Ningning Wei

Thermal runaway tests of nominally similar lithium-ion cells can yield substantially different consequences. We examine severity variability using 365 tests from an open battery failure databank and an empirical Thermal Runaway Severity Uncertainty Amplification (TR-SUA) descriptor, defined as the ratio of post-failure to recorded pre-test coefficients of variation (CVs), not as a measured physical gain. Forty-six near-replicate groups (349 tests) were defined by recorded cell description, format, and trigger ( n ≥ 3). The median available three-descriptor input CV was 0.104%, versus output CVs of 7.577% for corrected total energy, 14.730% for ejecta energy, 21.943% for unrecovered mass, and 21.903% for unrecovered-mass fraction. With a 1%-point input-CV floor, median ratios ranged from 7.577 to 21.943 and remained sensitive to alternative input definitions and floors. Exhaustive within-group log decomposition identified 1,564 valid test pairs; among 1,094 pairs with at least 10% ejecta-energy contrast, the absolute EPR term exceeded the total-energy term in 58.0%, which describes an algebraic partition, not mechanistic causality. Cluster-bootstrap P95 intervals were appreciably broad, and model-group/format holdouts revealed poor predictive transfer despite high random-fold classification AUC. The results support reporting severity dispersion and uncertainty-qualified upper-tail outcomes, while unobserved cell/protocol conditions prevent a causal physical-amplification claim.

Additional sensor modalities may improve short-horizon lithium-ion battery thermal-runaway (TR) forecasting within familiar experimental domains, but whether that benefit persists in unseen cell/source families is unclear. This study evaluated cross-family transferability using 67 mechanically induced TR experiments from six cell/source families. Past 3-s histories of temperature (T), terminal voltage (V), and penetrator load/force (F) were used to forecast 1, 2, and 5 s temperature increments. At the focal 2 s pre-onset horizon, Extra Trees utilities for T + V, T + F, and T + V + F were 11.2%, 17.5%, and 12.4% under experiment-grouped random validation but −3.5%, −6.3%, and −14.3% under leave-one-family-out validation. Removing absolute V(t) reduced the voltage transfer penalty point estimate from 14.7 to 1.8 percentage points; further relative scaling produced −0.8 percentage points but no consistent incremental benefit. Across six families, joint voltage–force feature discrepancy showed a conditional, small-sample association with transfer penalty under the dynamic maximum target, (Spearman ρ = 1.00, exact p = 0.0028); this association attenuated when the median thermocouple (TC) temperature or first thermocouple channel (TC1) was used. Mechanical sensor transfer estimates likewise changed materially with target construction. Within this mechanically induced TR dataset, these results show that apparent multimodal gains depend on validation design, sensor representation, and target temperature definition. Cross-family claims should therefore be supported by family domain holdout validation and harmonized input and target representations. The focal pre-onset analysis is short-horizon regression conditional on a retrospectively defined window (−10 ≤ τ < −2 s relative to operational TR onset), not an autonomous prospective warning trigger.

To mitigate localized erosion of pantograph–catenary system (PCS) contacts caused by random migration of arcs during contact loss, a periodic transverse magnetic field generated by a rotating permanent magnet was proposed to regulate arc motion. A synchronized high-speed imaging and electrical-signal acquisition platform was established to compare arc dynamics, current-collection characteristics, and carbon-electrode erosion under no field, a static field, and periodic fields of 5, 10, and 30 Hz at 120 V DC and 25 A, while SEM/EDS characterized surface morphology and elemental composition. The static field shortened the arcing duration and continuously stretched the arc column toward one side, but produced directional surface grooves. Under periodic fields, the arc column alternately deflected and contracted. Among rotating-field cases, 10 Hz yielded a normalized RMS current ratio of 72.923% and the lowest current fluctuation coefficient of 13.509%, whereas 30 Hz showed no further improvement in electrical or erosion performance. These results indicate that erosion mitigation is associated with both arcing energy and arc-root attachment-site renewal, and that the half-cycle at 10 Hz provides a favorable balance among arc migration, current-collection stability, and thermal-load dispersion, providing experimental guidance for selecting magnetic-drive parameters for PCS arcs.

Lithium-ion batteries in unmanned aerial vehicles (UAVs) experience highly dynamic electrical loads, but thermal warning criteria are rarely interpreted against non-abuse aviation-oriented operating states. This study integrated three independent datasets: 196 DJI Matrice 100 flights, 13,326 validated Sony-Murata VTC6 laboratory missions from 16 cells, and 67 high-temperature-rise abuse experiments from six cell/source families. Real UAV operation showed sustained electrical demand, with a median mission P₉₅ power of 643.1 W; each additional 250 g payload was associated with a 51.1 W increase in P₉₅ power. Among non-abuse missions, 942/13,326 (7.07%) reached at least 60 °C, and median mission maximum temperature increased by 14.28 °C from the first to final within-cell progression decile. However, the conservative maximum non-abuse heating rate at T ≥ 60 °C was 0.451 °C·s⁻¹, below the minimum same-thermocouple value of 0.900 °C·s⁻¹ in the abuse cohort. A literature-based criterion of T ≥ 60 °C and dT/dt ≥ 1 °C·s⁻¹ produced no observed non-abuse activations and activated in 66/67 abuse experiments. Thus, temperature-domain overlap increased across recorded mission progression, whereas empirical thermal-rate separation persisted.

High-throughput computational databases have become central to the discovery and screening of inorganic materials, yet the reproducibility of candidate selection across database versions has rarely been quantified. Here, a frozen 2018 Materials Project snapshot was harmonized with database version 2026.04.13 to evaluate the cross-version stability of thermodynamic and electronic screening. Among 78,756 historical mp-* records in the primary longitudinal cohort, 76,931 remained traceable (97.68%). Despite high cross-version property rank agreement, screening reversal rates ranged from 1.2% to 8.1% across the 16 combined threshold pairs. Under the representative criterion E_hull ≤ 0.05 eV·atom⁻¹ and E_g ≥ 2 eV, 15,754 materials were selected in 2018 and 14,191 in 2026; 1818 original candidates exited and 255 entered, corresponding to 88.46% candidate retention and a Jaccard similarity of 0.871. Instability was concentrated near decision boundaries and remained chemically structured after multivariable adjustment. Cl-, Br-, and N-containing materials were dominated by energy-above-hull threshold crossings, whereas the 3d-series metal group showed a larger contribution from band-gap changes. These results distinguish property reproducibility from decision reproducibility: strong cross-version property agreement does not guarantee stable candidate sets. The database version should therefore be treated as essential reproducibility metadata in high-throughput inorganic materials screening.

Component calorimetry is widely used to assess the thermal stability of lithium-ion battery cathode materials, yet it remains unclear which material-scale descriptors retain consistent relationships with whole-cell self-heating kinetics across independent datasets and test protocols. Here, raw accelerating rate calorimetry (ARC) and thermogravimetric/differential scanning calorimetry (TGA/DSC) data for lithium cobalt oxide (LCO), lithium iron phosphate (LFP), and nickel–cobalt–aluminum oxide (NCA) cells were reanalyzed using a physics-guided event framework. Cathode decomposition onset, event-specific apparent exothermic energy, and associated mass loss were compared with the maximum whole-cell self-heating rate. In the Sandia dataset, cathode onset and event-specific exothermic energy showed large matched-state-of-charge correlations with whole-cell self-heating kinetics (r = −0.983 and 0.980, respectively). In an independently generated dataset, the cathode–electrolyte onset relationship was substantially weaker (r = −0.376), whereas cathode–electrolyte reaction enthalpy retained a large association (r = 0.886). Rank-based sensitivity analysis preserved this contrast, whereas leave-one-chemistry-out stress tests were unstable for all three Sandia linear models, indicating that the strong within-dataset associations did not extrapolate reliably to an unseen chemistry. Because the two heat-release metrics were obtained under different sample configurations and calorimetric protocols, the cross-study comparison is interpreted at the descriptor-class level rather than as direct metric equivalence. Across the two conventional lithium-ion datasets examined, cathode-associated exothermic magnitude showed greater cross-dataset consistency than onset temperature alone. However, neither descriptor supported a universal chemistry-independent scaling law, indicating that component-to-cell relationships remain dependent on experimental protocol, electrode interactions, cell architecture, and battery platform.

Thermal runaway (TR) of lithium-ion batteries generates both gaseous and solid ejecta, yet the material characteristics of the expelled solids and their surrounding thermal-flow environment are often examined separately. In this study, solid ejecta recovered from three repeated thermal-runaway tests of 58 Ah Li(Ni_0.8Co_0.1Mn_0.1)O₂ (NCM811) cells under nitrogen were characterized using scanning electron microscopy (SEM) and energy-dispersive X-ray spectroscopy (EDS). A separate two-dimensional reacting-jet model was established to provide representative near-field thermal-flow context under a high-intensity ejection condition in ambient air. SEM observations at 500×–10,000× revealed irregular fragments, lamellar structures, agglomerates, and fine-particle deposits, indicating pronounced morphological heterogeneity of the recovered solids. In the local selected EDS region, Ni, Mn, and Co collectively accounted for 54.61 wt.% of the normalized elemental signal, together with substantial C, O, and F. The numerical model employed a reduced CO₂-CO-H₂-O₂ mixture, a 12.5 mm vent inlet width, an inlet velocity of 90 m/s, and an inlet temperature of 941.2 °C. The maximum resolved velocity reached approximately 104 m/s near the vent, whereas the maximum simulated temperature was approximately 1806 °C, which occurred farther downstream. These results provide complementary insight into the post-failure material characteristics of solid ejecta and the representative near-field thermal-flow behavior associated with severe lithium-ion battery thermal runaway. The numerical results are intended as representative spatial context rather than a direct reconstruction of the nitrogen-atmosphere experiments.

Reliable management of lithium-ion battery thermal runaway requires knowing not only when abnormal heating begins, but also how far future temperature evolution can be predicted with useful accuracy. This study evaluates that question using 67 abuse experiments from six cell/source families under strict leave-one-family-out validation. An interpretable temperature-only forecasting model was developed from the recent thermal trend and compared with four causal reference methods, including short-memory autoregressive and Kalman state-space approaches. Forecasting error increased markedly with prediction horizon: the mean experiment-normalized MAE of the proposed model was 2.09% at 1 s, 9.73% at 5 s, and 38.59% at 20 s. The median predictability horizons were 2 s and 4 s for 5% and 10% error tolerances, respectively. Forecastability also varied strongly within each event. Using an independent causal temperature-rate criterion, forecasting error reached a pronounced maximum around the transition into rapid heating and decreased after rapid escalation became established. This pattern remained consistent across alternative onset thresholds. These results indicate that, for the temperature-only cross-family setting examined here, deterministic forecasts are most reliable over short horizons and become particularly uncertain around rapid thermal escalation. The framework therefore provides an interpretable benchmark for assessing practical thermal-runaway forecastability rather than a universal limit on all possible forecasting methods.

Thermal runaway (TR) severity can be ranked using absolute or energy-normalised heat and by the pathway through which heat leaves the cell, but whether these definitions identify the same high-hazard cells remains unclear. We analysed 98 non-ISC heater-triggered tests from 13 commercial or near-commercial cylindrical cell models in the Battery Failure Databank. Cathode chemistry was not analysed because it was incompletely documented. Total, body, and ejecta heat were evaluated using allometric scaling, S-FTRC-generation-adjusted fixed-effects models, hierarchical bootstrap ranking, and isometric log-ratio analysis. Scaling exponents for total, ejecta, and body heat were 0.899, 1.003, and 0.879, respectively, with all 95% bootstrap intervals including b = 1. Despite this approximate proportional scaling, generation-adjusted absolute-total and total-kJ/Wh rankings disagreed for 46.2% of cell-model pairs. Body- and ejecta-heat rankings diverged even more strongly, with 62.8% point discordance, a bootstrap median of 61.5% (95% interval 53.8–67.9%), and 28 of 78 pairs showing reversal probabilities ≥95%. These findings show that absolute heat, energy-normalised heat, body heat, and ejecta heat address different engineering questions and should be reported and interpreted separately rather than combined into a universal cell-safety ranking.

Silicone oil is a promising dielectric coolant for battery immersion cooling, yet the ability of commercially available conductivity sensors to detect battery-related contamination remains poorly quantified. In this study, simulated carbonaceous particles, electrolyte, and mixed solid–liquid contaminants were introduced into silicone oil under controlled conditions using a closed-loop circulation platform, and conductivity was monitored in real time. Pristine silicone oil exhibited a baseline conductivity near the instrumental detection limit (approximately 1.26 μS·cm⁻¹). No measurable conductivity increase was observed for particle concentrations up to 10 g·L⁻¹ or electrolyte additions up to 3.0 vol%. Only under an intentionally extreme condition involving 20 vol% electrolyte and vigorous mixing were transient conductivity spikes of 350–550 μS·cm⁻¹ detected. Thus, within the application-relevant concentration range examined, conductivity monitoring showed limited sensitivity to progressive contamination. These findings concern the response of a commercial low-field conductivity sensor and do not constitute a complete assessment of leakage current, dielectric strength, or full thermal-runaway conditions.