Electrochemical impedance is widely recognized as one of the most information-dense, non-destructive signals for diagnosing battery health, degradation mechanisms, and safety risk. Despite this, impedance-based diagnostics — particularly Electrochemical Impedance Spectroscopy (EIS) — remain underutilized outside laboratory and specialized manufacturing environments due to hardware requirements, test duration, and integration complexity.
This article presents an impedance-native, software-defined diagnostic framework developed at Energsoft that shifts the focus from impedance measurement to impedance interpretation. By unifying full EIS, partial EIS, DC internal resistance (DCIR), pulse response, dQ/dV behavior, and inferred impedance within a common analytical framework, the approach enables earlier, mechanistically interpretable diagnostic decisions using heterogeneous data already generated across the battery lifecycle.
Limitations of Telemetry-Centric Battery Analytics
Most deployed battery analytics systems rely on telemetry signals — voltage, current, temperature — and derived quantities such as state of charge (SOC) and state of health (SOH). While effective for control and monitoring, these signals provide only indirect visibility into internal electrochemical state.
Voltage integrates multiple internal processes, obscuring causal mechanisms. Distinct degradation modes can project onto similar voltage trajectories, and telemetry-based machine learning models often generalize poorly across chemistries and usage regimes because they infer condition indirectly. As batteries scale in cost, energy density, and safety criticality, these limitations increasingly constrain early detection and confident decision-making.
Impedance as a First-Order Diagnostic Observable
Electrochemical impedance directly reflects internal processes such as charge-transfer kinetics, ion diffusion, interfacial layer growth, and electronic contact integrity. These processes evolve continuously throughout battery life and often change measurably before observable capacity loss or voltage deviation.
For this reason, EIS is widely used in R&D to study degradation mechanisms. However, full-spectrum EIS faces operational constraints, including specialized instrumentation, long measurement times, difficulty integrating into high-throughput workflows, and limited availability at scale. As a result, impedance diagnostics are often viewed as valuable but impractical outside controlled environments.
Impedance Is Rarely Measured Fully — but Widely Sampled
Although full EIS is uncommon in production, impedance-related behavior is already sampled across most battery workflows.
Laboratory and R&D environments generate full or partial EIS, hybrid pulse power characterization (HPPC), pulse tests, and rest-step relaxation data. Manufacturing and formation processes routinely capture DCIR, limited-frequency AC resistance checks, and end-of-line resistance measurements. In the field, battery management systems infer resistance through pulse response, diagnostic routines, and load transients.
Each measurement probes impedance over a specific time or frequency regime. Individually incomplete, together they encode a substantial portion of the battery’s impedance behavior. The primary challenge is coherent interpretation across these heterogeneous signals.
Interpretation Rather Than Measurement
The framework developed at Energsoft reframes impedance diagnostics as an interpretation problem rather than a measurement problem. Instead of introducing new instrumentation, the system focuses on extracting diagnostically relevant impedance features from data that already exists.
This decouples diagnostics from hardware, enables immediate use of historical datasets, and allows consistent reasoning across laboratory, manufacturing, and field contexts.
At the center of the system is a software-defined “Virtual EIS” layer that represents impedance as a structured, uncertainty-aware object.
Unified Impedance Representation
All impedance-related observations are mapped into a common representation capturing measurement modality (EIS, pulse response, DCIR, inferred), characteristic frequency or time constant, magnitude and phase (or response shape), temperature and SOC, and associated confidence bounds. This allows heterogeneous inputs to be aligned in a shared impedance feature space rather than treated as unrelated metrics.
From Time-Domain Signals to Impedance Features
Consider a current pulse commonly available in formation or BMS data. The resulting voltage response can be decomposed into an instantaneous drop (ohmic and high-frequency behavior), a shortterm exponential decay (charge-transfer processes), and a longer relaxation tail (diffusion-dominated behavior).
Using equivalent circuit approximations and physics-informed constraints, these components can be mapped to impedance regimes typically observed in EIS spectra. While this does not reproduce full spectral resolution, it preserves diagnostically relevant distinctions with explicit uncertainty. Analogous mappings are applied to DCIR, HPPC, and limited-frequency AC measurements.
Relationship to dQ/dV Diagnostics
Incremental capacity (dQ/dV) analysis is widely used to infer degradation mechanisms such as loss of active material, lithium inventory loss, and phase-transition shifts. While often treated separately, dQ/dV and impedance diagnostics probe complementary manifestations of the same underlying electrochemical processes.
Within Energsoft’s framework, dQ/dV features — such as peak position, width, and symmetry — are treated as slow-times-cale impedance proxies. Many degradation mechanisms produce coupled signatures across both domains. Aligning impedance and dQ/dV features enables cross-validation, improved diagnostic confidence, and graceful degradation of inference when one signal is unavailable.
Why Generative AI Is Structurally Necessary
Operational impedance data is incomplete by construction. Measurements vary by environment, operating condition, and lifecycle stage, leaving large regions of impedance space unobserved.
Generative models are used to learn the joint distribution of impedance features across frequency, SOC, temperature, aging state, and cell cohort. This enables probabilistic completion of missing impedance regimes, uncertainty-aware inference, and cohort-level comparison. These models are constrained by physics-informed priors and equivalent circuit structure, ensuring physical plausibility.
Knowledge Graph as Diagnostic Infrastructure
To support contextual reasoning, Energsoft maintains a battery knowledge graph that encodes relationships between impedance behavior, dQ/dV features, electrochemical mechanisms, operating conditions, and observed outcomes.
The graph links chemistries, materials, and form factors to impedance regimes, dQ/dV archetypes, degradation mechanisms, and operational risks. Rather than serving as a static ontology, the graph constrains inference, resolves ambiguities, and improves interpretability. Its defensibility arises from multi-modal alignment, operational grounding in real datasets, and direct integration into inference and decision logic.
From Impedance Features to Decisions
The objective of impedance interpretation is operational decision support. One evaluated application is early-life quality assessment during formation. By tracking impedance evolution rather than voltage alone, cells can be categorized into nominal trajectories, elevated but potentially recoverable risk, and anomalous behavior indicative of latent defects.
Across historical datasets analyzed internally; impedance-derived indicators surfaced anomalous behavior on the order of tens of cycles earlier than voltage-based SOH metrics, with clearer separation between degradation modes. Similar logic is applied to fast-charge derating, safety monitoring, and predictive maintenance.
Hardware-Agnostic Architecture and Trade-Offs
Because the framework operates on interpretation rather than measurement, it is inherently hardware-agnostic. No additional devices are required, and the same analytical pipeline can be applied retrospectively to existing datasets.
Full-spectrum EIS remains the most information-rich diagnostic signal when available. The virtual EIS framework does not replace it, but extracts meaningful diagnostic insight when only partial impedance information is practical.
Conclusion
Impedance has long been recognized as a powerful diagnostic signal for batteries, but its operational use has been constrained by measurement complexity. By shifting the focus from measurement to interpretation, the impedance-native framework developed at Energsoft demonstrates how existing impedance-related data — including pulse response and dQ/dV behavior — can be unified, enriched, and translated into earlier, mechanistically interpretable diagnostic decisions. This approach provides a practical path toward scalable battery diagnostics that complement telemetry-based analytics without requiring new hardware.
This article was written by Slava Agafonov, CEO, Energsoft Inc. (Seattle, WA). For more information, visit here .
