A deformation prediction based on deep learning for sintering process of ceramic core

doi:10.1007/s40436-025-00549-w

Abstract

Abstract: Ceramic cores play a significant role in determining the cavity structure of hollow turbine blades during precision casting. However, the sintering process in preparing ceramic cores may result in shrinkage and deformation due to high temperature, presenting significant challenges in controlling the dimensional accuracy of ceramic cores. In this work, we develop a framework based on deep learning to predict the three-dimensional deformation of ceramic cores during sintering under varied sintering parameters. A finite element thermo-elasto-viscoplastic model is developed to compute sintering deformation and generate the three-dimensional deformation database. The numerical model is validated using a sintering experiment, and the maximum deviation in the deformation between the numerical and experimental results is 0.383 mm, which is 3.19% relative to the diameter of the largest inscribed circle of the ceramic core section, and satisfies the precision requirement of the third level of dimensional casting tolerance grade (DCTG3). The developed framework slices each of the three-dimensional shapes of the sintered ceramic core in sequence to obtain the two-dimensional image data for training the deep learning network. A parameter-embedded U-net network is established and trained to learn the intricate relationship between sintering parameters and deformation in sliced images. A VTK reconstruction algorithm is applied to the slice sequence to restore the predicted images from the U-net to the three-dimensional shape of the ceramic core. A metric for evaluating the model accuracy based on the error of deformation prediction (EDP) is proposed specific to the image character of ceramic core sintering, and the score of EDP for the developed U-net is 4.31%, indicating a high accuracy in predicting the sintering deformation in sliced images. An unseen combination of process parameters is numerically computed, and the entire three-dimensional deformation is compared to the prediction from the developed framework. The result shows that the relative maximum deviation in deformation is 2.93%, demonstrating the overall good performance of the developed framework in predicting sintering deformation.

The full text can be downloaded at https://doi.org/10.1007/s40436-025-00549-w

Key words: Ceramic core sintering, Deformation prediction, Deep learning, Numerical simulation

Song-Zhe Xu, Qi-Yu Yuan, Zhi-Fan Tang, Chao-Yue Chen, Tao Hu, San-San Shuai, Wei-Dong Xuan, Zhong-Ming Ren. A deformation prediction based on deep learning for sintering process of ceramic core[J]. Advances in Manufacturing, 2026, 14(2): 499-513.

TrendMD

References

[1] Poullikkas A (2005) An overview of current and future sustainable gas turbine technologies. Renew Sustain Energy Rev 9(5):409-443
[2] Dong YW, Li XL, Zhao Q et al (2017) Modeling of shrinkage during investment casting of thin-walled hollow turbine blades. J Mater Process Technol 244:190-203
[3] Zhang L, Liu JP, Sun G (2012) Preparation and properties of silica-based ceramic core for single crystal blades. Foundry 8:941-943
[4] Pan ZP, Guo JZ, Li SG et al (2022) Experimental study on high temperature performances of silica-based ceramic core for single crystal turbine blades. Ceram Int 48(1):548-555
[5] Gromada M, Swieca A, Kostecki M et al (2015) Ceramic cores for turbine blades via injection moulding. J Mater Process Technol 220:107-112
[6] Zheng W, Wu JM, Chen S et al (2023) Preparation of high-performance silica-based ceramic cores with B4C addition using selective laser sintering and SiO₂-Al₂O₃ sol infiltration. Ceram Int 49(4):6620-6629
[7] Li H, Liu YS, Liu YS et al (2020) Influence of sintering temperature on microstructure and mechanical properties of Al₂O₃ ceramic via 3D stereolithography. Acta Metall Sin (Eng Lett) 33:204-214
[8] Li H, Hu KH, Liu YS et al (2021) Improved mechanical properties of silica ceramic cores prepared by 3D printing and sintering processes. Scripta Mater 194:113665. https://doi.org/10.1016/j.scriptamat.2020.113665
[9] Li H, Liu YS, Colombo PL et al (2021) The influence of sintering procedure and porosity on the properties of 3D printed alumina ceramic cores. Ceram Int 47(19):27668-27676
[10] Zhao ZJ, Yang ZG, Yu ZH et al (2020) Influence of yttrium oxide addition and sintering temperature on properties of alumina-based ceramic cores. Int J Appl Ceram Technol 17(2):685-694
[11] Olevsky EA (1998) Theory of sintering: from discrete to continuum. Mater Sci Eng R Reports 23(2):41-100
[12] Kraft T, Riedel H et al (2004) Numerical simulation of solid state sintering, model and application. J Eur Ceram Soc 24(2):345-361
[13] Braginsky M, Tikare V, Olevsky E (2005) Numerical simulation of solid state sintering. Int J Solids Struct 42(2):621-636
[14] Zhang R (2005) Numerical simulation of solid-state sintering of metal powder compact dominated by grain boundary diffusion. The Pennsylvania State University, State College
[15] Kakanuru P, Pochiraju K (2022) Simulation of shrinkage during sintering of additively manufactured silica green bodies. Addit Manuf 56:102908. https://doi.org/10.1016/j.addma.2022.102908
[16] Ni Y, Liu K, Wang J (2021) Establishment of constitutive models and numerical simulation of dry pressing and solid state sintering processes of MgTiO₃ ceramic. Ceram Int 47(7):8769-8780
[17] Long J, Shelhamer E, Darrell T (2015) Fully convolutional networks for semantic segmentation. IEEE Trans Pat Anal Mach Intell 39:640-651
[18] Badrinarayanan V, Handa A, Cipolla R (2015) Segnet: a deep convolutional encoder-decoder architecture for robust semantic pixel-wise labelling. arXiv:1505.07293. https://doi.org/10.48550/arXiv.1505.07293
[19] Krizhevsky A, Sutskever I, Hinton GE (2012) Imagenet classification with deep convolutional neural networks. In: Advances in neural information processing systems 25, Curran Associates Inc, New Orleans
[20] Ronneberger O, Fischer P, Brox T (2015) U-net: convolutional networks for biomedical image segmentation. In: The 18th international conference on medical image computing and computer-assisted intervention (MICCAI). Springer, Germany
[21] Chen J, Viquerat J, Hachem E (2019) U-net architectures for fast prediction of incompressible laminar flows. arXiv:1910.13532. https://doi.org/10.48550/arXiv.1910.13532
[22] Ribeiro MD, Rehman A, Ahmed S et al (2020) DeepCFD: efficient steady-state laminar flow approximation with deep convolutional neural networks. arXiv.2004.08826. https://doi.org/10.48550/arXiv.2004.08826
[23] Li YZ, Wang YP, Qi ST et al (2020) Predicting scattering from complex nanostructures via deep learning. IEEE Access 8:139983-139993
[24] Jiang ZH, Tahmasebi PM, Maoand ZQ et al (2021) Deep residual U-net convolution neural networks with autoregressive strategy for fluid flow predictions in large-scale geosystems. Adv Water Resour 150:103878. https://doi.org/10.1016/j.advwatres.2021.103878
[25] Sun NL, Zhou ZM, Li Q et al (2022) Spatiotemporal prediction of monthly sea subsurface temperature fields using a 3D U-net-based model. Remote Sensing 14(19):4890. https://doi.org/10.3390/rs14194890
[26] Wang J, Berger D, Mattie D et al (2021) Multichannel input pixelwise regression 3D U-nets for medical image estimation with 3 applications in brain MRI. In: Medical Imaging with deep learning, PMLR, Lübeck
[27] Paszke A, Gross S, Massa F et al (2019) Pytorch: an imperative style, high-performance deep learning library. In: Advances in neural information processing systems 32, Curran Associates Inc, New York
[28] Wee LK, Chai HY, Supriyanto E (2011) Surface rendering of three dimensional ultrasound images using VTK. J Sci Ind Res 70(6):421-426
[29] Sarbandi B (2021) Finite element simulation of ceramic deformation during sintering. Dissertation, Ecole Nationale Suprieure des Mines de Paris
[30] Markus W, Reiterer KG et al (2006) An arrhenius-type viscosity function to model sintering using the skorohod-olevsky viscous sintering model within a finite-element code. J Am Ceram Soc 89(6):1930-1935
[31] Zhang Z (2018) Improved adam optimizer for deep neural networks. In: ACM 26th international symposium on quality of service (IWQoS), IEEE, Banff, pp 1-2