نگين عرفانيان ارا

Silicon with a theoretical capacity of 4200 mAh/g has been introduced as a suitable alternative to graphite (372 mAh/g) in the anode of lithium-ion batteries, but its main challenge is the swelling of about 300% during the charging an‎d discharging process, which leads to a reduction in capacity an‎d ultimately structural destruction of the anode. One solution to overcome this challenge is the use of silicon composites with other materials. In this project, silicon-nickel-carbon composite anodes were synthesized by conventional ball milling an‎d silicon-nickel-graphite composite anodes were synthesized by high-energy ball milling. Then, their electrochemical behaviors were investigated by cyclic voltammetry (CV) an‎d galvanostatic charge-discharge (GCD) tests. From the results obtained from high-energy ball milling synthesis, the first discharge capacity for the composite anode with PVDF binder was about 1740 mAh/g with a Coulombic efficiency of 58% an‎d for the CMC binder it was about 1500 mAh/g with a Coulombic efficiency of 78%. After returning to the initial current density, the composite anode was able to maintain a discharge capacity of about 680 mAh/g, which showed that this electrode has a very good capacity an‎d stability compared to other electrodes. The main objective of this research was to use the experimental data obtained from the experiments conducted in this study to predict the electrochemical behavior of the lithium-ion battery anode system including cyclic voltammetry (CV) curves an‎d galvanostatic charge-discharge diagrams using machine learning methods based on artificial neural networks. According to the results obtained, the neural network architecture with two hidden layers an‎d 50 neurons in each layer with a coefficient of determination of 0.99, an average absolute error of 8.35, an‎d a root mean square error of 17.93 showed the best performance in predicting the cyclic voltammetry curve. Also, for predicting the charge-discharge curves, the architecture with three hidden layers an‎d 50 neurons in each layer with a coefficient of determination of 0.99, an average absolute error of 0.73, an‎d a root mean square error of 1.17 was optimal. The output of these models showed a very favorable overlap between the experimental an‎d predicted curves in all areas.

باتري ليتيوم يوني , آند سيليكوني , كامپوزيت سيليكون-نيكل , شبكه عصبي مصنوعي , بال ميلينگ

Lithium-ion battery , Silicon anode , Silicon-nickel composite , Artificial neural network , Ball milling

Negin Erfanian Ara

Dr. Mohammad Reza Zamani Mimian