Convolutional Neural Network battery pack classification - Gramian angular field vs. Markov Transition Field

Henrik Andersen*, Kasper Mayntz Paasch

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

Abstract

In battery pack manufacturing the main time-consuming aspects is the long test time for discharge and charge cycles. Making an AI that can assist by predicting the result of these tests while they are running is the end goal of this series of papers, where this is step one. To achieve this, the classification of the different battery pack types and the different tests must be performed. In collaboration with the company Banke Aps., who is a battery pack manufacturer, a database of tests has been build forming the foundation of this paper. Using the Gramian Angular Field (GAF) and Markov Transition Field (MTF) methods to transform the time series data into image form gives the possibility to utilize the standard convolutional neural network (CNN) structures to classify the battery pack and test type. Furthermore, building an algorithm that can distinguish between pass or fail tests with as high accuracy as possible, is important. Some question, that needs to be investigated, is which mathematical technique is best, GAF or MTF? Can these mathematical techniques identify important test features in the generated images and be used to estimate the test outcome?

The performance of the AI is measured in percent accuracy and the aim is to achieve as high accuracy as possible (above 90%) giving the confidence to apply the AI in the battery factory test system. Here the AI will evaluate the battery pack in real time and give the operator an indication if the battery pack will fail in the future or not.

To get above 90% accuracy, 10 pretrained neural networks will be tested and the best selected for this specific task. The networks range from “simple” (25 layers) to more complex (1243 layers) giving a good overview of the whole range of networks.
Original languageEnglish
Title of host publication7th E-Mobility Power System Integration Symposium (EMOB 2023)
PublisherInstitution of Engineering and Technology
Publication date2023
Edition18
Pages205-211
ISBN (Electronic)978-1-83953-957-2
DOIs
Publication statusPublished - 2023
Event7th E-mobility power system integration conference - Anker Engelunds Vej 1, 2800 Kgs. Lyngby, Denmark
Duration: 25. Sept 202326. Sept 2023
https://mobilityintegrationsymposium.org/

Conference

Conference7th E-mobility power system integration conference
LocationAnker Engelunds Vej 1
Country/TerritoryDenmark
City2800 Kgs. Lyngby
Period25/09/202326/09/2023
Internet address

Keywords

  • DEEP LERNING
  • GRAMIAN ANGULAR FIELD
  • MARKOV TRANSITION FIELD
  • BATTERY PACK
  • LIFEPO4

Fingerprint

Dive into the research topics of 'Convolutional Neural Network battery pack classification - Gramian angular field vs. Markov Transition Field'. Together they form a unique fingerprint.

Cite this