Automated battery database with innovative Visualization Tools for Big Battery Aging Data

AI Quick Summary

The focus of this project will be the development of a battery aging database with an analytical graphical user interface and innovative dashboard tool that provides comprehensive visualization of the complex landscape of battery aging. Based on the currently developed automated battery database by Battronics AG, these visualization tools will be developed by an interdisciplinary consortium of OST and Battronics. The goal is to visualize the battery aging for the Swiss and European Battery Value chain. Today, each battery company needs to test battery aging. That means that ca 40-70 companies test the same e.g., LG Chem battery cell and need to evaluate the 2-3 years aging data by educated people. This is very time consuming and costly for each company.We are currently in the process of developing a battery aging database where instead of 40-70 companies measuring similar data each time, we provide a detailed dataset and advanced battery aging prediction tools. This effort will be further supported by adding an adaptive user interface (AUI) that will learn from expert user decisions, how to present battery data and visualizations most effectively. These adaptive user interfaces will be based on current state-of-the-art knowledge visualizations and graphical presentations in battery research. Currently, cycle life analysis is described through a series of graphs, 2D plots representing the battery degradation over time (see Fig.1a)), 3D plots representing the temperature vs. state of charge (see Fig.1b)) and plots representing the electrode potential and state of charge (see Fig.1c)). These plots allow the battery experts to predict our use-cases: (1) the optimal battery material for a novel product, (2) life expectancy of a battery model and (3) the risk of a sudden death.Here we aim to: (A1) develop adaptive user interfaces to manipulated and present battery data (A2) deal with the big-data of the battery-cells for online presentation (A3) understand most acceptable user interface to analysis batteries to determine (1)-(3) (A4) optimize the presentation of battery data to improve the analysis reliability and reduce uncertainty in case of (1)-(3) In the current workflow, as presented in Fig. 2.a, all battery data plots are created manually and offline in an on-demand fashion and prepared by expert-analysts. This comes with high cost and slow information transfer. Keep in mind that for a publication usually programs like Origin & MATLAB are used where the whole publication process of such a paper needs at least 8 weeks to analyze and plot all data plus another 2-3 weeks to plot the graphs so that the quintessence is displayed in one single plot. And especially in Chemist programs like Origin, a new analyzed battery cell means directly another 10-11 weeks for evaluation with maybe 3 weeks savings from human experience gain from previous data analysis. This amount of time would not be justified in industry where we need to analyze many different cells and often even compare small differences btw them, e.g., Fig 1a at 25°C shows only 2% capacity loss difference after 10months of aging test. So, differences btw cells are small, but very important for industry, however it takes a long time to evaluate today by hand of experienced scientists even with a PhD. Thus, we strive to provide the required plots from the database provided by Battronics automatically through our adaptive interfaces (see Fig. 2.b.1) (A1).One dataset of a battery-cell contains approximately 650.000 data points with 15 to 124 cells tested per dataset [Severson,2019], for comparing different cells in the case of (1) we assume that we will need approximately 10-80 Mio data points at one given time. This provides us with the challenge of handling big-data though our web-service. Our data service is required to allow not only static display of the data plots to allow experts to determine solutions for (1)-(3) (A2), but also to change their display and evaluate different 2 and 3D graphs for small and often zoom features. To ease the work of the experts AUI’s are required that are easy to use and need to represent the data in an optimal way. Therefore, we will analyze user acceptance of our user interfaces both through mouse tracking and eye-gaze tracking of experts and novice users (A3). Once created these plots are used by experts to determine the usefulness of a given battery series regarding the issues (1)-(3). When determining the ideal battery solution for a novel product the expert is considering multiple parameters and uses his/her expert knowledge to select the optimal battery type. During this process experts focus on particular aspects of the presented graphics, which can provide insights into their decision making [Orquin, et al., 2013; Smith, et al., 2009]. Therefore, we are collecting physiological data like eye gaze, blinking rate and cognitive load during the interaction of expert and novice users with our interfaces to understand their reasoning and optimization process as well as how to get quickest to a decision based on the provided data (see Fig. 2.b.2). Here the expertise of Prof. Lohan will provide a significant benefit towards the successful data collection and analysis [Ahmad, et al., 2020; Hastie, et al. 2018]. This data in conjunction with machine learning techniques and pattern recognition is the basis for the adaptivity of our interfaces (see Fig. 2.b.3) [Henning 2019] (A4)The total addressable market of Li-ion batteries was $44bn dollar in 2020 and expected to growth with CAGR of 16% which means doubling of market every 4.5 years projected by MarketsandMarkets [Markets 2020]. The analytics and aging market will cover ca 14% of the total market and grow together with the total market thus being ca. $6.2bn.Our customer base is pure B2B, with customers in the battery industry value chain and more specifically our co-members of the European Battery Alliance (EBA250). The most upstream customers are in the mining industry. But most of our customers are players in the field of active material production, cell manufacturing and the final OEM/application companies, which need our data the most. Here also the adaptive visualization of battery aging data will come in as it directly serves our customers' demand for massive and reliable battery aging data and comparison + fast visualization btw different cell types. Acquisition of new customers are planned to be approached by our established direct acquisition channel and via on- & offline conference and trade fairs.The CO2 equivalent of an electric vehicle is based on the energy necessary for production of the car itself, the battery pack, and the type of energy (renewable, fossil, nuclear) used for charging during use. Even in countries like Switzerland, where most of energy in the grid comes from renewable sources, a large part of the CO2 footprint of the electric vehicle is based on the battery itself.Even though the CO2 footprint strongly depends on the country in which the cells are manufactured (as shown in Fig. 3), the most influential factor is the aging of the battery and therefor the expected cycle life, since it determines how long a specific battery pack or cell can be used before its capacity has degraded to a point, where the usage in an electric vehicle is no longer feasible. At that point, either the battery pack or the entire car (in case of non-removable battery packs) must be replaced, and the CO2 footprint of the new battery must be accounted for.The influence of the aging mechanism on the CO2 footprint of batteries can be well described with the Nissan Leaf as an example. While the old version of the car only had a battery capacity of 24 kWh, it showed good capacity retention by dropping to 80% capacity after roughly 5.5 years. The newer, upgraded version, has a significantly increased battery capacity of 30 kWh, but reaches its end of life after ~2.4 years, due to changes in chemistry and corresponding aging behavior. This “upgrade” of the battery sacrifices life span for driving range, resulting in an overall increase of carbon footprint. At an estimated production cost of 106 kg CO2 per kWh battery, 30 and 24 kWh capacity per battery pack respectively and ~112000 models sold per year, the CO2 production per year increases from:24 kWh model: (106 kg(CO2)/kWh * 112000)*(24 + 72)kWh /5.6 = 203’520 t CO230 kWh model: (106 kg(CO2)/kWh * 112000) * (30 +72) kWh /2.4 = 504’560 t CO2This shows an increase in CO2 output of 301’040 tons per year, due to worse battery aging. Since the battery of a Nissan Leaf cannot be swapped, the difference becomes so significant.This example clearly shows that a better understanding of aging mechanisms in battery cells is necessary to further reduce the carbon footprint of the related industries.Even though the difference in carbon footprint is different based on difference between models and manufacturers, it is far from optimal and needs improvements, like our database.As part of the Innosuisse project 48394.1 data collection for the database has already started and we are working on structuring and classification of the obtained data. A pilot SQL database is already in use and can answer queries on the intranet servers.The ongoing and future work over the next two years can be divided into several work packages:1. Calculator for linear aging & visualization developmentDeveloping and coding the linear aging calculator based on physical equation set and verification of correct dissection to fit the error corrected data from WP2. Also including standard fitting by linear and sqrt(t) least-square fitting procedure to allow customers use their standard technique. Implementation of physics-based algorithm to error-corrected data and dissection into individual aging mechanisms internally in order to use them for forecasting.2. Machine-learning for non-linear aging predictionThe heart of the Aging calculator is the machine-learning based calculator for non-linear aging onset prediction with training & verification on all database data. Final results will be used with reimplementation the experimental errors into the corrected data, that customers can compare their experimental data directly with Battronics Aging Calculator and directly compare with earlier linear, sqrt, physics-based calculation.3. Battery recommender & reverse searchIn order to allow users also to search for the best available battery for a given requirement set, we implement a reverse-search algorithm which uses the original database to list all cells that fulfill a certain aging requirement at a specific rate, temperature, State-of-charge, death-of-discharge. This uses already direct data from database plus the prediction from physics-based aging calculator for linear aging part and the predicted onset of non-linear aging calculator for non-linear onset as end-of-life criterion as cells in packages would diverge from each other4. Prototype design and user testingThe interfaces for battery analysis will be derived, implemented, and evaluated. User test groups will be involved in the iterative improvement of the prototypes.5. Framework/Interface developmentMost software development will be done in WP4, both prototypes will be implemented here. The multi-tier architecture, from Database to the Adaptive User Interface, is implemented.6. Numerical analysis toolkit measurements and validationThe numerical analysis toolkit will be designed and adapted to most realistic user requirements tested in the previous two work packages.7. User-centric task-based measures and validationDevelopment and implementation of user testing for feedback and adaptive interface training and adjustment Since all these tasks require the involvement of specially trained personnel (data analysts, IT specialists), the estimated cost of the project will be ~400000 CHF. Since the possibility in CO2 reduction from the use of this database would be massive, a financial contribution by the Energy Lab would be greatly appreciated.Video pitch

Project Idea Metadata

Project Idea Name: Automated battery database with innovative Visualization Tools for Big Battery Aging Data
Date: 3/2/2022 3:41:45 PM
Administrators:
- Dr. Michael Hess

Project Idea Description

Today, each battery company needs to test battery aging. That means that ca 40-70 companies test the same e.g., LG Chem battery cell and need to evaluate the 2-3 years aging data by educated people. This is very time consuming and costly for each company.

We are currently in the process of developing a battery aging database where instead of 40-70 companies measuring similar data each time, we provide a detailed dataset and advanced battery aging prediction tools. This effort will be further supported by adding an adaptive user interface (AUI) that will learn from expert user decisions, how to present battery data and visualizations most effectively. These adaptive user interfaces will be based on current state-of-the-art knowledge visualizations and graphical presentations in battery research.

Currently, cycle life analysis is described through a series of graphs, 2D plots representing the battery degradation over time (see Fig.1a)), 3D plots representing the temperature vs. state of charge (see Fig.1b)) and plots representing the electrode potential and state of charge (see Fig.1c)). These plots allow the battery experts to predict our use-cases: (1) the optimal battery material for a novel product, (2) life expectancy of a battery model and (3) the risk of a sudden death.

Here we aim to:

(A1) develop adaptive user interfaces to manipulated and present battery data

(A2) deal with the big-data of the battery-cells for online presentation

(A3) understand most acceptable user interface to analysis batteries to determine (1)-(3)

(A4) optimize the presentation of battery data to improve the analysis reliability and reduce uncertainty in case of (1)-(3)

In the current workflow, as presented in Fig. 2.a, all battery data plots are created manually and offline in an on-demand fashion and prepared by expert-analysts. This comes with high cost and slow information transfer. Keep in mind that for a publication usually programs like Origin & MATLAB are used where the whole publication process of such a paper needs at least 8 weeks to analyze and plot all data plus another 2-3 weeks to plot the graphs so that the quintessence is displayed in one single plot. And especially in Chemist programs like Origin, a new analyzed battery cell means directly another 10-11 weeks for evaluation with maybe 3 weeks savings from human experience gain from previous data analysis. This amount of time would not be justified in industry where we need to analyze many different cells and often even compare small differences btw them, e.g., Fig 1a at 25°C shows only 2% capacity loss difference after 10months of aging test. So, differences btw cells are small, but very important for industry, however it takes a long time to evaluate today by hand of experienced scientists even with a PhD. Thus, we strive to provide the required plots from the database provided by Battronics automatically through our adaptive interfaces (see Fig. 2.b.1) (A1).

One dataset of a battery-cell contains approximately 650.000 data points with 15 to 124 cells tested per dataset [Severson,2019], for comparing different cells in the case of (1) we assume that we will need approximately 10-80 Mio data points at one given time. This provides us with the challenge of handling big-data though our web-service. Our data service is required to allow not only static display of the data plots to allow experts to determine solutions for (1)-(3) (A2), but also to change their display and evaluate different 2 and 3D graphs for small and often zoom features.

To ease the work of the experts AUI’s are required that are easy to use and need to represent the data in an optimal way. Therefore, we will analyze user acceptance of our user interfaces both through mouse tracking and eye-gaze tracking of experts and novice users (A3).

Once created these plots are used by experts to determine the usefulness of a given battery series regarding the issues (1)-(3). When determining the ideal battery solution for a novel product the expert is considering multiple parameters and uses his/her expert knowledge to select the optimal battery type. During this process experts focus on particular aspects of the presented graphics, which can provide insights into their decision making [Orquin, et al., 2013; Smith, et al., 2009].

Therefore, we are collecting physiological data like eye gaze, blinking rate and cognitive load during the interaction of expert and novice users with our interfaces to understand their reasoning and optimization process as well as how to get quickest to a decision based on the provided data (see Fig. 2.b.2). Here the expertise of Prof. Lohan will provide a significant benefit towards the successful data collection and analysis [Ahmad, et al., 2020; Hastie, et al. 2018].

This data in conjunction with machine learning techniques and pattern recognition is the basis for the adaptivity of our interfaces (see Fig. 2.b.3) [Henning 2019] (A4)

The total addressable market of Li-ion batteries was $44bn dollar in 2020 and expected to growth with CAGR of 16% which means doubling of market every 4.5 years projected by MarketsandMarkets [Markets 2020]. The analytics and aging market will cover ca 14% of the total market and grow together with the total market thus being ca. $6.2bn.

Our customer base is pure B2B, with customers in the battery industry value chain and more specifically our co-members of the European Battery Alliance (EBA250). The most upstream customers are in the mining industry. But most of our customers are players in the field of active material production, cell manufacturing and the final OEM/application companies, which need our data the most. Here also the adaptive visualization of battery aging data will come in as it directly serves our customers' demand for massive and reliable battery aging data and comparison + fast visualization btw different cell types.

Acquisition of new customers are planned to be approached by our established direct acquisition channel and via on- & offline conference and trade fairs.

The CO₂ equivalent of an electric vehicle is based on the energy necessary for production of the car itself, the battery pack, and the type of energy (renewable, fossil, nuclear) used for charging during use. Even in countries like Switzerland, where most of energy in the grid comes from renewable sources, a large part of the CO₂ footprint of the electric vehicle is based on the battery itself.

Even though the CO₂ footprint strongly depends on the country in which the cells are manufactured (as shown in Fig. 3), the most influential factor is the aging of the battery and therefor the expected cycle life, since it determines how long a specific battery pack or cell can be used before its capacity has degraded to a point, where the usage in an electric vehicle is no longer feasible. At that point, either the battery pack or the entire car (in case of non-removable battery packs) must be replaced, and the CO₂ footprint of the new battery must be accounted for.

The influence of the aging mechanism on the CO₂ footprint of batteries can be well described with the Nissan Leaf as an example. While the old version of the car only had a battery capacity of 24 kWh, it showed good capacity retention by dropping to 80% capacity after roughly 5.5 years. The newer, upgraded version, has a significantly increased battery capacity of 30 kWh, but reaches its end of life after ~2.4 years, due to changes in chemistry and corresponding aging behavior.

This “upgrade” of the battery sacrifices life span for driving range, resulting in an overall increase of carbon footprint. At an estimated production cost of 106 kg CO₂ per kWh battery, 30 and 24 kWh capacity per battery pack respectively and ~112000 models sold per year, the CO₂ production per year increases from:

24 kWh model: (106 kg(CO₂)/kWh * 112000)*(24 + 72)kWh /5.6 = 203’520 t CO₂

30 kWh model: (106 kg(CO₂)/kWh * 112000) * (30 +72) kWh /2.4 = 504’560 t CO₂

This shows an increase in CO₂ output of 301’040 tons per year, due to worse battery aging. Since the battery of a Nissan Leaf cannot be swapped, the difference becomes so significant.

This example clearly shows that a better understanding of aging mechanisms in battery cells is necessary to further reduce the carbon footprint of the related industries.

Even though the difference in carbon footprint is different based on difference between models and manufacturers, it is far from optimal and needs improvements, like our database.

As part of the Innosuisse project 48394.1 data collection for the database has already started and we are working on structuring and classification of the obtained data. A pilot SQL database is already in use and can answer queries on the intranet servers.

The ongoing and future work over the next two years can be divided into several work packages:

1. Calculator for linear aging & visualization development

Developing and coding the linear aging calculator based on physical equation set and verification of correct dissection to fit the error corrected data from WP2. Also including standard fitting by linear and sqrt(t) least-square fitting procedure to allow customers use their standard technique.

Implementation of physics-based algorithm to error-corrected data and dissection into individual aging mechanisms internally in order to use them for forecasting.

2. Machine-learning for non-linear aging prediction

The heart of the Aging calculator is the machine-learning based calculator for non-linear aging onset prediction with training & verification on all database data. Final results will be used with reimplementation the experimental errors into the corrected data, that customers can compare their experimental data directly with Battronics Aging Calculator and directly compare with earlier linear, sqrt, physics-based calculation.

3. Battery recommender & reverse search

In order to allow users also to search for the best available battery for a given requirement set, we implement a reverse-search algorithm which uses the original database to list all cells that fulfill a certain aging requirement at a specific rate, temperature, State-of-charge, death-of-discharge.

This uses already direct data from database plus the prediction from physics-based aging calculator for linear aging part and the predicted onset of non-linear aging calculator for non-linear onset as end-of-life criterion as cells in packages would diverge from each other

4. Prototype design and user testing

The interfaces for battery analysis will be derived, implemented, and evaluated. User test groups will be involved in the iterative improvement of the prototypes.

5. Framework/Interface development

Most software development will be done in WP4, both prototypes will be implemented here. The multi-tier architecture, from Database to the Adaptive User Interface, is implemented.

6. Numerical analysis toolkit measurements and validation

The numerical analysis toolkit will be designed and adapted to most realistic user requirements tested in the previous two work packages.

7. User-centric task-based measures and validation

Development and implementation of user testing for feedback and adaptive interface training and adjustment

Since all these tasks require the involvement of specially trained personnel (data analysts, IT specialists), the estimated cost of the project will be ~400000 CHF. Since the possibility in CO2 reduction from the use of this database would be massive, a financial contribution by the Energy Lab would be greatly appreciated.

Video pitch