Subcategory: Physics (not Nanoscience)
Arnav Mazumder - University of North Texas
Co-Author(s): Dmytro Shymkiv, University of North Texas, Denton, TXArkadii Krokhin, University of North Texas, Denton, TX
Obtaining parameters for higher efficiency or optimization is one of the most demandable problems that people are solving in every area of life and science. Enhancing performance in solar panels, engines, and batteries is just the surface of the scientific limits that can be exceeded. In this work, we study phononic crystals – periodic elastic structures. We focus on nonreciprocity and one-way transmission (acoustic diodes), which are related to the difference in sound propagation in the forward and backward directions. Using machine learning, which gave great results for similar systems [1, 2], we optimize the parameters of the structures. A phononic crystal of asymmetric aluminum rods in a water background is a passive nonreciprocal medium . We varied the level of geometrical asymmetry in the scatterers by changing the radii of the cylinders, removing certain slices, and rotating each scatterer (rotational angle). Sound propagation through all scatterers was numerically calculated in COMSOL for a finite number of parameters’ values. Machine learning was used to predict the level of nonreciprocity and acoustical diode quality factor for the uncovered values. From the predictions we searched for the maximum nonreciprocity and minimum quality factor (“0” corresponds to the best performance).Totally, we obtained a dataset with 3 parameters: type of geometry, radius, and rotational angle. Before processing this large dataset with more than 1000 samples, the smaller one for 1 type of the geometry and 2 free parameters was analyzed. It helped to check the proof of the concept and to determine techniques, which might be useful for a full dataset, for instance, dealing not with the actual rotational angle, but with the trigonometric functions of the angle instead. The working model was developed and the predicted optimal parameters for the better quality factor were used to evaluate the performance of the model. The minimum quality factor in the dataset was 8.5% and the best predicted by the model was 7.5%. Comsol check gave 8.4% with the optimized parameters. Although it is not the same as the prediction said, it shows enhancement in the diode performance. Currently, we are developing a larger model with a full dataset. The results for the small dataset with less than 200 samples give a hope to achieve higher nonreciprocity and diode efficiency. In the future work, we plan to vary wave frequency and consider other kinds of scatterer geometries.References:1. I. Malkiel et al., 2018, Plasmonic nanostructure design and characterization via Deep Learning, Light Sci Appl 7:60.2. Zh. Kudyshev et al., 2020, Machine-learning-assisted metasurface design for high-efficiency thermal emitter optimization, Applied Physics Reviews, 7, 021407. 3. E. Walker et al., 2018, Nonreciprocal Linear Transmission of Sound in a Viscous Environment with Broken P Symmetry, Phys. Rev. Lett. 120, 204501.
Funder Acknowledgement(s): This work was supported by an Emerging Frontiers in Research and Innovation grant from the National Science Foundation (Grant No. 1741677).
Faculty Advisor: Dr. Arkadii Krokhin, email@example.com
Role: In this research, I set up and ran COMSOL simulations, collected the data samples, and developed all machine learning models. I also selected the scatterer geometries and parameters that would vary (radii and angle of rotation). All data and model visualizations were generated by me.