Master Thesis:Battery management system control and estimation design

An energy storage system (ESS) based on lithium-ion batteries is one of the most important but expensive and safety-critical components in the electrified powertrain. These batteries have complex nonlinear dynamics and need a battery management system (BMS) with advanced estimation and control algorithms to ensure their optimal performance and long lifetime. In this regard, the systems and control community have shown a lot of research interest in recent years. The overall goal is to develop a knowledgebase to design battery health conscious BMS for optimal utilization of currently available cells to guarantee their long lifetime. One of the core BMS functions is to estimate the battery’s internal state (state-of-charge [SOC], dynamic polarization, internal State-of-Temperature [SOT], etc.) and parameters (state-of-resistance [SOR], state-of-capacity [SOQ], etc.) using voltage, current, and temperature measurements. These estimates are used to provide critical predictions about maximum available battery energy and power (i.e., state-of-energy [SOE] and state-of-power [SOP]) during driving or charging. These predictions are then used to decide the maximum battery load to guarantee optimal, reliable, and safe operation (i.e., to respect voltage, current, and temperature limits)


Suitable background
Must have a strong educational background in electrical engineering, engineering physics, computer science, or mechatronics with very good grades in master level courses like machine learning, reinforcement learning, nonlinear filtering/estimation, linear control systems, nonlinear and adaptive control systems, model predictive control, etc.
Must have high proficiency in MATLAB and Simulink.
You must be self-motivated and meticulous in your problem-solving approach.
Familiarity with electro-thermal dynamics of lithium-ion batteries and some experience with dSpace embedded control software development tools will be considered a merit

Please send your application including a CV, Cover Letter, and Transcript of grades.

Description of thesis work
A battery system in electric vehicles contains multiple cells connected in series or in parallel. These cells may exhibit different dynamic behaviours, due to electrochemical variances from the manufacturing process, locations of cells in a multi-cell system, and inevitable variations or imbalances in their internal parameters and operating conditions. The different dynamic behaviours lead to SOC and temperature imbalances among cells. The degree of imbalances over SOC and temperature is also affected by the different drive cycles and operating conditions. The cell imbalances are harmful to batteries’ energy and power utilization, and hence their lifetimes. To mitigate cell imbalance, one typical approach is to lower the energy of cells with a relatively high SOC to have the lowest SOC in a multi-cell system when the maximum SOC difference among all the cells is greater than a certain threshold. This method is simple, but not energy and power-optimal considering different drive cycles and operating conditions.


In this thesis, we will design a uniform approach to balance both SOC and temperature for different optimization objectives using reinforcement learning, considering measurement data (current, voltage, temperature), SOX, and drive cycle data. This thesis deals with a part of this puzzle with the scope confined to the following research tasks:
Topology analysis for passive and active cell balancing of series-connected and parallel-connected multi-cell systems for lithium-ion batteries. The characteristics of the degree of cell imbalance in SOC and temperature will be analysed.
Reinforcement learning problem formulation and solution for cell balancing. Define states, actions, and rewards of the Markov decision process in a drive cycle. An efficient decision-making policy should be drawn using a reinforcement learning method (such as Q-learning, temporal difference learning, or hybrid approaches).
Interpretability analysis of the proposed method regarding inputs, rewards, and decision-making. We will evaluate and visualize the proposed method for the specific multi-cell system from the user’s point of view.
Investigation of deployment methods for the proposed reinforcement learning solution in real applications. We will estimate and evaluate the performance of the newly learned agents in the target environment, namely, electric vehicles.
Simulation and verification of the proposed learning methods for real drive cycles and operating conditions. The numerical results should be compared with the performance of benchmark methods. The main purpose is to thoroughly evaluate the performance, potential benefits, and costs associated with the operation of the proposed approach.



Thesis Title: Reinforcement Learning-Based Cell Balancing for Electric Vehicles


Thesis Level: Master

Language: English


Starting date: 2024-01-14


Number of students: 2


Tutor
Dr. Yang Xu
Senior Control Engineer, +46 739023241
BMS Controls
Department of Electromobility
CampX, Gothenburg, Sweden


Kindly note that due to GDPR, we will not accept applications via mail. Please use our career site.