Scholarlink

Accurate health diagnostics are crucial for ensuring battery reliability in critical applications, preventing potential malfunctions, and reducing the risk of accidents. This research proposes a Data-Driven approach using Convolution Neural Network (CNN) model to diagnose battery health. Lithium-ion batteries (LIB) are a type of battery that is widely used in many industries. Battery datasets from NASA are used as training and testing datasets to build the proposed model. This study proposes SoH diagnostics based on data pre-processing method and CNN model. In data pre-processing, the capacity degradation data is visualized, highly related features are selected by Pearson Correlation Coefficient (PCC). Then, all features are normalized by the min–max feature scaling method, which will speed up the training process in achieving the minimum cost function and then data splitting is performed. After pre-processing, all selected features are incorporated into the CNN model. There are four main features in this study i.e. cycle (t), voltage (V), current (I), and temperature (T) as input data. The layering structure is configured by referring to Multi-Channel-CNN (MC-CNN) and LeNet-5. The output for predicting the battery uses State of Health (SoH). The proposed CNN show better results compare with MC-CNN, has result RMSE (Root Mean Square Error) as 1.58%.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Health Diagnostics for Lithium-Ion Batteries Using Convolutional Neural Network Model

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Ikhsan Romli,
  • Mohammad Mi’radj Isnaini,
  • Rachmawati Wangsaputra,
  • Bermawi Priyatna Iskandar

摘要

Accurate health diagnostics are crucial for ensuring battery reliability in critical applications, preventing potential malfunctions, and reducing the risk of accidents. This research proposes a Data-Driven approach using Convolution Neural Network (CNN) model to diagnose battery health. Lithium-ion batteries (LIB) are a type of battery that is widely used in many industries. Battery datasets from NASA are used as training and testing datasets to build the proposed model. This study proposes SoH diagnostics based on data pre-processing method and CNN model. In data pre-processing, the capacity degradation data is visualized, highly related features are selected by Pearson Correlation Coefficient (PCC). Then, all features are normalized by the min–max feature scaling method, which will speed up the training process in achieving the minimum cost function and then data splitting is performed. After pre-processing, all selected features are incorporated into the CNN model. There are four main features in this study i.e. cycle (t), voltage (V), current (I), and temperature (T) as input data. The layering structure is configured by referring to Multi-Channel-CNN (MC-CNN) and LeNet-5. The output for predicting the battery uses State of Health (SoH). The proposed CNN show better results compare with MC-CNN, has result RMSE (Root Mean Square Error) as 1.58%.