In today's and future sustainable mobility, electric powertrains play a pivotal role. Among all the components of electric vehicles, the battery holds the highest value. For manufacturers and mobility providers, the competition is determined by the total cost of ownership (TCO). To balance the reduction of TCO and battery costs while ensuring optimal performance, range, efficiency, and most importantly, lifetime, the continuous monitoring of the battery during operation is inevitable.

Many organizations, including Electra Vehicles, are using AI/ML to deliver greater business value as the ever-evolving technology plays an increasingly vital role in data communication and management. Join us to learn how Electra’s Adaptive Digital Twin technology works with the power of AI/ML to unlock the next big innovations that are moving the energy and EV industries forward at record pace.

Lithium metal batteries using solid electrolytes show superior performance potential, but only stringent quality control will prevent issues like dendrite growth and explosions. This study presents batteryNet, an iterative residual U net-based deep learning model tested on HPC NERSC’s Perlmutter for detecting lithium plating dynamics in solid-state batteries using in-operando x-ray computed tomography. The algorithm converts high-resolution x-ray data from DOE experimental facilities into detailed measurements, identifying battery defects and quantifying durability. This technique also evaluates new polymer electrolyte designs under investigation at UC Irvine’s National Fuel Cell Research Center, marking a significant stride toward battery inspection and safety validation.

This presentation will describe advances in machine learning-based techniques for addressing systems problems that arise for lithium-ion batteries. The specific systems problems include the prediction and classification of battery cycle lifetime (aka remaining useful life) and the determination of optimal charging protocols. The development of the techniques and their application are in collaboration with materials science, physics, and computer science researchers at Stanford University, Toyota Research Institute, and MIT.

We demonstrate an early prediction model with reliable uncertainty estimates, which utilizes an arbitrary number of initial cycles to predict the whole battery degradation trajectory. Our model will enable accelerated battery development via uncertainty-guided truncation of cell cycle experiments once the predictions are reliable.

A fast, accurate, and reliable SOH diagnostic algorithm based on a combination of time-domain diagnostics, parameterization, and machine learning (ML)-based regression and optimization will be presented. This unique algorithm can be utilized for a variety of applications such as mobile, stationary, marine, aviation, and EVSE, etc.

Model-based Battery Management System (BMS) can be used to improve the performance of current and next-generation lithium batteries. Recent results for life improvement, reduction in charging time for cells, packs, and modules, and next-generation batteries will be presented. The importance of physics-based electrochemical models, suitable algorithms, and experimental validation will be discussed.

It’s go time for the battery sector. Companies across the value chain are racing to secure materials supply, ship new products, ramp new gigafactories, and optimize post-sale value for customers. Missteps at this stage can mean delayed revenue, costly reengineering, or safety incidents that can risk lives and trigger huge recalls. To succeed, companies must learn faster than the competition. Many organizations now understand that enterprise battery intelligence offers a path to world-class battery excellence, but consider it a nice-to-have until more urgent initiatives are addressed. This talk will illuminate how companies can’t afford to wait, as battery intelligence delivers immediate ROI and is in fact the key to accelerating in the midst of developing supply chains, evolving chemistries, and ever more demanding applications.

This talk will discuss recent research in battery modeling, focusing on diagnostics from field data, including combining of machine learning and circuit models to allow flexibility in fitting from data while retaining the transparency of physical models. It will conclude with some thoughts on how data-driven models can accelerate progress in batteries.

How can the time-to-market of new batteries and electrical vehicles shortened? Testing hundreds or thousands of batteries and cells simultaneously in a multi-vendor lab is a huge challenge and time consuming. Efficient processes and an open toolchain for automated scheduling, monitoring, energy management, and data analytics are the key to success as part of that virtual testing can cut development time and cost, e.g. with model based cell aging prediction or AI based battery fleet data modelling.

Despite their merits, lithium-ion batteries still face significant performance and safety bottlenecks. Physics-informed machine learning has proven as a useful way to make batteries work better, and live up to their potential. In this talk, we will share our explorations on this topic, especially for modelling and condition monitoring for lithium-ion battery systems. We will further discuss prospective opportunities and challenges.

Health prognostics are essential for smarter battery management. It is difficult to guarantee performance under disparate distributions of lifetime and degradation patterns however, due to a variety of battery types and application scenarios. The effectiveness of various transfer learning strategies for performance improvement in battery health prognostics will be discussed in light of the accessibility of operating data and capacity labels. The fundamental concept and the applicability of various transfer learning strategies will be shown. Specifically, the sparsely labeled data fine-tuning, unsupervised domain adaptation, and no source labeled data condition-driven self-supervised strategy will be discussed in this presentation.

This study presents an advanced strategy for State of Charge (SOC) and State of Health (SOH) estimation that has achieved errors of less than 1%. This strategy includes combined spectral and temporal characterization of cells. It uses the Smooth Variable Structure Filter together with the Interacting Multiple Model concept for estimation.

Lithium-ion (LIB) battery degradation is often characterized at three distinct levels: mechanisms, modes, and metrics. ML can provide a unique multi-level perspective on characterizing LIB degradation and it can improve accuracy especially when combined with perturbation techniques. This pulse injection aided machine learning (PIAML) technique can be used for battery diagnostic and prognostics with high fidelity. LIB management systems can leverage this estimation framework to extend lifetime and reduce costs. Battery degradation: impedance and incremental capacitance. Estimation of mechanisms, modes, and metrics including SoC, SoH, SoP.