NAME: Valentin Borisavljevic ID: 2140173 Problem: Brake calliper Method: D2

Application of the method to the problem and Results This report concerns the optimization problem of a 1
̂𝑋 = 𝑋̂ 𝑇 𝑋̂, with X = rows of
the D-criterium for the sample space. The matrix 𝑀 𝑁
specific mechanical structure. The problem can be described in five steps: 2 2
- Definition: The structure is a brake calliper, a specific part of the brake mechanism of a bike. [1 𝑥𝑖,1 … 𝑥𝑖,𝑘 𝑥𝑖,1 … 𝑥𝑖,𝑘 𝑥𝑖,1 𝑥𝑖,2 … 𝑥𝑖,𝑘−1 𝑥𝑖,𝑘 ] for each i-th of the N experiments (xi,j = values of design
It is made of Aluminium (5000 or 6000 series). It is confronted to a certain load: Fc (-y-direction), Fw (z- factor j in the i-th out of N experiments) and its det(-1/p) (p=(1/2)(k+1)(k+2) ) are calculated for each sample
direction), FR (-x-direction) and reaction forces in all directions (Rx, Ry and Rz). Figure 1 shows the free- space, and the sample space with the minimal value of det(-1/p) is defined as the optimal one. Furthermore,
body diagram of the brake calliper, with a reference frame xyz. Figure 2 shows the dimension quantities of the report looks into the use of the least square method (LSM) to produce the analytical function from the
the structure. The goals are a minimal mass and a minimal cost of the structure. sample space. The function is 𝑦 = 𝛽0 + 𝛽1 𝑥1 + ⋯ . +𝛽𝑘 𝑥𝑘 , and the ᵝ coefficients are derived from the
- Information collection: the properties of Aluminium 5000 series are: Young’s modulus of 70 GPa, a sample space: 𝛽⃗ = (𝑋̂ 𝑇 𝑋̂)−1 𝑋̂ 𝑇 𝑦⃗, with 𝑦⃗ = vector of the response variable values for each of the N
yield stress of 170 MPa, a density of 2700 kg/m3 and a cost of £3/kg. 6000 has a different yield experiments, and 𝑋̂ = matrix rows of [1 𝑥𝑖,1 … 𝑥𝑖,𝑘 ] for each i-th of the N experiments (Cavazutti 2013,
stress (250 MPa) and cost (£3.2/kg). pages 39-46).
A first important aspect of the structure is the internal stresses within it. From the points of actions and
orientations of the forces acting on the structure, certain internal stress components are defined as the The report considers the implementation of the method by a Matlab code, which is provided in the
most critical. Other stress components are neglected. Fc leads to a normal stress on the upper surface of Appendix. It is described in details in the code file itself. The code only considers design factors L and t.
𝐹 1
the structure (in xz plane) of 𝜎𝑁 = 𝐴𝐶 , with 𝐴 = 𝐿. 𝑡 − 4 𝜋𝑑22 . FR and Fw lead to a maximal bending stress of The first step is the determination of the optimal sample space. The code generates a matrix containing all
𝑀 .𝑑 possible sample spaces (each sequence of N rows is a sample space formed by the N experiments), and
𝜎𝑀 = 𝑚𝑎𝑥 𝑚𝑎𝑥 . The bending situation is simplified to a cantilever beam. Its length is d+f, the force is acting calculates the D-criterium for each of them. The optimal value of the criterium is then linked to the
𝐼
on its furthest point and the cross section is a filled rectangle. As a result, dmax and I for FR and Fw corresponding sample space, which is thus the optimal one and stored in a new matrix (N rows).
𝑔 𝑡.𝑔3 𝑡 𝑔.𝑡 3 Thereafter, the code creates the approximating function by calculating the ᵝ coefficients. The last step is to
respectively are and , and and . The second moment of area is overestimated, because the
2 12 2 12 find the optimal value of the analytical function (and thus of the response variable) and the corresponding
cross-section is actually hollow over the majority of the beam length. Mmax is overestimated as FR.(d+f) and
combination of design factors. The response variable (using the analytical function) and the constraints are
Fw.(d+f) (because the force is actually not acting on the furthest point on the beam), which compensates
calculated for all possible combinations of design factors (now more than L values) and stored in a matrix
the approximation of the cross section.
(each row is a different combination of design factors). All rows that do not satisfy one or more constraints
A second important aspect of the structure is its deformation. Only FR and Fw seem to lead to significant
𝐹.(𝑑+𝑓)3
are then deleted by a succession of if statements. The code also stores how many times the strength or
deformation. With the same assumptions as for the calculation of the bending stresses: ∆ = 3.𝐸.𝐼 . stiffness constraints were not met, by increasing a variable when an if statement is true. Finally, a simple
- Design factors to optimize: Figure 2 showed the dimension parameters of the structure. search in the updated matrix allows to find the optimal response variable value and the corresponding
All dimensions are design factors, which gives 22 design factors. However, a constant shape of the design factors. Furthermore, the code implements a simple statistical check, by computing the relative
structure is assumed (constant ratios between all dimensions in the xy plane and L, and between all other error between the exact value for the mass (using the exact formula) and the approximated value via the
dimensions and t). As a result, L and t are the only two design factors related to dimensions. The material analytical function, for each combination of design factors that satisfy the constraints.
can be Aluminium 5000 or 6000 series, so the material is a third factor.
- Optimization criteria: as mentioned before, there are two goals: minimal mass and minimal cost. The method is applied to the problem using following procedure. Firstly, arbitrary values of N=10, L=4
Mass is f(factors): 𝑚 = 𝜌. 𝑉, with 𝑉 = 𝑡 . ( (𝐿 + 𝑖). (𝑎 + 𝑐 + 𝑑 + 𝑓) − 𝑏. 𝑐 − 𝑒. 𝑓 − ℎ. 𝑖 − 𝑖. 𝑘 − (𝑙 + 𝑖). 𝑅 − (number of levels, not the design factor L) and domains of design factors 𝐿 ∈ [30𝑚𝑚; 100𝑚𝑚] and
𝑡 ∈ [10𝑚𝑚; 30𝑚𝑚] are chosen as characteristics of the sample spaces. Secondly, it is assumed that there
1 1 1 1
(𝑙 + 𝑅 + 𝑖). 𝑓1 − 𝜋𝑅 2 − 𝜋𝑑1 2 ) − 𝑔. (𝑙ℎ𝑜𝑙𝑒 . 2𝑅ℎ𝑜𝑙𝑒 + 2. 𝑅ℎ𝑜𝑙𝑒 . 𝑏ℎ𝑜𝑙𝑒 + 𝜋𝑅ℎ𝑜𝑙𝑒
2
) − 𝑎. 4 𝜋𝑑22 − 𝑗. 4 𝜋𝑑32 . are 100 possible values of the design factors L and t in the analytical function, evenly spaced within the
4 4
The minimalization of mass is equivalent to that of cost, because the cost of a structure is proportional with domain. All those values can be changed in one click and the code will automatically apply them. Lastly, as
the amount of material used. Other considerations are neglected (for example, the cost could be higher mentioned above, the material (third design factor) is not considered by the code. Therefore, the method is
within a certain interval of dimensions if less efficient machines had to be used within that interval). applied two times: once for Aluminium 5000 series and once for 6000. In that way, the method produces
- Constraints: the structure must satisfy a few criteria in order to functionate properly. an optimal design for the structure using that material (minimal mass and thus cost). The method gives
Firstly, the internal stresses must be smaller than the yield stress. Secondly, the deformation must be following results: the best design of the structure with 6000 has L = 30.71mm and t = 19.49mm which
limited to ensure contact with the wheel. The maximum value is set arbitrarily to dcrit=0.5mm. gives m = 0.00056kg and cost = £3.2/kg * m = 0.0018£, and L = 30.00mm and t = 20.30mm for 5000,
which gives m = 0.0031kg and cost = £3/kg * m = 0.0093. The absolute value of the relative error varies
Multiple methods to solve the optimization problem exist. The goal of the method is to find the combination from 0 to 1 as a function of the (L, t) combinations, but is mostly smaller than 0.5, as shown in Figure 3.
of design factors giving the best response variable. The latter is the quantity to minimalize or maximize. Furthermore, the strength constraint is never reached with 6000 as material, unlike with 5000. Also, about
While doing that, the method must take constraints into account. Those define a so-called design 30% of the possible (L, t) combinations doesn’t satisfy the stiffness constraints, for both materials.
envelope, which is the ensemble of allowed combinations of design factors (Minton 2022). This report
looks into the use of the Optimal design method (D2) to solve the problem, which is a Design of
Experiments method (DoE). The starting point of DoE methods is a set of experiments, or sample space:
a number of design factor combinations (N) are tested, each giving a certain value of the response
variable. The i-th experiment (sample) gives a response variable 𝑦 (𝑖) = 𝑓(𝑥⃗ (𝑖) ) (𝑖 = 1 … 𝑁), 𝑥⃗ (𝑖) being a
vector containing the k design factors 𝑥1 , … 𝑥𝑘 . The design factors can only take a limited number of
values (L) for the experiments, which are called levels. The sample space is then used to create an
analytical function that approximates the response variable, typically via response surface modelling
(RSM). Thereafter, it is easy to find the optimal value of the analytical function (and thus of the response
variable) and the corresponding combination of design factors. The general idea of DoE methods is to
efficiently use the experiments, in other words to get a maximum amount of data with a minimum amount
of experiments (N) (Cavazutti 2013, page 6). The optimal design DoE method searches for the optimal
sample space, on the basis of a certain criterium. This requires a high computing power, because it
necessitates the consideration of all possible sample spaces for the given k, N and L. This report considers
Figure 1: free-body diagram Figure 2: dimensions of the structure Figure 3: relative error of the method
 NAME: Valentin Borisavljevic ID: 2140173 Problem: Brake calliper Method: D2

Discussion

It is interesting to reflect on the correctness of the results. On the one hand, the resulting optimal
mass does not seem realistic. This is confirmed by the relative error (Figure 3): the selected design is one
of the first (L,t) combinations, and the relative error for the first combinations can be very high (up to
100%). On the other hand, the optimal combination of design factors L and t is likely to be fairly accurate.
This conclusion is based on following reasoning: the smallest dimensions give the smallest mass, and the
code took the (or one of the) smallest dimensions (L, t) for which the constraints were satisfied. The
verification of constraints is not based on the approximating function, so the selected dimensions are likely
to be close to the optimal ones. However, as mentioned at the beginning of the first paragraph (information
collection about the problem), a few assumptions were made to get the formulas related to the constraints
(stresses and deformations). As a result, some uncertainty remains about the verification of the constraints.

The applied optimization method causes the mentioned relative errors. Firstly, a higher number of
experiments (N) and possible values of the design factors for the experiments (levels L) could possibly lead
to a better sample space, leading to a more accurate approximating function for the response variable.
Furthermore, the D-criterium to define the optimal sample space is not the most popular one. The most
widely used criterium is the so-called I-criterium (Cavazutti 2013, page 38). Also, the used technique to
calculate the approximation function from the optimal sample space may not be the best. The least square
method is a relatively “basic” method. Other techniques exist (Cavazutti 2013, pages 49-70).

As mentioned in last paragraph, the code is applied two times: once for Aluminium 5000 series, once for
6000. This leads to an optimal structure design for 5000, and one for 6000. Those designs differ: bigger
optimal dimensions for 5000, and thus a higher optimal mass, because 5000 and 6000 have equal
densities. This is because the strength constraints were never reached with Aluminium 6000 as material,
unlike with 5000. Therefore, designs with smaller dimensions are possible with 6000 as material. For a
certain interval in which the dimensions of 5000 are only slightly higher than 6000, a situation
necessitating a trade-off between the two objectives arises: 5000 has a higher mass, but a smaller cost
than 6000. In that case, the relative importance of the two objectives needs to be defined in order to
choose the optimal design (the optimal 5000 design or 6000 design).

The comparison with the original design of the structure (given by the CAD file as an Appendix) is as
follows: L≈30mm vs 61mm, and t≈20mm vs 16mm for the original design. t is little bit higher, but L is much
smaller, which definitely leads to a smaller mass and cost. However, possible problems of integration
within the global design of the braking system require attention. A brake calliper with the calculated optimal
dimensions may be incompatible with the other structures of the braking system.

References
Cavazutti, M 2013, Optimization methods: from theory to design, Springer, Heidelberg.

Minton, T 2022, ME5542 Advanced modelling and design, lecture notes. Brunel University, London.

Appendices

Appendix A: Matlab code (.m), ‘Design_code’

Appendix B : CAD file (.step), ‘brake_calliper’