PSI - Issue 57

Khashayar Shahrezaei et al. / Procedia Structural Integrity 57 (2024) 711–717 K. Shahrezaei et al. / Structural Integrity Procedia 00 (2023) 000–000

716

6

4.3. GSA of the arbitrary uncertainty profiles

Results shown in Figure 3 are based on actual experimental data of the material using the experimentally extracted distribution profiles. Using the experimental data from the material that is under investigation helps in assessing what might be relevant for this specific use case and material. However, when looking at uniform distributions and their results (Figure 4), it can be seen that the e ff ect of void angle is increased. The uniform distribution reveals that the void angle has an e ff ect on the material sti ff ness that in some cases might not be negligible and could be important if the sti ff ness in Y and Z direction was crucial for the specific design. This circumstance doesn’t weigh as heavily in the case where we use the actual experimental distributions of variables as input. In that case, it can be concluded that the size of the void is what is the most important.

Exx Eyy Ezz Gxy Gyz Gxz

Exx Eyy Ezz Gxy Gyz Gxz

Exx Eyy Ezz Gxy Gyz Gxz

0.8

0.8

0.8

0.6

0.6

0.6

0.4

0.4

0.4

0.2

0.2

0.2

0

0

Ratio

Ratio

Ratio

Angle

Angle

Angle

MajorAxis

MajorAxis

MajorAxis

(a)

(b)

(c)

Fig. 4: The Sobol’ sensitivity indices using uniform distributions for (a) ANN, (b) PCE, and (c) Kriging.

4.4. Outlook

The metamodeling framework is applied to a case study for a CFRP material predicting macro properties. The usage of the framework streamlines part of a MSM simulation process used to investigate fatigue behavior. The framework shows great possibilities, and one should not limit oneself to only one kind of material or one kind of simulation process. The study presents a framework that could support many di ff erent kinds of design development processes providing a great amount of knowledge and deeper understanding when the limitations are low. Replacing computationally heavy simulations leads to the possibility of a deeper investigation of design possibilities early in the design development process. This would help designers base their choices on worst- and best-case scenarios and lead to more flexibility early in the design development process. An improvement of the metamodels would be to identify the least influential factors, which is as important as identifying the most influential parameters, because this knowledge may potentially be exploited in a factor-fixing setup. Practitioners may use such information and keep the least influential input factors constant if the parameters have negligible influence on the material properties.

5. Conclusions

The following conclusions may be drawn from this study:

• All the considered metamodels were e ff ectively performing well with su ffi cient accuracy using only 240 training samples and 60 testing samples.