Fig. 5: Ranking performance and computation cost.
From: Deep learning for non-parameterized MEMS structural design
a Comparison of ranking obtained from the DL model and FEA simulation for the entire testing dataset. The red box features the top 10% of data in ranking. b Magnified view of the boxed region in a. c Patterns of the top eight designs based on high Qanchor values in the testing dataset as calculated using the DL model. The numbers in the bottom square brackets are FEA Rankings/DL Rankings. d The histogram shows the computation time costs in terms of seconds for the FEA simulation for frequency and Qanchor, the data preprocessing, and the training and testing process for the double-output DL model. The pie chart shows the percentage of the computation time costs from the FEA simulations, data preprocessing, DL training and testing processes to analyze the frequency and Qanchor of the MEMS disk resonators.
