Battery Ladder Technical Notes#

This page describes the packaged battery benchmark suite as a design-research benchmark ladder, not as a production battery-pack simulator. The key public entries are:

  • battery_18650_t1_rectangular_surrogate_*

  • battery_18650_t2_pose_surrogate_*

  • battery_18650_t3a_topology_surrogate_*

  • battery_18650_t3b_netlist_explicit_*

  • battery_18650_t4_thermal_hybrid_*

  • battery_fast_charge_dfn_anchor_opt

Representation And Evaluation Modes#

The battery suite now treats representation and evaluation mode as separate concepts.

Representation modes describe what a user designs:

  • rectangular: classical S x P pack sizing.

  • pose_layout: per-cell 3D pose with fixed rectangular electrical semantics.

  • topology_allocation: pose variables plus active-cell count and stage-slot assignment.

  • explicit_netlist: explicit cells, terminals, and interconnects.

  • thermal_topology: topology-allocation plus thermal-system variables.

  • fast_charge_cell: continuous electrochemical cell-design parameters.

Evaluation modes describe how the design is scored:

  • analytic_surrogate: closed-form pack electrical equations and compact thermal proxies.

  • explicit_circuit: projected or native explicit-netlist scoring through the shared circuit backend.

  • hybrid_thermal: explicit-circuit electrical scoring plus the Tier-4 thermal network.

  • electrochemical_anchor: direct PyBaMM DFN evaluation for the fast-charge anchor.

Pack Compatibility Matrix#

For the pack ladder, the public matrix is intentionally sparse but ordered. Each cell is written as electrical path / thermal path. The legend is:

  • native: the evaluator works directly on the representation.

  • projected: the representation is deterministically projected before that evaluator component runs.

  • promoted: missing detail is filled in by documented deterministic defaults.

  • unsupported: the pairing is intentionally rejected.

Representation

analytic_surrogate

explicit_circuit

hybrid_thermal

rectangular

native / native

promoted / native

promoted / promoted

pose_layout

native / native

promoted / native

promoted / promoted

topology_allocation

native / native

projected / native

projected / promoted

explicit_netlist

unsupported

native / native

native / promoted

thermal_topology

native / native

projected / native

projected / native

The only intentional public pack-matrix gap is explicit_netlist + analytic_surrogate. The package rejects that pairing instead of silently reducing a general netlist to a surrogate topology model.

Two row-specific notes matter:

  • thermal_topology + analytic_surrogate is supported, but analytic scoring ignores candidate-specific thermal-control variables by contract.

  • thermal_topology + hybrid_thermal stays projected / native because the candidate already contains native thermal-network variables; only the electrical side is projected to the shared explicit-circuit backend.

Shared Physical Scope#

The 18650 pack ladder uses one fixed cylindrical packaged cell with nominal:

  • V_cell = 3.7 V

  • C_cell = 2.5 Ah

  • R_int = 0.05 ohm

  • C_rate,max = 10 C

  • diameter = 18 mm

  • length = 65 mm

All pack benchmarks enforce hard requirements on:

  • target voltage within tolerance,

  • minimum capacity,

  • minimum current capability,

  • maximum width/depth/height,

  • minimum inter-cell clearance.

Geometry from Tier 2 upward uses finite oriented cylinders. Clearance is based on minimum surface-to-surface distance, not center distance. The reported design volume is the axis-aligned bounding-box volume.

Tier Contracts#

Tier 1#

Question:

How well do methods handle discrete rectangular pack sizing when geometry and wiring are fixed?

Physically modeled:

Canonical rectangular S x P pack relations, pack envelope, cell count, and a steady-state thermal proxy.

Deliberate surrogates:

Topology is fixed to a full rectangular family, electrical behavior is summarized analytically, and thermal behavior is represented by a compact Joule-heating proxy.

Tier 2#

Question:

How well do methods handle continuous geometric freedom once rectangular pack sizing is no longer enough?

Physically modeled:

Per-cell 3D pose, finite-cylinder clearance, and bounding-box volume.

Deliberate surrogates:

Electrical and thermal scoring remain analytic pack-level surrogates.

Tier 3A#

Question:

How well do methods handle asymmetric topology allocation once cell count and stage assignment matter?

Physically modeled:

Active-cell count, pose, stage-slot assignment, geometric feasibility, and an optional projected explicit-circuit check.

Deliberate surrogates:

The default electrical abstraction uses an imbalance surrogate instead of a general circuit solve. The default min_stage model is intentionally conservative and penalizes uneven stage populations.

Tier 3B#

Question:

How well do methods synthesize explicit pack netlists when topology is itself the representation?

Physically modeled:

Explicit cells, terminals, interconnects, graph validation, and constant-load explicit-circuit discharge scoring.

Deliberate surrogates:

Thermal behavior still uses a compact pack-level proxy rather than a full electro-thermal pack transient model.

Tier 4#

Question:

How well do methods co-design topology, geometry, and thermal controls when temperature becomes a first-class design axis?

Physically modeled:

Tier-3A representation plus cooling coefficient, passive cooling, ambient temperature, and a Tier-4 thermal network using PyBaMM-derived priors.

Deliberate surrogates:

Candidate representation is still topology-allocation based; only the evaluator fidelity changes across modes.

Fast-Charge Anchor#

Question:

How well do optimization methods handle a higher-fidelity electrochemical battery-design problem?

Physically modeled:

A PyBaMM DFN with lumped thermal dynamics, plating, SEI growth, and CC-CV fast-charge evaluation.

Solver role:

The packaged solve() method is a deterministic baseline/reference search, not a claim of strong optimization performance.

Shared Surrogates And Substitution Rules#

Analytic pack electrical surrogates use:

\[V_{pack} \approx S \cdot V_{cell}\]
\[C_{pack} \approx P_{eq} \cdot C_{cell}\]
\[I_{limit} \approx P_{eq} \cdot C_{cell} \cdot C_{rate,max}\]

The effective parallel support P_eq depends on the benchmark:

  • Tier 1 and Tier 2: P_eq = P.

  • Tier 3A and Tier 4 surrogate modes: P_eq is derived from the stage populations.

Two imbalance surrogates are currently supported for topology-allocation style benchmarks:

  • min_stage: use the least-populated stage.

  • harmonic_mean_stage: use the harmonic mean of non-empty stage counts.

Adaptation rules are intentionally one-way in this release:

  • topology-allocation candidates may be projected to a canonical explicit netlist for explicit_circuit scoring;

  • lower-detail rectangular and pose-layout candidates may be promoted to a canonical series-parallel circuit for explicit scoring;

  • hybrid thermal scoring may promote lower-detail pack representations to one deterministic thermal context rather than inventing representation-specific heuristics at runtime;

  • arbitrary explicit netlists do not automatically reduce back to surrogate topology metrics unless a deterministic reduction is defined.

Backend Provenance#

Battery problems accept shared backend configuration through [parameters.battery_backend] and now report evaluation provenance explicitly. Provenance records:

  • representation mode,

  • evaluation mode,

  • imbalance model when applicable,

  • requested backend config,

  • resolved backend config,

  • honored vs ignored backend fields,

  • electrical path (native, projected, or promoted),

  • thermal path (native, promoted, or none),

  • cell-model source,

  • thermal-prior source,

  • promoted-only assumed defaults when applicable,

  • adaptation notes that explain any projections or promotions.

This is meant to make battery benchmark fidelity legible without pretending that every public problem uses the same evaluator.

Electrical Backend Modes#

For explicit-circuit and hybrid-thermal evaluation, the shared backend now distinguishes between a compact default ECM path and a medium-fidelity pulse path:

  • auto -> pybamm_ecm keeps the existing one-RC Thevenin-style surrogate as the default for backwards compatibility and fast screening.

  • pybamm_ecm_2rc adds a second relaxation branch and is fit from live single-cell PyBaMM SPM traces over a compact SOC/temperature pulse-rest design.

  • pybamm_direct runs a direct PyBaMM SPM discharge for ideal series-parallel packs when a higher-cost reference evaluation is worth the runtime.

The intended claim for pybamm_ecm_2rc is deliberately narrow:

scientifically grounded for conceptual pack ranking and first-order transient voltage prediction under benchmarked discharge and pulse/rest conditions, with explicit limits outside those regimes.

The backend still does not claim to cover:

  • strong-hysteresis chemistries unless a later hysteresis mode is added;

  • aggressive charge or regenerative operating histories;

  • arbitrary asymmetric explicit-netlist current-split problems through the pybamm_direct path;

  • production BMS use, safety-critical prediction, or plating-risk claims.

Backend-Config Example Variants#

Three packaged catalog variants now pin one non-default backend configuration directly in their manifests:

  • battery_18650_t3a_topology_explicit_2rc_opt: projected explicit-circuit scoring on the topology-allocation rung.

  • battery_18650_t3b_netlist_explicit_2rc_grammar: native explicit-netlist grammar evaluation with the same backend choice.

  • battery_18650_t4_thermal_hybrid_2rc_opt: hybrid thermal scoring with the same electrical backend selection.

All three variants use the same manifest payload:

[parameters.battery_backend]
cell_model_mode = "pybamm_ecm_2rc"
thermal_mode = "isothermal"

[parameters.battery_backend.parameterization]
parameter_set = "Marquis2019"

The intent is to surface a concrete medium-fidelity backend choice in the public catalog without changing the underlying representation contracts of the base tiers.

Validation Matrix#

The suite is validated as a benchmark family rather than via a full battery-validation campaign:

Benchmark

Validation scope

Tier 1

analytically checked surrogate consistency

Tier 2

geometry validity and monotonicity checks

Tier 3A surrogate

topology-abstraction sanity checks

Tier 3B explicit

explicit-circuit consistency checks

Tier 4

qualitative thermal trend and mode-consistency tests

Fast-charge DFN anchor

PyBaMM model and solver reproducibility checks

For the shared explicit battery backend, validation is split into two layers:

  • always-on invariant checks for KCL/KVL consistency, SOC/state boundedness, discharge monotonicity, long-rest OCV convergence, and energy/loss sanity;

  • marked pybamm_real oracle checks against live PyBaMM single-cell pulse/rest and short dynamic traces, plus small symmetric/asymmetric pack sentinels using the fitted backend models.

Background References#