Formation charging reveals hidden manufacturing defects in lithium-ion batteries

Technical Summary: Formation Charging Reveals Hidden Manufacturing Defects in Lithium-Ion Batteries

Problem Statement
Manufacturing defects in lithium-ion batteries (LIBs) present a persistent safety challenge, often leading to abrupt spontaneous fires in electric vehicles (EVs). These defects, which include metallic foreign matter, electrode cracking, and separator pore closure, are introduced during fabrication and remain sealed within the cell. Crucially, they frequently produce no discernible abnormalities in routine current, voltage, or temperature signals before a hazardous failure occurs. Existing detection strategies generally fall into two categories: dedicated inspection hardware (e.g., X-ray, ultrasound, visual cameras) which increases instrumentation complexity, or electrical feature extraction using multivariate classification of capacity, voltage, and resistance. However, these electrical approaches often require complex, defect-specific decision rules and fail to provide a unified signature across diverse defect morphologies.

Methodology
To address the heterogeneity of defects, the authors constructed a controlled dataset of 55 pouch cells (2 Ah) manufactured on an industrial-scale production line at CALB Group. The set included 30 normal cells and 25 defective cells representing five distinct defect classes derived from real failure cases:

1. Metallic foreign matter
2. Edge active-material loss
3. Tab tearing
4. Electrode cracking
5. Separator pore closure

All cells underwent an identical, standardized formation protocol involving multi-step constant-current (CC) charging, a 50-hour rest, constant-current–constant-voltage (CCCV) charging, and discharge. The study focused on analyzing the electrical response during the formation stage, specifically the CCCV charging phase.

The authors introduced a new metric, the charge-acceptance partitioning index ($r_{CV}$), defined as:
$$r_{CV} = \frac{Q_{CV}}{Q_{CC} + Q_{CV}} \times 100\%$$
where $Q_{CC}$ and $Q_{CV}$ are the capacities charged during the constant-current and constant-voltage stages, respectively. This index quantifies the fraction of total charge accepted during the CV stage. The authors compared $r_{CV}$ against a broad set of conventional production-line features (capacity, terminal voltage, K-value, resistance, incremental capacity) and developed a mechanistic model using three descriptors: side-reaction charge consumption ($M_S$), local heterogeneity ($M_H$), and transport–kinetic deviation ($M_T$).

Key Results

1. Unified Detection via $r_{CV}$: Despite the diverse morphologies and failure pathways of the five defect classes, all defective cells exhibited a consistent downward shift in $r_{CV}$ compared to normal cells.

   • Normal cells (Group F): $r_{CV}$ range 4.117%–4.640% (Median: 4.358%).
   • Defective cells (Groups G–K): $r_{CV}$ range 3.357%–4.075% (Median: 3.805%).
   • A single threshold applied to $r_{CV}$ completely separated all 25 defective cells from the 30 normal cells with zero false positives.

2. Superiority Over Conventional Metrics: Conventional quality metrics (capacity, terminal voltage, K-value, resistance) and features extracted from other stages (such as the three-stage CC charge) showed type-dependent responses. While some features detected specific defects, they failed to provide a unified separation across all defect classes under a single decision rule. Furthermore, calculating $r_{CV}$ from the first-cycle CCCV charge (post-formation) failed to maintain this separation, highlighting the unique sensitivity of the formation stage to defect-induced abnormalities.

3. Mechanistic Interpretation: Mechanistic analysis revealed that while the defects are morphologically distinct, they converge on the same electrical consequence by perturbing terminal charge acceptance through three distinct routes:

   • Side-reaction charge consumption ($M_S$): Dominant in metallic foreign matter (parasitic reactions/leakage).
   • Local heterogeneity ($M_H$): Prominent in edge material loss, tab tearing, and electrode cracking (state mismatch).
   • Transport–kinetic deviation ($M_T$): Evident in separator pore closure and other structural defects (ion transport obstruction).
     A linear model combining these three descriptors successfully reproduced the observed decrease in $r_{CV}$ (Pearson correlation: 0.871), confirming that the index reduction is a shared consequence of diverse defect mechanisms.

Significance and Claims
The paper claims that formation charging offers a "unified, interpretable, and low-cost production-line signature" for detecting manufacturing defects. The primary contribution is the demonstration that heterogeneous manufacturing defects converge onto a common electrical response—the reduction of the charge-acceptance partitioning index during the CV stage of formation charging.

Unlike previous methods that rely on adding inspection hardware or expanding the feature space with complex multivariate models, this approach utilizes a single, process-informed feature ($r_{CV}$) derived from data already recorded during standard formation. This allows for the separation of defective cells from normal ones using a simple threshold, addressing the challenge of defect diversity without increasing testing complexity. The authors position this as a step toward simpler, more robust quality control that captures early-stage abnormalities before they develop into pronounced failures, while noting that future work is needed to validate generalizability across larger datasets, varying defect severities, and diverse cell chemistries.