Degradation modeling
One model that predicts capacity loss, resistance rise, and irreversible expansion together: tuned once, and still right on cycling conditions it never saw.
Research Fellow · Battery Control Lab, University of Michigan
Sravan Pannala, Ph.D.
Li-ion battery degradation modeling · expansion sensing · state estimation
I build physics-based models that predict how lithium-ion cells lose capacity, gain resistance, and expand irreversibly over their life, and the low-cost sensors that measure it. Nine years at Michigan's Battery Control Lab; the model code from my papers is public.
Open to battery R&D and modeling roles in industry and at national labs.
+201 µm irreversible after 339 cycles
107 µm swelling per charge, reversible
397citations
10h-index
14peer-reviewed
1patent +1 pending
Google Scholar, Oct 2026
What I work on
Expansion sensing
An inductive sensor costing under $20 that tracks cell thickness to within about 3 µm of a $2,000 lab-standard LVDT, fine enough to follow irreversible expansion from the first formation charge.
Accelerated aging simulation
Simulating a battery's whole life in seconds instead of minutes or hours, by extrapolating between cycles instead of solving every one.
State estimation and control
Reading internal state (lithium plating, SEI growth, electrode health) from voltage and thickness a pack can actually measure.
The same cell, 339 cycles apart
Every charge, a cell expands and then contracts again. Underneath that reversible expansion it grows thicker cycle after cycle and never shrinks back: irreversible expansion. Here that permanent growth reaches 201 µm, 67% of it from SEI growth, more than the cell's whole reversible expansion on a single charge. Predicting both from the same degradation mechanisms that cost the cell its capacity is what my thesis is about.
- SEI growth
- Active-material loss
- Lithium plating (2.8 µm, too thin to see)
Show the data as a table
| Cycle | SEI | Active material | Plating | Total (µm) |
|---|---|---|---|---|
| 1 | 2.7 | 0.0 | 0.0 | 2.7 |
| 41 | 22.6 | 7.4 | 0.0 | 30.1 |
| 81 | 41.1 | 14.9 | 0.2 | 56.2 |
| 120 | 58.0 | 22.2 | 0.3 | 80.5 |
| 160 | 74.1 | 29.7 | 0.6 | 104.4 |
| 200 | 89.1 | 37.3 | 1.0 | 127.4 |
| 240 | 103.1 | 44.9 | 1.4 | 149.4 |
| 279 | 115.8 | 52.5 | 1.9 | 170.2 |
| 319 | 127.9 | 60.3 | 2.5 | 190.7 |
| 339 | 133.6 | 64.2 | 2.8 | 200.6 |
five ages: brightest new, faintest cycle 339
five ages: brightest new, faintest cycle 339
+201 µmirreversible
−23%reversible
−23%capacity
Model output, not raw measurement. Pannala et al., J. Electrochem. Soc. 171 010532 (2024). Model code
Commercialization
Low-cost battery expansion measurements for formation and safety
Adapting the lab's low-cost inductive expansion sensor for use in battery manufacturing. Formation is where the SEI first grows, and it shows up as the cell's first irreversible expansion; measured on every cell, it can reveal defects and let formation protocols be shortened. Selected through a competitive process ending in a pitch to the MTRAC oversight committee.
- Sensor cost
- under $20 vs about $2,000 for an LVDT
- Accuracy
- within 3.2 µm of the LVDT
- Deployment
- 120 fixtures
- Maturity
- TRL 6
Selected work
Managing silicon burn-out via onboard material diagnostics for durable high-energy density batteries
Degradation and expansion of lithium-ion batteries with silicon/graphite anodes: Impact of pretension, temperature, C-rate and state-of-charge window
How mechanical pretension, temperature, and how hard you cycle a cell change the way silicon-graphite anodes age and expand.
Consistently tuned battery lifetime predictive model of capacity loss, resistance increase, and irreversible thickness growth
One model, tuned once, that predicts capacity loss, resistance rise, and irreversible expansion together, and holds up on cycling conditions it never saw during training.
Software, in and out of the lab
Battery modeling code released with the papers, plus the website and wiki a Linux distribution runs on and ten public analytics apps on ARM servers I manage myself.
batmod-gpu
Faster degradation-parameter fitting for PyBaMM
Gradient-based fitting uses roughly 15× fewer model evaluations than gradient-free, and batched solves run ~8.7× faster than serial on a 32-candidate DFN batch. The benchmarks also publish the negative result: for these ~100-state systems, SUNDIALS on 32 CPU threads still beats the GPU, so the CPU path is the recommendation.
PythonPyBaMMPyBOPJAX
Battery degradation and expansion model
Open-source code for the 2024 JES lifetime paper
Splits irreversible expansion into SEI growth, lithium plating, and active-material loss, and predicts it on cycling conditions it was never fitted to. The irreversible-expansion chart on this site comes from it.
PythonPyBaMMNumPyJupyter
endeavouros.com and Discovery wiki
The live website and wiki of a Linux distribution
Shipped to production for an established open-source project, not a demo.
AstroStarlightTypeScript
shinyNBA
Ten deployed analytics applications
Shipped and operated, not just written.
PythonShinypandasParquet
Honors and awards
2025Best Poster, Electric Vehicle Center Symposium, University of Michigan
2022Best Paper Runner-Up, Modeling, Estimation and Controls Conference (MECC)
2021Best Paper Runner-Up, Modeling, Estimation and Controls Conference (MECC)