Dataset
OHTPD Dataset is generated based on typical 88 lines of codes of topological optimization and has been released at BaiduPan （Password: u8fv）.
In terms of dataset construction, three random layout distributions with different component numbers （10, 15, and 20） are mainly considered, as shown in the second row of Table 1; For each layout sample, the SIMP method is used to obtain the corresponding optimal heat transfer structure topology, as shown in the third row of Table 1, so as to form the input and output data pairs of the thermal component layout-structure topology optimization scheme. Corresponding to the number of components of each class, 10,000 samples are generated for training and testing.
Evaluation Criteria
In terms of evaluation criteria, according to the characteristics of deep learning and heat transfer design tasks, the evaluation indicators for image performance and physical performance are proposed, which are designed to measure the different attribute characteristics of heat transfer design results. Image performance mainly includes pixel-level average error and similarity to evaluate the accuracy and accuracy of the design results. The physical properties mainly include the average relative temperature field error and the average relative volume error, which are used to evaluate the heat conductivity and the amount of structural materials used in the design results.
