1 Introduction
2 Fail-safe design
x1 |
x2 | x3 | Effective volume | Compliance | |
---|---|---|---|---|---|
Standard | 7.07 | 0.00 | 7.07 | 1000.00 | 47.60 |
Failsafe | 5.22 | 5.22 | 5.22 | 1000.00 | 64.42 |
Failure 1 | 0.00 | 5.22 | 5.22 | 630.60 | 174.00 |
Failure 2 | 5.22 | 0.00 | 5.22 | 738.79 | 64.42 |
Failure 3 | 5.22 | 5.22 | 0.00 | 630.60 | 174.00 |
3 Fail-safe formulation for general 3D structural continuum
4 Solution strategy for fail-safe topology optimization
4.1 Effect of damage population size
-
Damage Series A:(a)Level 1: A total number of N D damage cubes of size d are distributed evenly to cover the entire structural domain Ω. Grey cubes in Fig. 5 represent the base Level 1 damage population. The centers of damages are evenly spread in Ω, with diagonal distance \( \sqrt{3}d \) between neighboring cubes. We denote the population size of Level 1 at PA1 = N D .(b)Level 2: Next we aim to double the density of damage zones to create evenly spread damage zones with distance between neighbors reduced to \( \sqrt{3}d/2 \). To achieve that we only need to double the grid points of cube centers along X, Y and Z, resulting in a damage population PA 2 = 23 × N D = 8 × N D .(c)Level L: We aim to double the evenly spread damage population density from Level (L−1), producing a damage population PA L = 23(L − 1) × N D . It can be easily established that for a 3D domain the total population of level L is always eight times the population of the previous level, i.e., PA L = 8 × PA (L − 1). Thus the increase of population from one level to the next is always ΔPA L = 7 × PA (L − 1).Now we construct a slight variation of the above damage series, termed PB, as a partial set of PA at all levels except Level 1. We will show later that PB has some interesting characteristics.
-
Damage Series B:(a)Level 1: The damage population starts exactly the same as PA, i.e., PB1 = PA1.(b)Level 2: We keep only a subset of Level 2 population in PA. PA2 can be constructed by moving seven copies of PA1 into bisection combinations along XYZ. For PB we only keep the copy moving diagonally in space, shown in red color in Fig. 5. The resulting damage population is PB 2 = 2 × PA 1 = PA 2/4.(c)Level L: Following the logic in (b) we construct PBL as PA(L−1) enriched with its copy shifted diagonally into bisection location in space, i.e. in all three dimensions XYZ. The total damage population is PB L = 2 × PA (L − 1) = PA L /4. Therefore, the population size of the partial set for PB is only a quarter of the complete set of PA at any level in the series.
Damage level | Damage population xN
D
| Maximum survival rate of a refresentative cube | ||||
---|---|---|---|---|---|---|
PA | PB | PA sectional | PB sectional | PA volumetric | PB volumetric | |
1 | 1 | 1 | 75.0 % | 75.0 % | 87.5 % | 87.5 % |
2 | 8 | 2 | 43.8 % | 50.0 % | 57.8 % | 62.5 % |
3 | 64 | 16 | 23.4 % | 25.0 % | 33.0 % | 34.4 % |
4 | 512 | 128 | 12.1 % | 12.5 % | 17.6 % | 18.0 % |
5 | 4096 | 1024 | 6.2 % | 6.3 % | 9.1 % | 9.2 % |
… | … | … | … | … | … | |
L
|
P
L
= 23(L − 1)
| 2 × P
(L − 1)
|
\( \frac{2\times {2}^n-1}{2^{2n}} \)
| 2/2
n
|
\( \frac{3\times \left({2}^{2n}-{2}^n\right)+1}{2^{3n}} \)
|
\( \frac{3\times {2}^n-2}{2^{2n}} \)
|
L → ∞ | ∞ | ∞ | 2/2
n
| 2/2
n
| 3/2
n
| 3/2
n
|
4.2 Effect of member cross-section length scale
Damage level | Damage population xN
D
| Maximum survival of a representative cube | ||||
---|---|---|---|---|---|---|
PA | PB | DS-A sectional | DS-B | DS-A volumetric | DS-B volumetric | |
1 | 1 | 1 | 75.0 % | 75.0 % | 87.5 % | 87.5 % |
2 | 8 | 2 | 0.0 % | 50.0 % | 0.0 % | 50.0 % |
4.3 Practical considerations on damage population size
5 Computational scheme
-
Damage zones containing any point load are eliminated to preserve load conditions. Partial elimination of distributed loads by a damage zone is considered acceptable.
-
If a damage zone increases the compliance by a significant margin compared to that of the undamaged structure at the start, the process terminates. Such case indicates that the structure’s function depends on a narrow pathway that doesn’t allow redundancy to be built. The margin threshold can be defined by the user and 10 times compliance increase is set as default.
-
For reducing computation cost, a threshold on the material volume inside a damage cube can be applied to reduce the total damage population. Ten percent threshold is used for numerical examples in this paper. In addition for second layer of damage cubes added by PA2 or PB2 we only include those that are fully inside the structural domain.
6 Numerical examples
6.1 Example 1: rectangular plate under shear force
Standard | Failsafe PA1
| Failure Zones | |||
---|---|---|---|---|---|
1 | 2 | 3 | 4 | ||
58.72 | 84.28 | 161.50 |
183.10
|
183.10
| 161.50 |
5 | 6 | 7 | 8 | ||
161.50 |
184.02
|
184.02
| 161.50 |
Standard | Failsafe PB2
| Failure Zones | |||
---|---|---|---|---|---|
1 | 2 | 3 | 4 | ||
58.72 | 82.96 | 164.46 |
184.44
|
184.44
| 164.46 |
5 | 6 | 7 | 8 | ||
165.04 |
186.34
|
186.34
| 165.04 | ||
9 | 10 | 11 | |||
165.08 |
183.44
| 165.08 |
Unrestricted | Enforced symmetry | |||
---|---|---|---|---|
PA1
| PB2
| PA1
| PB2
| |
Iteration | 55 | 61 | 59 | 59 |
Compliance | 87.18 | 86.94 | 87.20 | 89.56 |
Max comp | 140.18 | 155.84 | 164.04 | 171.42 |
6.2 Example 2: rectangular plate under bending force
Design | Damage size | Population | Compliance | Max Comp | Active Zone | Iteration |
---|---|---|---|---|---|---|
Fig. 16(a)
| none | 202 | ||||
Fig. 16(b)
| none | 222 | 42 | |||
Fig. 18(a)
| 10 | 7701 | 239 | 352 | ||
Fig. 18(b)
| 10 | 108 | 300 | 405 | 30 | 88 |
Fig. 18(c)
| 10 | 193 | 304 | 413 | 42 | 79 |
Fig. 20(a)
| 22 | 5421 | 304 | 714 | ||
Fig. 20(b)
| 22 | 26 | 350 | 644 | 18 | 85 |
Fig. 20(c)
| 22 | 42 | 358 | 656 | 20 | 86 |
Fig. 21
| 22 | 42 | 348 | 643 | 20 | 80 |
6.3 Example 3: 3D control arm
7 Conclusion
-
The fail-safe topology optimization concept and formulation is developed with reference to well established fail-safe design approach vital to aircraft design. However, the general approach should be applicable to any industry where structural failure can result in catastrophic accident. We only established the basic concept with generic sphere and cube shaped damages. In practice damage can occur in different forms. For example a ballistic impact could cause a penetration of a given shape and size. It should be an interesting research topic to explore more broad damage scenarios found in engineering practice of various industries.
-
From the perspective of confidence quantification of the failure representation, we only provided quantitative study about maximum material survival rate of a representative member section for a given damage population. Statistics based study on reliability of the entire structural system under a given damage population can provide further insights.
-
Some basic problems studied herein can be interesting for mathematicians as well. We used an engineering approach to establish exact survival properties under a given damage population through analyzing geometric interactions of a representative member section with intersecting damage cubes. It can be of theoretical value to provide formal mathematical proofs of the findings presented. Also it can be an interesting problem to derive geometric properties for sphere and other damage shapes. Another very challenging theoretical problem is to study the uniqueness and global optimality of fail-safe topology. Though the solution is not unique under finite damage population, the limiting case of infinite damage population seems unique by definition. Mathematical insights into this complex problem can be highly valuable from both theoretical and practical perspectives.