State estimation for Battery Management Systems (BMSs) using Artificial Intelligence (AI) is a rapidly growing field of research. It involves using AI algorithms to accurately estimate the State of Charge (SOC), State of Health (SOH), and other parameters related to the health and performance of batteries in electric vehicles, consumer electronics, and stationary energy storage applications. By leveraging advanced machine learning techniques such as deep neural networks, researchers can develop models that can predict SOC and SOH with high accuracy while also providing insights into how different operating conditions affect battery performance over time. This technology has enabled more efficient utilisation of batteries by allowing users to better understand their current state and plan for future usage scenarios. As AI technologies continue to advance, state estimation is expected to become an increasingly important tool for managing battery life cycles in various applications.
Researchers have developed effective state estimation techniques using AI by combining physics-based and empirical-based modeling techniques, such as discriminative support vector machines and Recurrent Neural Networks (RNN). These techniques aid in estimating state operational variables and utilize online-gathered process-level variables, along with their own estimated internal nodes, while controlling parameterized changes offline. This approach, therefore, forms data streams that will allow the use of multi-objective optimisation techniques, leading towards improved performance and safer operating conditions for each BMS function.
The use of AI for state estimation for Battery Management Systems is undoubtedly advantageous and can benefit process and safety critical systems. Some recent innovative solutions for state estimation using AI for BMS operation are given in the section below.

Coral IoT: Advanced Analytics for State Estimation in BMSs

Coral IoT, headquartered in the US, is an innovative technology firm specializing in taking advantage of AI and the Internet of Things (IoT) to create a unique solution for state estimation for BMSs. Utilising advanced analytics, Coral IoT creates estimates regarding energy production and storage as well as provides forecasts for future BMS performance needs over time. This helps in making better decisions on capacity planning with increased accuracy while still minimising risk regarding component failure and potential operating issues that unexpected environmental conditions can cause.

What sets Coral apart from other organisations regarding Battery Management Systems is its prediction capabilities. By leveraging AI-driven algorithms, they can accurately predict future behaviours, which gives improved insight into lifetime efficiency gains across systems such as solar panels and electric vehicles. Along with real-time status monitoring through online dashboards, users have intimate knowledge of the performance and efficiencies of the battery systems and even ageing over time, allowing them greater flexibility. Reporting visualisation also includes heat maps to help identify disparities related to predicted and real regeneration levels. The developers have heavily emphasized minimizing upfront investment and running costs in the development of the system.

Cutting-Edge AI Algorithm for State Estimation in Battery Management Systems

Tansai Technologies is a leading BMS provider that has implemented an innovative and cutting-edge AI algorithm for state estimation in Battery Management Systems. Tansai’s proprietary AI techniques focus on precisely predicting the status of batteries using its signature algorithms, even when only limited data is available. This enables the achievement of accurate information regarding battery health with minimal input and resources. The true power lies in its accuracy and speed. Research shows that this system can estimate states up to 50 times faster than traditional methods, drastically reducing time spent undertaking maintenance checks during operation or post-mortem analysis following failure events. Furthermore, their modelling approach benefits from excellent repeatability since, once trained properly, it guarantees excellent performance under all conditions, even when unexcepted interruptions occur.

Additionally, Tansai Technologies offers two major advantages over other competitive offerings:

1. It provides a more flexible tuning capability, allowing users to access raw values side by side in a clear and concise format.
2.The datasets used here are lightweight, enabling training and real-time execution almost simultaneously. This contrasts with data-dense matrices that require manipulation before sequential modeling and prediction can be undertaken.

Tesla: AI-Powered State Estimation for BMSs

The Turkish company Eatron Technologies has created an automobile-grade BMS that is AI-ready. To ensure precise and trustworthy data on the status and functioning of the battery, BMSTAR® combines highly accurate battery models with AI and sophisticated estimation techniques. The edge-based BMSTAR® software intends to include a cloud counterpart that incorporates built-in connection and analytics. These allow for continual and adaptive software changes as well as over-the-air updates to provide greater performance and reliability (for example, for SOC and SOH) throughout the vehicle’s lifetime.

The software accomplishes this through AI-powered and physics-based algorithms, which apply to both low-voltage and high-voltage Battery Management Systems. Its AI-based functions estimate the remaining time until the end of the useful life, and users can utilize it for both cloud-based fleet data analysis and onboard the vehicle as part of the BMS. It also guides Original Equipment Manufacturers (OEMs) and their cell suppliers on the real-life fleet performance of batteries. Furthermore, it analyzes leading indicators to enable the diagnosis of battery cell issues months in advance. Additionally, it utilizes sophisticated data augmentation techniques to support the training of machine learning models. Users can use the created models to run in real-time for in-vehicle applications and in the cloud for large-scale data analysis.

AI-Powered State Estimation for Tesla’s BMSs

Tesla, US, uses AI to develop a state estimation system for their Battery Management Systems. This system uses AI algorithms to accurately estimate the battery pack’s SOC, temperature, and other parameters in real-time. The AI algorithms train on data collected from Tesla’s fleet of vehicles, enabling them to continuously improve accuracy over time. The unique selling point of Tesla’s approach is that it allows for more accurate predictions than traditional methods, such as Kalman Filters or Extended Kalman Filters, which rely on linear models and assumptions about the environment.

By leveraging deep learning techniques, Tesla can make more accurate predictions even when faced with nonlinear dynamics or unexpected environmental conditions. Additionally, this approach enables faster response times since it does not require manual tuning or calibration as traditional approaches do. Finally, Tesla ensures that its estimates are tailored specifically to its products and use cases by training its AI models on data from its fleet of vehicles, rather than relying on generic models developed by third parties.

What Next?

With the development of big data, intelligent algorithms, and cloud platforms, a trend of smart and networked BMSs is becoming increasingly obvious, which will effectively improve the battery state estimation accuracy and thus improve the life and safety of batteries. New (and in some cases, current) research will focus on intelligent sensing using IoT, model and signal enhancement, cloud-to-edge-based state estimation, battery state estimation under vehicle network interaction, and life cycle intelligent management.
Future research on state estimation for Battery Management Systems will also focus on developing digital twin models to automate and optimise the BMS state estimation process by utilising machine learning algorithms and cloud computing. By recording high-quality data from the battery system under test, researchers can create high-fidelity empirical models of attributes such as charge/discharge rate, voltage, current, state of health (SOH), and state of charge (SOC). These models can then be used to predict these attributes for thousands of operational scenarios without extensive physical testing. In this case, creating a digital twin of a battery management system can save time and effort in the development process as compared to physical testing.