Data-driven battery degradation study
< Project overview >
This project, with the University of Sheffield working in collaboration with an energy resource management company, focussed on data-driven battery energy storage health estimation and prediction. The knowledge exchange and feasibility parts of this project used the project lead’s expertise in battery energy storage systems to establish the challenges and requirements of using the company’s data for the estimation of battery states including prediction of battery cell degradation.
Another aspect of the project, undertaken at the University of Oxford, was the further development and testing of an open-source database system for battery test data management.
The project aimed to deliver methods for estimating battery states that the company would evaluate on its data sets. These methods have been developed using the Sheffield project lead academic’s research on the 2MW/1MWh Willenhall battery energy storage system and data collected on battery cells in the lab.
The company’s data sets are commercially sensitive (being effectively the property of the company’s customers) which therefore presents a barrier for researchers looking to have full access to data in order to develop their algorithms.
In this project, the aim was to develop a working methodology to enable Sheffield to provide the company with packaged code to run against their data with only the validation data shared back with Sheffield. It is critical for the battery estimation research to validate the algorithms against different batteries and new data sets to ensure the methods used are not limited to success only on the Willenhall battery.
Large batteries cost millions of pounds, a prohibitive cost for the University of Sheffield; this project enables access to the additional data required for this work.
Finally, to facilitate data sharing between HEIs for algorithm validation, improvements will be made to a flexible, interoperable open-source database software developed by Oxford for storing laboratory battery test data.
What was done?
This project enabled different state estimation methods to be validated against a range of different battery systems whilst meeting the requirements that underlying datasets cannot be shared outside the company. Using packaged Python code, algorithms were delivered to the company with training on how to use them and guidance on the expected data formats. Company engineers extracted the appropriate data sets and evaluated the results output from the code
Based on the feedback given by the engineers, development of the algorithms was carried out through an iterative process. At Oxford, the database system to simplify lab battery testing was rewritten to enable easy installation and metadata to be added through an intuitive web interface, and a network of international collaborators has been built up around this.
For Sheffield to evaluate the effectiveness of its battery state estimation algorithms it is essential to validate against a wide variety of energy storage assets. The company has produced validation data against a number of assets using the minimum viable data streams. This has shown good results and enabled Sheffield to demonstrate the potential improvements in accuracy that could be achieved through the capture of additional data streams from the storage assets.
The working methodology enabled the algorithms to be run without sharing the data; however, there were challenges interpreting unexpected results without the full debugging capabilities that become available when in possession of the data.
Sheffield has had the opportunity to develop sensitive-data collaborative methods, establish the criteria for use and assess their limitations. The database system developed at Oxford now uses a React JS interface to enable easy web-based configuration and a REST API was written to enable interoperability with other projects around the world, such as batteryarchive.org.
Deliverables and outputs
Source code (Python) to demonstrate the battery state estimation against data sets.
Software for the collection of data from lab-based equipment and central repository.
The project will have an impact on future product development by the company. The potential to estimate and predict battery states has been shown and through collaboration with Sheffield the additional asset data streams that would be required to achieve this successfully have been identified. The database work has resulted in an ongoing open collaboration with various US National Labs, RWTH Aachen and others.
During the project the company was purchased by a multinational energy supplier. There is no immediate commitment to future activity with the company at present but we hope to develop a strong relationship with the new owners.
Sheffield is continuing its research into battery state estimation algorithms and is preparing to use the algorithm-sharing methodologies developed in this project directly with battery asset owners.
Oxford is continuing development on the database system, and is building a community around it.
The challenges to development with commercially sensitive data were overcome by the working method.
At the beginning of the project a series of online knowledge-exchange sessions were provided by Sheffield to enable the company’s engineers to evaluate the results.
At Oxford we learnt to start with user needs (and adapt requirements throughout), involve professionals (research software engineering) and build wider links internationally (it turns out that many others around the world have been working to address a similar challenge).
It would have been useful to have at least one of the company’s clients involved with the project so that one of the data sets could have been shared. This would have helped with the debugging and analysis.
On the database side, we were somewhat limited by the available software engineering resource at the beginning of the project, but this was addressed later. Professional software expertise was supplied by the Research Software Engineering (RSE) team at Oxford. The development of professionally engineered and crafted software enhances the end-user experience which is critical to any eventual further exploitation outside academia.
What has Pitch-In done for you?
This funding has provided the researchers with the opportunity to test their algorithms on significant data sets that would otherwise not be possible. The effort put into packaging the code for use outside of the research lab means that it will be possible to collaborate with more asset owners in the future.
Professor Dan Gladwin – University of Sheffield
An energy resource management company