Physics Informed Methodology Using Neural Network to Match Measurements in Sensor Devices
Vishal Venkata Raghavendra Nandigana
*
Issue: Volume 10, Issue 4, August 2025
Pages: 84-95
Received: 21 July 2025
Accepted: 4 August 2025
Published: 21 August 2025
Abstract: In this paper we develop physics informed neural network model to solve battery technology. The first model uses physics from the theory. The voltage of the battery is related to the charge carrier, frequency term and power. The theory is used to obtain 15 different voltages. The parameters charge carrier, frequency term, power and voltage are our 15 training data. The training data is trained using Recurrent Neural Network (RNN) Long Short-Term Memory (LSTM). The algorithm determines the weight based on the training data. We study for 50, 100, 150, 200 and 500 epoch. We predict for 15 test cases. The predict file has variables [charge carrier given, frequency given, voltage not given and power given]. We obtain huge error when the training set is given as one file with 15 rows of 4 variables in each row. However the physics from the theory matches with the predict answer for the voltage when the training file has one row of 4 variables that is repeated to study multiple times. We have 15 different training files. We study for 50, 100, 150, 200 and 500 epoch. The dependency on the epoch is visible until 200. The accuracy is 95% for few predict test case results. The predict voltage correlates with the theory. Thus, the model is physics from theory included in the neural network for the first time. Next, we study the physics informed partial differential equation with the neural network. We use 15 training sets. Each training set have 10 rows with variables [grid location, concentration, voltage distribution, frequency term and current]. We use our model to test for one case. The test variables are [same grid location, 1 new concentration, predict the voltage distribution, same frequency term and same current]. We obtain good output matching the actual and the physics. The accuracy is 80%. We study for epoch 50.
Abstract: In this paper we develop physics informed neural network model to solve battery technology. The first model uses physics from the theory. The voltage of the battery is related to the charge carrier, frequency term and power. The theory is used to obtain 15 different voltages. The parameters charge carrier, frequency term, power and voltage are our ...
Show More
