Page 23 - IJAMD-2-1

International Journal of AI for
Materials and Design
Predicting thermal conductivity of sintered Ag

ideal prediction trendline (red dashed line), indicating that (i). Image processing and modeling simulation methods for
the predicted parameters are relatively close to the actual SEM images of sintered nano-Ag microstructure were
parameters. The value of R is as high as 0.96, indicating established by MATLAB and Ansys software. Batch
2
that the ANN model based on Bayesian optimization has a calculation of the thermal conductivity from 2D images
good predictive ability for the correlation analysis between of sintered nano-Ag microstructure was performed
microstructure characteristics and thermal conductivity accordingly. Based on the SEM image characteristics of
of sintered nano-Ag. The 95% confidence region also actual sintered nano-Ag microstructure, 186 images of
confirms that the proposed optimal ANN model has good the microstructure, obtained from using 3-pixel sizes
predictive performance. The confidence interval represents and different sintering times, were reconstructed using
the knowledge level of the best fitting line and determines the Gaussian random model. The thermal conductivity
that the true linear fitting output is within the interval. In in the x- and y-directions of the microstructure plane
Figure 12, when the trained ANN model was used to predict model was obtained by finite element simulation.
thermal conductivity, the confidence interval between the (ii). Given the advantages and disadvantages of both the
predicted output and the real output was compared; it was numerical parameters and the image input types for
found that most of the predicted points overlapped with neural network analysis, the numerical dataset and
the actual value, with smaller differences corresponding to the image dataset are established, respectively. An
higher prediction accuracy. image was obtained from the reconstructed model of
The factors that affect the thermal properties of sintered the sintered nano-Ag microstructure, and the dataset
nano-Ag, such as sintering process parameters, affect the was enhanced four times. The model accuracy can be
microstructure of sintered Ag nanoparticles. However, in further improved by enhancing the amount of data
existing studies, microstructure parameters are rarely used used for model training.
in thermal conductivity models. Notably, our study utilized (iii). Based on the Bayesian optimized ANN model, the
30
the microscopic reconstruction method, which combines average particle size, circumference, and porosity were
Gaussian random reconstruction and finite element, taken as input parameters to predict the equivalent
effectively studying the thermal conductivity changes as thermal conductivity of sintered nano-Ag. The final
the microstructure changes under the influence of sintering determination coefficient of the ANN model for the
parameters. However, due to limitations in the dataset and equivalent thermal conductivity prediction is 0.96.
the difficulty of obtaining it, we only analyzed the evolution The results are of great significance for evaluating the
of microscopic grains at different sintering times using the microstructure and physical properties of sintered
open operation method, without considering the effect nano-Ag. Hence, future studies should focus on the
of sintering temperature. Nevertheless, if the influence of physical mechanisms based on existing models.
sintering temperature changes on thermal conductivity
is included, the model would remain valid. Rong et al. Acknowledgments
29
mentioned in their study that features selected from a The authors would like to express their gratitude for the
large descriptor space may limit the predictive accuracy support from the National Natural Science Foundation
of machine learning models. It is worth noting that we of China (No. 12272012) and acknowledge the AI tool
simultaneously used both the image dataset and the feature SparkDesk (iFLYTEK) for enhancing the language quality
parameter dataset for training, effectively mitigating this and readability of the manuscript.
issue and enhancing the model’s generalization ability for
different types of structures. We are working to integrate Funding
the aforementioned factors to study the microstructure
evolution of sintered nano-Ag, aiming to obtain a more This research was financially supported by the National
comprehensive dataset that captures the changes in thermal Natural Science Foundation of China (No. 12272012).
characteristics with respect to microstructure parameters. Conflict of interest
4. Conclusion The authors declare that they have no known competing

We presented the image of sintered nano-Ag microstructure financial interests or personal relationships that could have
based on Gaussian reconstruction, where feature parameters appeared to influence the work reported in this paper.
and image data are separately extracted to form a dataset. Author contributions
The equivalent thermal conductivity of sintered nano-Ag
is then predicted using a machine-learning model. In Conceptualization: Jiahui Wei, Yanwei Dai
summary, our key findings are as follows: Formal analysis: Libo Zhao, Jiahui Wei

Volume 2 Issue 1 (2025) 17 doi: 10.36922/ijamd.5744