Understanding battery systems presents a complex modelling paradigm mathematically and computationally. After highlighting the need for better battery management systems, this article presents the current limitations and challenges, and discusses the prospective future for artificial intelligence to be used within the battery industry.
The need for better battery management systems
The energy storage industry is currently going through a technological renaissance as the perturbations of the Electrical Vehicle (EV) sector continue to snowball and increase demands for battery management modelling. To add gravity, the EV sector has a concoction of environmental and economic drivers catapulting the market into the forefront of energy storage, providing significant incentives to accelerate the research of battery management systems. The current evolution of government policies is in favour of constricting CO2 emissions and controlling the collateral damage caused by environmental pollution. Consequentially there have been strong investments seen in the EV sector and the most significant demand for increasing battery range, and augmented cooling systems. Due to these factors, the force applied to battery research to increase the accuracy and efficacy of management systems has also grown exponentially.
The current research focuses on Lithium-ion chemistry due to the advantages posed relative to other available chemistries:
-Lithium-ion batteries operate at higher voltages generating high power density.
-Lithium-ion batteries have a low self-discharge rate relative to other chemistries.
-Lithium-Ion batteries have a wide operating temperature range whilst also being compact allowing for a longer lifespan.
-Lithium-Ion batteries operate using an intercalation system, where the cathode and anode structures allow for reversible insertion and extraction of Lithium ions.
-Lithium has a high electrochemical potential.
The current limitations and challenges
However, as it stands, EVs and the current existing chemical and physics-based models are confronted with a range of drawbacks. Regarding vehicle battery modelling, a range of models exist designed to return diagnostics of the battery, predicting for example voltage, current, battery temperature, overpotential, state of charge, state of health, and so forth. Although these variables are critical to battery longevity and performance, the existing models still harbour a degree of inaccuracy due to the non-linearity of the governing physics and chemistry of batteries. This is especially true for longer term prediction, which clashes with the current drive from EV to increase battery life.
There exists a complex and dynamic challenge intertwining electrochemistry and microscale physics to produce battery performance predictions. The modelling complexity arises due to the technical nature of model parameterization: the model must account for dependencies that exist within the macro and micro scales. For example, lithium diffusivity and electrical conductivity have a dependence on lithium concentration. The technical multiscale dependencies weave together a multiplex modelling challenge for battery models. The infrastructure of battery research is therefore not parsimonious in the quantity of tests delegated and conducted.
Why is AI a good candidate for the battery industry?
Compounding the model sensitivity and coupled non-linear complexity there exists a logical drive for a machine learning approach. Interchangeable physics battery modelling libraries present a gulf of data for machine learning to digest and optimize. This is further augmented by the fundamental nature of advanced deep learning techniques enabling the discovery of higher dimensional relationships which are difficult to parameterize.
Machine learning approaches to battery modelling can produce key diagnostics with relatively high accuracy whilst also computing at an accelerated rate. Therefore, serving as an enhanced simulation environment for engineering users who wish to conduct experimental tests or optimizations. AI capabilities can also enable battery research users to explore analytical relationships within the cell infrastructure which can aid in reducing the time taken for producing optimal performing batteries.
How can AI augment the battery industry?
Machine learning models thrive with access to large datasets due to the mathematics governing the algorithms, thus, enabling the large datasets that are gathered through battery or vehicle drive tests to be effectively integrated. Real-time data that is derived from vehicle driving tests contains information such as the acceleration, ambient temperature, velocity, state of charge, all of which are a dynamic function of time. Understanding the relationships that exist quantitively between these variables and each other with time stimulates a need for AI application in this industry. The attraction for AI here exists as it may contribute to understanding complex higher-dimensional relationships which are present between the features, and as a result leading to a greater understanding of the ground truth physics.
Battery testing equipment produces substantial amounts of data that machine learning models can use to decipher relationships. Understanding the interrelation between the physics defining a battery model allows for intelligent iterations to be made to physical models, which in turn can optimize battery performance. The current trajectory for the integration of AI and battery physics modelling is to enhance the user’s ability to intelligently engineer changes within the battery cell. The perturbations of intelligent modification and refinement of a battery system are to optimize battery performance. The additional advantage of using AI with physics-based modelling comes from the non-linearity and complexity of the governing chemistry and physics on the atomistic scale. This high degree of parameterization presents mathematical friction to modelling. Deep learning techniques therefore can be supplementary in producing digital battery twins sometime in the future.
AI opportunities for the battery industry: Conclusion
More than ever, there is a need for performant and long-lasting batteries, mainly driven by the EV industry. However, the complexity of the physics of such systems means that battery performance models are not accurate enough and that lots of tests are still required to design optimal battery systems. This is where AI comes into place. By capturing higher dimensional relationships from data gathered from driving tests and other physical tests, AI can better inform battery designers on what their next step should be to design robust and performant batteries.