Edwin R. Gilliland Professor, Massachusetts Institute of Technology
Machine Learning-based Identification, Prediction, and Control of Lithium-ion Batteries
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), the determination of optimal charging protocols, and the identification of fundamental physicochemical expressions for electrochemical kinetics, thermodynamics, and mass transfer from real-time video imaging. The development of the techniques and their application are in collaboration with materials science, applied physics, and computer science researchers at Stanford University, Toyota Research Institute, and MIT.
Assistant Professor, Tsinghua University
An Electrochemical-Thermal Coupled Battery Thermal Runaway Model for Battery Safety Design
Thermal runaway is always a troublesome problem that hinders the safe application of high energy lithium-ion batteries. There is an urgent need to interpret the voltage and temperature changes and their underlying mechanisms during thermal runaway, in order to guide the safe design of a battery system. This research is dedicated to building a coupled electrochemical-thermal model that can well predict the voltage drop and temperature increase during thermal runaway. The model can capture the underlying mechanism of 1) the capacity degradation under high temperature; 2) the internal short circuit caused by the thermal failure of the separator; and 3) the chemical reactions of the cell components that release heat under extreme temperature. The model is validated using by experimental data, therefore the modeling analysis has high fidelity. We employ the model to analyze 1) the capacity degradation under extreme temperature; 2) the influence of the SEI decomposition and regeneration on the thermal runaway behavior; 3) the heat generation by internal short circuit in the thermal runaway process. The discussions presented here help extend the usage of lithium-ion batteries at extreme high temperature (>80◦C), and guide the safe design of lithium-ion batteries with less hazard level during thermal runaway.
Associate Professor, University of Oxford
Data-Driven Battery Health Diagnosis in Real-World Applications
Accurate diagnostics and prognostics of battery health improves overall system performance. This allows industry to unlock value by detecting faults and improving maintenance, extending operational range, and understanding asset depreciation. However, battery aging is complex and caused by many interacting factors. Two key questions arise: first, how to handle modelling challenges, including parameter variability and nonlinearities, in methods for online estimation of state of health. Second, how to develop validated predictions of future health, where key issues include coping with variable usage scenarios, and cell-to-cell behavioural differences. This talk will discuss recent approaches to tackle some of these exciting topics, including the use of Bayesian non-parametric models for health prediction, and the combining of non-parametric and parametric models to allow flexibility in model fitting from data, whilst retaining the benefits of equivalent circuit and physical models.
Assistant Professor, University of California, Davis
Sensitivity: Decoding the Data Signature of Battery States and Parameters
Estimation of battery internal states and parameters, e.g. state of charge (SOC), state of health (SOH), and state of power (SOP), is one of the most important tasks of battery management. Estimation is typically performed using sensor data including current, voltage, and temperature. It is noted that different states/parameters will influence the measured output (e.g. voltage) differently, and for each variable, there will be certain input (current) patterns under which it has significant impact on the output and hence are highly observable from the data. These patterns, or data signatures, can be extracted by studying the sensitivity of the variables to the data based on the physical models governing the underlying input-output relationship. In this talk, we will discuss how to analytically derive the data signatures based on sensitivity analysis for battery phenomenological and electrochemical states/parameters and the potential use of them to optimize the accuracy of estimation.
Associate Professor, University of California, Berkeley
Unleashing the Full Potential of Batteries: Learning and Control
Batteries are ubiquitous. However, today’s batteries are expensive, range-limited, power-restricted, die too quickly, charge too slowly, and susceptible to safety issues. For these reasons, advanced model-based battery management systems (BMS) are of extreme interest. In this talk, we discuss eCAL’s recent research electrochemical-based BMS, which are modeled by nonlinear partial differential equations (PDEs). Specifically, we discuss the safe-fast charging control problem. Two methods are reviewed: (i) reinforcement learning-based methods, and (ii) reference governors. Finally, we close with exciting new perspectives for next-generation battery systems.
Assistant Professor, Stanford University
Pushing the Envelope in Battery Estimation Algorithms
The current battery research in the control engineering community has focused mainly on lithium-ion batteries at the cell-level. This has led to the development of detailed battery models and estimation/control algorithms for single cells in an isolated environment. Merely applying the existing knowledge of a single cell to a large-scale battery pack composed of numerous interconnected cells assumes ‘modularity’, wherein modularity is defined as the ability to extrapolate the behavior of a pack from a single cell. Recent experimental studies presented in the literature show evidence that the assumption of modularity, in terms of thermal and aging behavior, does not hold true. Expectedly, a large-scale battery pack ages faster than a single cell, the thermal/aging behavior of the pack is non-uniform and pronounced in comparison to a single cell, and the performance of a pack is adversely affected due to cell-to-cell variability induced by manufacturing variances. As a result, pack-level modelling and estimation/control must take into consideration the temperature distribution, manufacturing-induced parameter (impedance, capacity) variances, pack topology (electrical interconnections), and thermal interactions between cells.
Research at the Stanford Energy Control Lab aims to extend the current understanding of electrochemical-thermal-aging dynamics of a lithium-ion cell to the pack level by taking into account the electrical interconnections, thermal interactions between cells and cell-to-cell heterogeneity. The implications of the outcome of this research will open up opportunities for accurate diagnosis/prognosis of battery pack health, efficient thermal control, and management of battery pack performance via multi-objective optimization-based control. Those are critical tasks to tackle to establish energy storage technology in renewable energy infrastructure, electric vehicles, and grid storage applications.
J. “Lee” Everett Professor of Mechanical Engineering, Pennsylvania State University
Thermal and Flow Actuation for High Performance Battery Systems
The opportunities for control in battery systems have been limited by a lack of actuators that can change the battery response in real-time. Recently, thermal and flow actuation have become interesting new tools in the battery systems engineers toolbox. Temperature has long been known to influence battery performance, including aging, power, and safety. Flow has been used to enhance metal electroplating surface finish, a process similar to charging in lithium metal batteries where dendrite growth is a safety hazard. This talk will address the potential and opportunities for thermal and flow actuation to improve the performance and lifespan of lithium ion chemistries.
Assistant Research Scientist/Lecturer, University of Michigan
Simultaneous Identification and Control for Hybrid Energy Storage System using Model Predictive Control and Active Signal Injection
Online State of Charge (SoC) and State of Health (SoH) estimation is essential for efficient, safe, and reliable operation of Lithium ion batteries. The previous research on this topic mainly focuses on the development of battery models and estimation algorithms. However, the nature of the battery excitations also significantly influences the estimation performance. The impact of data on the estimation accuracy is investigated using Fisher information matrix and Cramer-Rao bound, considering measurement noise. A sequential algorithm, which uses the frequency-scale separation and estimates the parameters/states sequentially by actively injecting current signals with different frequencies, has been proposed and developed, it will be presented at this workshop. By incorporating a high-pass filter, the high-frequency and medium-frequency currents are injected to facilitate estimation of SoC, battery parameters, and SoH in a sequential form. Experimental results show that the estimation accuracy of the proposed sequential algorithm is much better than the concurrent algorithm where all parameters/states are estimated simultaneously. As a case study, the sequential algorithm is applied to the battery/supercapacitor hybrid energy storage system (HESS), which is over-actuated in the sense that there are two power sources providing power to the load. This over-actuation feature is exploited to achieve accurate identification of the battery states/parameters and high system efficiency simultaneously. To resolve the conflict between providing rich signal content for identification and achieving energy efficient operation, a model predictive control (MPC) strategy is used to incorporate both objectives to determine the optimal power distribution between supercapacitor and battery. The results showing the benefits of the Simultaneous Identification and Control framework will be discussed in the context of managing the competing requirements for estimation accuracy and energy efficiency.
Professor of Mechanical Engineering & William Clay Ford Professor of Technology, University of Michigan
The Price of Degradation — The Value of Prognostics — The Cost-Effectiveness Case of Electric Buses
Replacing a gasoline or diesel internal combustion engine vehicle (ICEV) with an electric vehicle (EV) will zero out the direct (tailpipe) GHG emissions. The net emission impacts as well as the financial impacts of a shift to EVs are important. Key barriers to EV adoption include the higher first cost of EVs compared to ICEVs and the up-front investment needed to provide charging infrastructure. The total cost of ownership and the cost of the fleet transition depends on the lower operating cost due to fuel and maintenance savings that depend on the route and utilization of the vehicle. A long route with en-route fast-charging charging can payback fast as long as the warrantee timeline is satisfied and the actual battery degradation is monitored and constrained by mixing or altering routes. The multi-objective optimization supported by real-time estimation of degradation and prognostics will be discussed. The possibility of revenue from a second life application of the aged batteries in buildings (behind the meter) will be also reviewed. Finally, the value of using the buses for resiliency beyond their nominal transportation mission will be considered.
Ernest Dashiell Cockrell II Professor of Engineering, The University of Texas at Austin
Model-Based BMS and the Design of Efficient Algorithms for Current and Next-Generation Batteries
In order to significantly expand the PHEV/ BEV market, and to increase the use of lithium-ion batteries in electric grids, there is a need to develop optimal charging strategies to utilize the batteries more efficiently and enable design for longer life. Advanced battery management systems (BMS) that can calculate and implement such strategies in real-time are expected to play a critical role for this purpose. This talk will present approaches for determining model-based optimal charging profiles for batteries, and experimental validation of the same. Model-based BMS enables > 100% improvement in cycle life, 30% reduction in charging time, and >50% reduction in temperature rise. Validation of the same for 6 different batteries/sizes/chemistries/form-factors will be presented.
In addition, the design and relevance of efficient algorithms for enabling model-based BMS will be discussed. To simulate solid-phase diffusion inside the electrode particles efficiently (along the radial direction), the concept of volume-averaged flux (with quartic polynomials) was introduced by us in 2005 and has been successfully adopted by the battery modeling and control community worldwide. To simulate the concentration, temperature, and potential fields along the electrode thickness (x-direction, measured from the cathode current collector to the separator) model-reformulation approach based on spectral methods was developed. This enables efficient simulation of lithium-ion battery models (15 ms for a discharge curve prediction) without losing fidelity. To address the frequent failures of standard DAE solvers (integrators in time) a robust fail-safe approach for simulating index-1 nonlinear differential-algebraic equations was developed. (As of today, this is the most robust approach for solving battery models and fails 0% of the time). To enable real-time simulation and control of 100s of cells in series/parallel configuration, a novel tanks-in-series approach was developed. This method has the same computational complexity as simple flowsheet models. Application (and the inherent presence) of control theory in enabling the development of some of these efficient algorithms will be highlighted.
Finally, challenges in simulating multiscale-multidomain-multiphysics-multiphase models for next-generation batteries will be presented.