About

Anna Stefanopoulou is the William Clay Ford Professor of Manufacturing and a distinguished faculty member in the Department of Mechanical Engineering at the University of Michigan. Her research interests encompass the estimation and control of internal combustion engines and electrochemical processes such as fuel cells and batteries. She has made significant contributions to the modeling, analysis, and control of advanced vehicle powertrain systems, with a focus on energy, controls, and mobility, particularly in automotive and transportation applications. Stefanopoulou has been recognized with numerous awards and honors, including being elected as a Fellow of SAE, IEEE, and ASME. She has served as an associate editor for the IEEE Transactions in Control Systems Technology and has held leadership roles such as serving on the IEEE Control Systems Society Board of Governors and the ASME Dynamic Systems Control Division Executive Committee. Her work has been acknowledged through awards like the Control System Technology Award from IEEE Control System Society, the U-M Faculty Recognition Award, and the Gustus L. Larson Memorial Award from ASME. Her contributions extend to battery management systems, energy systems, and automotive control technologies, and she has been actively involved in advancing battery safety and electrification of the automotive sector. Stefanopoulou has delivered keynote speeches at major conferences, served as director of the U-M Energy Institute, and has been honored with a Distinguished University Professorship for her scholarly and creative achievements, reputation for academic excellence, and her mentorship and service.

Research topics

• Computer Science
• Automotive engineering
• Engineering
• Mechanical engineering

Selected publications

• Robust emergency fast discharge algorithm for lithium-ion batteries

  Control Engineering Practice · 2026-03-03

  article
  PublisherDOI

• Quantifying imbalances in parallel-connected cell groups using group voltage and current

  University of Michigan Library · 2026-01-01

  otherOpen accessSenior author

  This simulated data was produced to investigate the impact of cell-to-cell imbalances on the voltage and current dynamics of a group of parallel-connected cells. From this simulated data, we show that features of the group’s differential voltage with respect to differential state of charge (dV/dz) can quantify imbalance in the capacity-resistance product (CR). We demonstrate that CR imbalance is proportional to imbalances in C-rate and SOC, and that dV/dz features can consequently quantify these imbalances. We develop a novel algorithm, OCP-informed Feature Identification, that accurately and precisely estimates these features from noisy voltage data to enable a robust assessment of imbalances in parallel-connected cell groups.

  PublisherDOI

• Quantifying imbalances in parallel-connected cell groups using group voltage and current

  Journal of Energy Storage · 2026-01-14

  articleOpen accessSenior author
  PublisherDOI

• Control-Oriented Statistical Model of In-Cylinder Pressure, Combustion Phasing, and Torque for Compression-Ignition Engines Operating with Low Cetane Fuels

  SAE technical papers on CD-ROM/SAE technical paper series · 2025-03-31

  article

  <div class="section abstract"><div class="htmlview paragraph">Sustainable aviation fuels are becoming more widely available for current and future engine powered propulsion systems. However, the diversity of ignition behavior in these fuels poses a challenge to achieving robust, efficient operation. Specifically, low cetane fuels with poor ignitability exhibit highly variable torque production unless fuel is injected earlier during compression. The tradeoff is that earlier injection may cause dangerously high in-cylinder pressure rise rates. Novel models that can simulate these competing behaviors are needed so that appropriate strategies may be developed for controlling combustion at low cetane fueling conditions. This work builds upon a previously developed model that simulates asymmetric combustion phasing (CA50) distributions as a function of fuel cetane, fuel injection timing, and electrical power supplied to an in-cylinder thermal ignition assist device. An extension of the model is presented in which the phasing output is used to reconstruct in-cylinder pressure traces, by which indicated mean effective pressure (IMEP) can then be simulated. Additionally, a functional form is parametrized for modeling maximum in-cylinder pressure rise rate (MPRR). The model’s parameters are regressed using a total 121,237 engine cycles of experimental data from a commercial CI engine operating with four fuel blends with cetane number ranging from 25 to 48. Relative to the data, the model simulates mean CA50 within a root mean square error (RMSE) of approximately 3 CAD over a range of 64 CAD, mean IMEP within an RMSE of 0.85 bar over a range of 5.8 bar, and mean MPRR within an RMSE of 6.1 bar/CAD over a range of 88 bar/CAD. Along with the mean-value trends, it accurately emulates the variability of all three combustion metrics. Ultimately, this work marks the first time a low-order, control-oriented model simulates statistical distributions of CA50, IMEP, and MPRR.</div></div>

  PublisherDOI

• Fast-charging lithium-ion batteries require a systems engineering approach

  Nature Energy · 2025-07-10 · 16 citations

  article
  PublisherDOI

• Mobile-Guided Gas Sensing for Vent Detection in Battery Energy Storage Systems

  ECS Meeting Abstracts · 2025-11-24

  article

  Recent Battery Energy Storage System (BESS) failures highlight the need to detect gases from venting cells as quickly as possible, as well as the vulnerabilities in existing monitoring infrastructure that would alert the occurrence and location of cell venting [1]. Conventional detection strategies, which rely on stationary gas sensors, can fail to identify low-volume gas release from single-cell early stage (first) venting, especially in large-scale BESS that have multiple racks and modules, such as Panel A in Figure 1. Suboptimal or improper fixed sensor placement, gas transport effects (e.g., thermal buoyancy), and dilution below detection thresholds further complicate failure detection. These limitations hinder emergency responders’ ability to accurately assess explosion and/or toxicity risks. To address these challenges, we propose integrating a guided mobile gas sensing platform to complement stationary gas sensors. This system will dynamically monitor spatial and temporal gas evolution during early failure stages by deploying a robotic platform equipped with commercial sensors to detect the most commonly emitted gases (CO2, H2, CO, electrolyte vapor). The robot will patrol accessible areas or be strategically guided to areas of interest to actively sample “suspicious zones” with high concentrations of vent gases. This target navigation will be informed by advanced state-of-health (SOH) diagnostics derived from electrical and thermal measurements typically available in the Battery Management System (BMS) data. A major component of our detection algorithm is to recognize how rapidly a fault is propagating to other cells or the time elapsed from the vent initiation. To this end, we first characterize venting gases in terms of composition, quantity, temperature, and time evolution. A literature review (Figure 1, Panel B) on gas characterization was conducted, and it was found that most prior work focused on global, post-thermal runaway (TR) gas analysis. However, the combustion of vent gases during TR makes their characterization challenging, leaving first venting behavior poorly understood. To address this gap, we aim to develop a systematic way to trigger gas generation without full thermal runaway by inducing controlled overheating and arresting the heat after first venting occurs. We will apply this method to two form factors: cylindrical (2.6 Ah) and prismatic (32 Ah), focusing on LFP cells due to their widespread adoption in grid-scale BESS. After characterizing single-cell behavior, we will construct a representative BESS rack (Figure 1, Panel C) with cells inside modules to study venting propagation from vented cells to the module venting channels and to the rack headspace where the stationary sensors are typically located. Our gas characterization will then inform models and algorithms that can recognize the various stages of an evolving failure, including the location and the number of cells undergoing slow discharge, venting, and/or the propagation of thermal runaway. This will guide the robotic sensor platform to reach the appropriate venting channel, providing a faster response and better guidance for first responders. By integrating robotic mobility with diagnostic and prognostic tools, this will not only improve the detection of first venting events, but also enhance situational awareness for emergency response. It enables high-confidence localization of first venting events, estimates the scale of the fault progression, and quantifies the number of cells involved in or propagating in a thermal runaway event, providing actionable data to guide safer and more rapid containment efforts. ACKNOWLEDGEMENT This project is supported by UL Research Institutes. REFERENCES [1] Srinivasan, L., Shaw, S. and Billaut, E. (2024). Insights from EPRI’s battery energy storage systems (BESS) failure incident database: analysis of failure root cause Figure 1: A typical large-scale Battery Energy Storage System (BESS) with multiple racks and modules (Panel A), an overview of literature showing the lack of vent (pre-thermal runaway) gas characterization for LFP cells (Panel B), and a representative BESS rack with a mockup of the mobile guided gas sensing platform directed to sample from a module vent (Panel C). Figure 1

  PublisherDOI

• Characterizing Inhomogeneity in Degraded Lithium-Ion Batteries across Operating Temperatures

  ECS Meeting Abstracts · 2025-11-24

  articleSenior author

  Inhomogeneity in lithium-ion batteries generally describes nonuniform distributions of materials, properties, and operating conditions that lead to local variation in aging factors (e.g., state-of-charge, current density, and temperature). In extreme cases, inhomogeneities can develop into safety risks, including localized overheating, overcharging, and internal short circuits, making it important for battery management systems (BMS) to characterize inhomogeneity and monitor its evolution. While numerous factors influence inhomogeneity, previous studies have shown that the resulting smoothing of differential voltage (DV) phase transition peaks can be captured by modeling inhomogeneity as a distribution of the state of lithiation or capacity [1,2]. However, there is limited data and discussion about the impact of operating temperature on the evolution of inhomogeneity, especially for cells with significant capacity loss. Our work quantifies inhomogeneity for NMC/graphite cells aged under different temperature and C-rate conditions up to 50% capacity loss and shows how inhomogeneity develops at different temperatures [3]. By applying a combination of algorithms from previous studies [1,2,4], we characterize inhomogeneity by augmenting a differential voltage analysis (DVA) algorithm, which typically assumes invariant half-cell open-circuit potential (OCP) taken from the fresh cell, with a modeled aged negative electrode OCP. The aged electrode OCP model captures the smoothing of graphite phase transition peaks using a Gaussian distribution of electrode capacities, where inhomogeneity (σ Cn ) is characterized as the standard deviation. Overall, the model describes the aged full cell open-circuit voltage (OCV) and DV well, capturing the DV peak smoothing behavior even after the peak corresponding to the Stage 2 transition, or high-SOC peak, is not observable (Fig. 1a). Across all cells, the average root-mean-square errors (RMSE) are low for voltage (RMSE V = 4.2 mV) and differential voltage (RMSE dVdQ = 17.5 mV/Ah). A comparison with the conventional voltage fitting (VF) algorithm shows a significant reduction in the RMSE dVdQ , while maintaining comparably low RMSEV as the cells age (Fig. 1b). Across cells cycled at C/3 and cold (0 o C), room (~25 o C), and hot (45 o C) temperature, we observed different increasing trajectories in inhomogeneity (Fig. 1c). After an initial decrease, σ Cn increases mostly linearly with growth rates in order of hot > cold> room temperature, until the cells transition to an accelerated degradation rate, or “knee” in capacity. After the knee, σ Cn grows notably faster in cold and hot temperature cells. In the cold temperature cells, the sudden knee transition corresponds to a significant increase in σ Cn . In contrast, the knee transition and the corresponding increase in σ Cn in the hot temperature cells are gradual and occur when the N:P ratio<1, indicating electrode saturation. While both cold- and hot-temperature cells have similar σ Cn (0.8 Ah) after 30% capacity loss, cold-temperature cells consumed almost half the amount of Li after the knee compared to hot-temperature cells. While the literature offers suggestions for possible underlying degradation mechanisms, future work will include post-mortem analyses to identify the dominant degradation mechanisms for a subset of cells. REFERENCES [1] Kirst, Cedric, et al. "Non-destructive electrode potential and open-circuit voltage aging estimation for lithium-ion batteries." Journal of Power Sources 602 (2024): 234341. [2] Fath, Johannes Philipp, et al. "Quantification of aging mechanisms and inhomogeneity in cycled lithium-ion cells by differential voltage analysis." Journal of Energy Storage 25 (2019): 100813. [3] Tran, Vivian, et al. “Estimating degradation modes and inhomogeneity in aged Lithium-ion batteries.” (In preparation) [4] Lee, Suhak, et al. "Electrode state of health estimation for lithium ion batteries considering half-cell potential change due to aging." Journal of The Electrochemical Society 167.9 (2020): 090531. Figure 1: Overview of the main modelling and parameterization results, including (a) an example of measured and modeled voltage and DV for a cell aged at C/3 and 45 o C from fresh to 50% SOH; (b) an error comparison to illustrate the impact of adding differential voltage error (V+DV) to the cost function and inhomogeneity (V+DV+σ) to the parameters; and (c) the parameterized degradation modes across temperature for cells cycled at C/3 and 2C. Figure 1

  PublisherDOI

• Lessons and Challenges from Crowd-Sourced Data in Rural Michigan to Assess EV Suitability based on Analyzing Personal Driving

  SAE technical papers on CD-ROM/SAE technical paper series · 2025-03-31 · 1 citations

  article

  <div class="section abstract"><div class="htmlview paragraph">Abstract</div><div class="htmlview paragraph">Real-world driving data is an invaluable asset for several types of transportation research, including emissions estimation, vehicle control development, and public infrastructure planning. Traditional methods of real-world driving data collection use expensive GPS-based data logging equipment which provide advanced capabilities but may increase complexity, cost, and setup time. This paper focuses on using the Google Maps application available for smartphones due to the potential to scale-up real-world driving data logging. Samples of the potential data processing and information that can be gathered by such a logging methodology is presented. Specifically, two months of Google Maps driving data logged by a rural Michigan resident on their smartphone may provide insights on their driving range, duration, and geographic area of coverage (AOC) to guide them on future vehicle purchase decisions. Aggregating such statistics from crowd-sourcing real-world driving data via Google Maps may also inform us of general characteristics in rural driving, along with placing public chargers, particularly in under-studied geographic regions like rural Michigan.</div></div>

  PublisherDOI

• Predicting Battery Remaining Useful Life for EV Resale: Switching from/to Cold/Hot Temperature

  2025-07-08 · 1 citations

  articleSenior author

  Used electric vehicles are driven and sold across states and countries where there can be a switch in both the driving pattern and environmental temperature in which the vehicle is parked and driven. Predicting battery remaining useful life (RUL) and associated fair resale value under such a switch is challenging as one cannot rely on extrapolating the state-of-health (SOH) trend observed from its first user.This paper presents Gaussian Process (GP) regressions that can predict capacity fade in NMC-graphite cells undergoing a switch in operating temperature (from -5°C to 45°C and vice versa) as they transition from first to second use. The GPs are trained on data collected from three cells cycled until 70% SOH in the laboratory at various temperatures (room: 25°C, cold: -5°C, and hot: 45°C). In addition, the GP is also trained on first-use data (before SOH reaches 80%), after which the operating temperature is switched. Training data consisted of temperature and total Amp-hours throughput collected during slow and full charge/discharge cycles which are performed approximately every 40 cycles, assuming such conditions will rarely (once every year) occur in the field. We compare two versions of GP: a baseline regression with a linear mean function and a domain-knowledge informed regression with nonlinear mean function. The nonlinear mean leverages and retrains a basis function based on an empirical degradation model previously developed by the authors. Our GP predicts RUL in second use after a large temperature swing, with an RMSE of 2.5% for another 3 years of operation (about 100 cycles) in the future.

  PublisherDOI

• Quantifying Imbalances in Parallel-Connected Cell Groups Using Group Voltage and Current

  SSRN Electronic Journal · 2025-01-01

  preprintOpen accessSenior author
  PublisherDOI

Recent grants

• GOALI- Fuel Cell Performance and Tolerance in Dead-Ended Anode Operation

  NSF · $300k · 2009–2013

• GOALI: Estimation and Control of Water in PEM Fuel Cells

  NSF · $252k · 2006–2010

Frequent coauthors

• Lorenzo Marconi

  University of Bologna

  470 shared

• Luca Zaccarian

  470 shared

• Hitay Özbay

  Bilkent University

  470 shared

• B. Pasik-Duncan

  470 shared

• Warren E. Dixon

  University of Florida

  470 shared

• Thomas Parisini

  469 shared

• M.L. Corradini

  Università di Camerino

  462 shared

• V Balakrishnan

  Institute for Plasma Research

  461 shared

Labs

• Battery Control GroupPI

Education

• PhD, Electrical Engineering and Computer Science

  University of Michigan

• MS, Electrical Engineering Computer Science

  University of Michigan

• MS, Naval Architecture and Marine Engineering

  University of Michigan

Awards & honors

• Fellow, SAE, 2018
• Control System Technology Award, IEEE Control System Society…
• Member IEEE Fellow: Evaluation Committee for IEEE Control Sy…
• Gustus L. Larson Memorial Award, ASME, 2009
• Fellow, Institute of Electrical and Electronics Engineers, 2…