Degradation decomposition
Total fade, split into the mechanisms that caused it.
“20% capacity lost” is never one thing. Decomposition attributes the fade to its mechanisms, lithium inventory loss, active material loss, and kinetic (impedance) growth, because each one points at a different fix.
What it measures
Three mechanism families account for most lithium-ion degradation, and they leave different fingerprints in the differential curves computed from ordinary cycling data:
- Lithium inventory loss (LLI): cyclable lithium consumed by SEI growth or plating. In dQ/dV, the peaks keep their shape but the curve shifts and the capacity between peak pairs shrinks.
- Active material loss (LAM): electrode sites disconnecting through particle cracking or binder failure. dQ/dV peak heights collapse while positions hold.
- Kinetic growth: rising impedance from film growth or contact loss. Peaks migrate apart with cycling (growing charge/discharge separation) and the same fade partially recovers at lower C-rate.
The raw ingredients are dQ/dV (incremental capacity) and dV/dQ (differential voltage) curves tracked across aging, computed with a resample → smooth → differentiate pipeline, differentiating raw cycler data directly amplifies noise into false peaks.
How to read the output
Overlay the differential curves from early, mid, and late life and watch what the peaks do. Shifting curve, stable peak heights: LLI, look upstream at formation, electrolyte, and anything that grows SEI. Collapsing peaks at fixed positions: LAM. Look at electrode mechanics, calendering density, particle cracking. Widening charge/discharge separation: kinetic, look at impedance growth and contact quality.
Real cells mix mechanisms, so the readout is a dominant mode plus secondary contributions, with the caveat that a full quantitative split needs half-cell reference curves. Relative trends across a population are usually enough to direct the fix.
A real use case
A pack integrator reports that cells from one production month fade noticeably faster in the field, while formation metrics and early capacity for that month look identical to the good months. Running dQ/dV evolution on returned cells shows peak heights holding steady while the curves shift, lithium inventory loss, not active material loss. That single distinction moves the investigation away from the electrode lines and onto formation and electrolyte, where a dry-room humidity excursion during that month turns out to have accelerated SEI growth. The decomposition did not just measure the fade; it picked which half of the factory to investigate.
Common mistakes
- Differentiating raw voltage-capacity data. Noise explodes under differentiation; resample and smooth first or every wiggle becomes a phantom peak.
- Claiming a quantitative LLI/LAM split from full-cell data alone. Without half-cell reference curves, only relative trends are defensible.
- Reading peak assignments from the wrong chemistry. The peak map for NMC811/graphite is not the LFP map, confirm the chemistry before interpreting.
- Comparing curves taken at different C-rates or temperatures; kinetic peak shifts will masquerade as aging.
- Stopping at state-of-health. Two cells at 85% SOH with different dominant mechanisms need different interventions and have different remaining lives.
Tiered diagnostics that scale with the data you have
Niobia runs degradation diagnostics in tiers that match the data available. With basic cycling data it establishes signal sanity and per-cycle trends; with multi-cycle capacity data it adds coulombic-efficiency trends, voltage-profile evolution, and dQ/dV-based mode indicators, computed through the resample → smooth → differentiate pipeline, separating lithium inventory loss from active material loss from kinetic growth; with resistance or impedance data it corroborates the kinetic component directly. It estimates state-of-health and remaining useful life with explicit confidence framing, maps peak signatures against the confirmed chemistry (NMC811, NMC622, NCA, LFP, LCO, LTO, graphite, Li-S), and will not fabricate a quantitative mode attribution when the data cannot support one, without half-cell references it reports relative trends and says so.
Frequently asked
Can degradation modes really be separated without opening the cell?
To a useful degree, yes. LLI, LAM, and kinetic growth leave distinct signatures in dQ/dV and dV/dQ evolution, which come from ordinary cycling data. The honest limit: a fully quantitative split needs half-cell reference data; full-cell analysis alone supports dominant-mode identification and relative trends.
What data does the analysis need?
Multi-cycle charge/discharge data with reasonable voltage resolution, ideally spanning enough of life to see evolution. Resistance or EIS data taken along the way strengthens the kinetic attribution considerably.
Why does the dominant mechanism matter if the fade total is the same?
Because the fixes differ. LLI points at formation and electrolyte; LAM points at electrode mechanics and calendering; kinetic growth points at impedance and contact quality. Treating a LAM problem with an electrolyte change wastes a development cycle.