Health Degradation analysis of Lithium ion batteries

Project Title

Health Degradation analysis of Lithium ion batteries

Project Area of Specialization Electrical/Electronic EngineeringProject Summary

Li-ion batteries are gradually becoming the most commonly used type of batteries. They are popular because of long lifetime, low weight, high energy density, large depth of discharge, wide temperature range and low self-discharge rate. This technology is now entering the realm of back-up power system storage because of the above reasons.

Pakistan is going to see a great influx of lithium-ion batteries because of the government’s vision to move toward more sustainable forms of energy like renewable energy sources and electric vehicles that can reduce the carbon footprint of the country. If the batteries in such electrical devices are not characterized and tested for their reliability, it may result in a significant amount of revenue lost for both the end user and investor.  

The project aims to address the problem of accurate battery characterization of lithium-ion batteries. There are two aspects of the problem a) diagnostics; where we try to determine the current failures in the device and the current SOH and state of charge (SOC) and b) prognostics; where we try to forecast what will happen in the future for example estimation of remaining useful life of the battery.

In the first phase of the project, our own, NASA and university of Maryland datasets on Li-ion batteries will be used. The parameters(Resistors and Capacitors) of the battery are determined using Thevenin equivalent 2-RC model.  Using a non-linear least square curve fitting tool the parameters, state of charge (SOC) and state of health (SOH) is measured.

The second approach is that we will use our own available dataset on the Li ion battery to build a Thevenin equivalent 2-RC model.  We will be using the pulse discharge test data available. We use neural networks (NN) in deriving a model between Output voltage and state of charge. At the end, we approximate the state of charge (SOC)

This project can help us determine the reliability of batteries and can help in developing a more reliable battery model and reduce the technology life cycle of the battery.

Project Objectives

In today’s world portability is the greatest feature that any technological item can achieve. And to attain portability in any electricity driven device batteries become the most important component. Within batteries, through rigorous trial and testing, Lithium-ion batteries are the emerging leader in the battery rally in the world right now and are used in various applications like Power backup and Electric Vehicles. Around the world many consumer markets have been shifted to battery power because of its eco-friendly nature and its long-term sustainability. One such market is the vehicle industry, an industry notoriously known for its negative effect on the environment, with the introduction of battery powered vehicles, the horizons of sustainability have been substantially broadened, and hence the need to find a sustainable battery to take the load of a motor engine also gives in to the need to have an efficient battery analysis system in order to check, maintain and regulate the state of health and charge of the batteries.

The reliability of battery is vital in these applications. The Lithium-ion battery has gained its status through a detailed process to test the reliability that revolves around battery characteristics such as State of Charge (SOC), State of Health (SOH) and battery modelling concepts. The objectives of this project are summarized below as well:

• To study literature previously available related to Li-ion batteries.

• Use data available to create mathematical models related to Li ion batteries soc and SOH estimation.

• To verify models by generating our own data through experimentation.

• To write and publish research paper.

• A tangible Hardware setup which can model time dependent SOH and RUL of the battery.

Project Implementation Method

The implementation of both approaches is given below:
1) Equivalent Circuit Model to predict State of Health (SOH):

To obtain an estimate of SOH we used the equation (1) to obtain a Qmax which will then be substituted in (2). The model takes an assumption that is the coefficients c0, c1, ...c5 are constant for Constant current charge/discharge cycle. The parameters for akref needs to be computed once for the initial cycle. These parameters then remain constant for the remaining cycles. c0....c5, SOC, and Qmax needs to be evaluated for all cycles.

                                

A constraint on c0 had to be placed where c0 ? [0.5,1.5]. The result was validated and it was proven accurate that beyond this bound the accuracy decreases. Likewise, Qmax ? [Qend, Q0], where Q0 is the max discharge capacity of the cycle, and Qend is the end capacity of the given CC Charge/Discharge cycle. We train on the initial cycle to compute value of a1ref , ...., akref . During the initial cycle we do a 70-30 train-test split. We randomize the data for the initial cycle to avoid over-fitting. However, this created a limitation as the accuracy varied according to randomization. To overcome it we repeated the cycle for a number of iterations and we averaged out the final value. The evolution of Qmax was computed for every cycle by fitting equation (1) to the V-Q profile for the given cycle. The value obtained is then used to compute SOH from equation (3). The model uses non-linear least square curve fitting to approximate SOH for a given cycle.

To validate our model, we tested it on CALCE dataset provided by UMD college park and publicly available NASA Ames Dataset. The SOH approximated is compared with the ground truth provided in the dataset. We then calculate the Root mean squared error as described in equation below:

                      

Where SOHˆ is the predicted value of State-of-health for the given cycle and SOH is the actual value. N is the total number of cycles for a given test dataset.

2) SOC Estimation using Data Driven Technique:

The approach we used to achieve our aim, can be divided into 5 steps.

1. Use MATLAB Simulink to create Thevenin based equivalent circuit model of Li ion battery.

2. Extract battery parameters using pulse discharge.

3. Create an accurate battery model.

4. Use battery model to generate synthetic data and use it to train neural network.

5. Use neural network to estimate state of charge.

Benefits of the Project

Social benefits are a point that is essential to the innovation and application of anything new. The question of how anything affects society is an important one, and it goes hand in hand with the relevance of said development or innovation. Even if something affects society positively, but has little relevance in the broader frame of reference, the innovation becomes futile. Assessing how the concept of Electric Vehicles affects the society of Pakistan has been key in selecting this area of research. Not only has this topic been conclusively beneficiary to the social landscape of Pakistan, but it is also highly relevant globally.

The biggest benefit of having an efficient EV system in our country is its eco-friendly nature. It would be a major step in making Pakistan a nature friendly country in the long run and help reduce pollution and global warming on a large scale. According to a 2019 study Pakistan houses around 24 million vehicles which is a large number when compared to the concentration of areas those vehicles operate in. So, a long-term solution to the pollution caused by these vehicles become more of a necessity than a luxury.

The lowered carbon emissions will help with the climate and the pollution problems major cities face currently. Electric vehicles will also reduce the consumption of non-replenished power resources such as petrol and diesel and in the long run will shift the entire system to a renewable energy driven system.

Now this project in itself leads to all the benefits above but before that it focuses on creating an efficient system for the batteries of the eventual Electric Vehicles. The system that this project aims to aid would create a time efficient sustainable and implementable system according to the social and economic system of the country it targets.

Pakistan is new to the Electric Vehicle industry, and hence there is little or no legal policy that has been drafted to regulate this industry. Now potential problems could arise when required regulations are implemented but until then this industry operates in a relative grey area which is optimal for the establishment of a long-term solution, free from political influence and restrictions.

The use of data driven analysis also sets a precedent for the propagation of efficient and modern analysis techniques. The use of machine learning makes the process time efficient and hence benefits the society at a micro level as the final system would be highly efficient.

Technical Details of Final Deliverable

Non-Linear least square curve fitting toolbox of MATLAB is used on NASA and UMD dataset profiles to extract parameters and evaluate State of Charge and State of Health for the given Cycle. The approach used is an extension of the work already done on 1 RC model.

A battery test bed which will be used to get Dynamic Discharge profiles on Battery Cells. The setup is an interface between Battery Cell, Variable DC load, Programmable DC Supply. The control circuit in this interface uses Relay circuit which is controlled through Arduino Mega 2560. Arduino SD Card module is used for Data Logging.

The Dynamic Discharge Profile will be collected for cycles enough to study the degradation of battery i.e. (approx. 200)

After getting Dynamic Discharge Profiles, we train our Neural Network model using this data and then use this to estimate State of Charge (SOC). This approach is fast and efficient as compared to the conventional approach.

Battery Parameters are extracted using Pulse Discharge Test. 2 Cycles of this test are embedded in the Dynamic Discharge Profiles to capture the transients and model the parameters.

Using the same Dynamic Discharge Profiles, we will be looking at the relation of Battery Parameters with Degradation of the Battery.

The above-mentioned approach will be used on different Cell capacities for accurate modelling and estimation.

Final Deliverable of the Project Hardware SystemCore Industry Energy Other Industries Transportation Core Technology Internet of Things (IoT)Other TechnologiesSustainable Development Goals Affordable and Clean EnergyRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	74700
SD Card	Equipment	3	500	1500
Batteries	Equipment	2	30000	60000
PCB	Equipment	2	500	1000
Wires	Equipment	100	1	100
Ardrino	Equipment	3	700	2100
Stationary	Miscellaneous	8	500	4000
Printing	Miscellaneous	6	1000	6000