PSI - Issue 68

Tuncay Yalçinkaya et al. / Procedia Structural Integrity 68 (2025) 325–331 Yalc¸inkaya et al. / Structural Integrity Procedia 00 (2024) 000–000

329

5

20

10

Preform

20

20

20

Roller

20

20

26

Mandrel

100

20

100

6.27

9

80

5

60

51.3

Ele. Num. = 6

45

30

26

25

App. Ele. Size = 0.7

App. Ele. Size = 10.25

(e)

(b)

(d)

(a)

(c)

Fig. 1: FE model and geometry of a) ST, b) NT, c) PST, d) ISS and e) FE model of flow forming process with mesh size.

the friction e ff ect. The friction coe ffi cient is 0.12 between the mandrel and the preform and 0.08 between the preform and the rollers. The FE model employed is a temperature-displacement connected model, where the material is heated due to plastic deformation and friction. For this reason, the outer surface of the preform is kept constant at 293K during the simulation using surface film condition interaction.

3. Results and Discussion

Plasticity and damage criteria are calibrated using the ST specimen, then applied to other specimen geometries to evaluate their predictive capabilities. The experimental and numerical fracture-initiation displacement comparisons for NT, PST, and ISS specimens reveal key insights into the performance of various damage models. For the NT specimen, all models except LR and Freudenthal predict fracture displacement values relatively close to the experimental value. In particular, the Ayada, Ayada-m, KH, MC, and RT models are successful in predicting fracture under notch tensile conditions, showing the closest agreement with experimental data. Similarly, for the PST specimen, these models provide predictions that align closely with experimental values. However, for the ISS specimen, the KH, RT, and Freudenthal models stand out as the most accurate in predicting fracture displacement, whereas the Ayada, Ayada-m, and MC models, which perform well in other geometries, are less successful in in-plane shear conditions. Overall, the analysis highlights that the accuracy of damage models varies under di ff erent stress states. The KH and LR models stand out for their consistency and reliability, while the Ayada and Ayada-m models may require further refinement to improve their accuracy, especially for in-plane shear conditions. In contrast, the RT and MC models exhibit the highest total error values, with significant errors observed especially in the ISS sample estimates. Subsequently, flow forming simulations are conducted, and the fracture limit and location predictions of the criteria are analyzed. Figs. 2 and 3 show the results of flow forming FE simulations with calibrated damage models. Erdogan et al. (2023) demonstrated and explained that forming was completed at 37.5% and 50% thickness reduction (TR), but the part failed at 70% TR. Additionally, it is stated that radial cracks occurred on the outer surface due to the process at a rate of 50% TR. Fig. 2 shows the damage distribution in the preform after flow forming at a 37.5% thickness reduction rate predicted by various damage models. This shows that if the damage prediction of models at the thickness reduction ratio exceeds one, these models cannot accurately predict the forming limits. Ayada, Ayada-m, KH, and MC models correctly predict that the damage values remain below one and successful forming is achieved at this reduction ratio. In contrast, the Brozzo, LR, OH, RT, CL and Freudenthal models show damage values exceeding 1. While cracks occur on the inner surface in the Brozzo, OH and CL models, LR, RT and Freudenthal also predict high damage accumulation throughout the entire part. Fig. 3 shows the damage distribution of flow forming at 50% thickness reduction for the calibrated models. Dam age distributions are shown for Ayada, Ayada-m, Brozzo, KH, MC, OH, and CL models, excluding models like LR,