5 9 10
188 4 6 10 11 228 1 3 9 11 268 2 5 6 11 308 3 4 9 10
189 3 6 10 11 229 1 2 9 11 269 1 5 6 11 309 2 4 9 10
190 2 6 10 11 230 6 7 8 11 270 3 4 6 11 310 1 4 9 10
191 1 6 10 11 231 5 7 8 11 271 2 4 6 11 311 2 3 9 10
192 4 5 10 11 232 4 7 8 11 272 1 4 6 11 312 1 3 9 10
193 3 5 10 11 233 3 7 8 11 273 2 3 6 11 313 1 2 9 10
194 2 5 10 11 234 2 7 8 11 274 1 3 6 11 314 6 7 8 10
195 1 5 10 11 235 1 7 8 11 275 1 2 6 11 315 5 7 8 10
196 3 4 10 11 236 5 6 8 11 276 3 4 5 11 316 4 7 8 10
197 2 4 10 11 237 4 6 8 11 277 2 4 5 11 317 3 7 8 10
198 1 4 10 11 238 3 6 8 11 278 1 4 5 11 318 2 7 8 10
199 2 3 10 11 239 2 6 8 11 279 2 3 5 11 319 1 7 8 10
200 1 3 10 11 240 1 6 8 11 280 1 3 5 11 320 5 6 8 10
14
TABLE 3. The unique number associated to the unique combination of 4 damaged
elements
No. Comb. No. Comb. No. Comb. No. Comb.
321 4 6 8 10 361 2 4 5 10 401 2 4 7 9 441 4 5 6 8
322 3 6 8 10 362 1 4 5 10 402 1 4 7 9 442 3 5 6 8
323 2 6 8 10 363 2 3 5 10 403 2 3 7 9 443 2 5 6 8
324 1 6 8 10 364 1 3 5 10 404 1 3 7 9 444 1 5 6 8
325 4 5 8 10 365 1 2 5 10 405 1 2 7 9 445 3 4 6 8
326 3 5 8 10 366 2 3 4 10 406 4 5 6 9 446 2 4 6 8
327 2 5 8 10 367 1 3 4 10 407 3 5 6 9 447 1 4 6 8
328 1 5 8 10 368 1 2 4 10 408 2 5 6 9 448 2 3 6 8
329 3 4 8 10 369 1 2 3 10 409 1 5 6 9 449 1 3 6 8
330 2 4 8 10 370 6789 410 3 4 6 9 450 1 2 6 8
331 1 4 8 10 371 5789 411 2 4 6 9 451 3 4 5 8
332 2 3 8 10 372 4789 412 1 4 6 9 452 2 4 5 8
333 1 3 8 10 373 3789 413 2 3 6 9 453 1 4 5 8
334 1 2 8 10 374 2789 414 1 3 6 9 454 2 3 5 8
335 5 6 7 10 375 1789 415 1 2 6 9 455 1 3 5 8
336 4 6 7 10 376 5689 416 3 4 5 9 456 1 2 5 8
337 3 6 7 10 377 4689 417 2 4 5 9 457 2 3 4 8
338 2 6 7 10 378 3689 418 1 4 5 9 458 1 3 4 8
339 1 6 7 10 379 2689 419 2 3 5 9 459 1 2 4 8
340 4 5 7 10 380 1689 420 1 3 5 9 460 1 2 3 8
341 3 5 7 10 381 4589 421 1 2 5 9 461 4 5 6 7
342 2 5 7 10 382 3589 422 2 3 4 9 462 3 5 6 7
343 1 5 7 10 383 2589 423 1 3 4 9 463 2 5 6 7
344 3 4 7 10 384 1589 424 1 2 4 9 464 1 5 6 7
345 2 4 7 10 385 3489 425 1 2 3 9 465 3 4 6 7
346 1 4 7 10 386 2489 426 5 6 7 8 466 2 4 6 7
347 2 3 7 10 387 1489 427 4 6 7 8 467 1 4 6 7
348 1 3 7 10 388 2389 428 3 6 7 8 468 2 3 6 7
349 1 2 7 10 389 1389 429 2 6 7 8 469 1 3 6 7
350 4 5 6 10 390 1289 430 1 6 7 8 470 1 2 6 7
351 3 5 6 10 391 5679 431 4 5 7 8 471 3 4 5 7
352 2 5 6 10 392 4679 432 3 5 7 8 472 2 4 5 7
353 1 5 6 10 393 3679 433 2 5 7 8 473 1 4 5 7
354 3 4 6 10 394 2679 434 1 5 7 8 474 2 3 5 7
355 2 4 6 10 395 1679 435 3 4 7 8 475 1 3 5 7
356 1 4 6 10 396 4579 436 2 4 7 8 476 1 2 5 7
357 2 3 6 10 397 3579 437 1 4 7 8 477 2 3 4 7
358 1 3 6 10 398 2579 438 2 3 7 8 478 1 3 4 7
359 1 2 6 10 399 1579 439 1 3 7 8 479 1 2 4 7
360 3 4 5 10 400 3479 440 1 2 7 8 480 1 2 3 7
15
TABLE 4. The unique number associated to the unique combination of 4 damaged
elements
No. Comb. No. Comb. No. Comb. No. Comb.
481 3456 485 1356 489 1 2 4 6 493 1245
482 2456 486 1256 490 1 2 3 6 494 1235
483 1456 487 2346 491 2 3 4 5 495 1234
484 2356 488 1346 492 1345
16