BATTERY SYSTEM DEGRADATION MODELLING FOR TWO-WHEEL ELECTRIC VEHICLE BASED ON MACHINE LEARNING

• Main Author: Rafi Elian, Faishal
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/54725
• Institution: Institut Teknologi Bandung

Today the world has begun to experience a major transformation in transportation, with the increasing number of electric vehicles sold. Batteries as energy storage, play a vital role in terms of electric vehicle efficiency. But, the battery cannot be used continuously so its degradation model becomes important to be known.. In order to model battery degradation, a suitable Battery Management System (BMS) needs to be developed. This study will discuss the integration of a Battery Management System that can provide sufficient data so that the battery State-of-Health (SOH) degradation of the battery pack in electric vehicles can be modeled. In this study, an Internet of Things (IoT) based BMS will be made on a two-wheeled electric vehicle with a Li-Ion battery pack of the LiNiCoMnO2 (Nickel-Manganese-Cobalt,NMC) cathode. This electric vehicle has 5 levels of assist as the motor power limit with level 5 as the highest level, using ESP32 as a communication module. and microcontroller. Data was collected using 3 scenarios in real-time, with scenarios 1 and 2 carried out on the ITB campus with assist levels 5 and 3 respectively, and scenario 3 carried out on the uphill route ITB-Jl. Ir. H. Djuanda to Bukit Dago with assist level 5, this data will be trained together with Li-Ion battery data of the LiFePO4 (Lithium-Ferro-Phosphat,LFP) and LiMn2O4 (Lithium-Manganese Oxide,LMO) cathode types which come from references with several machine learning models and the best-optimized modeling results will be compared to the NMC battery degradation model. In this study, the BMS that has been made has succeeded in producing voltage and current data during driving which is then processed to produce capacity data for several initial cycles using the Approximate Weighted Total Least Square (AWTLS) method. This data is successfully modeled by machine learning Supporting Vector Regression (SVR) using additional LFP type battery data for scenarios 1 and 2, and additional LMO battery type data for scenario 3. Scenario 1 and 2 models fail to achieve degradation capacity while scenario 3 succeeds in achieving degradation capacity on the 490th cycle. So the scenario 3 model can represent the NMC battery degradation model with R2=0.911. Keywords: Electric Vehicle, Battery Management System, Battery Degradation, Machine Learning, Internet of Things