the battery worked

in the lab

Problem statement

Lithium-ion battery systems behave fundamentally differently at grid scale than laboratory tests predict. Subtle electrical, thermal, and balance-of-plant interactions that barely register in controlled experiments become critical at megawatt scale, leading to accelerated degradation, unexpected capacity fade, and — in worst cases — catastrophic thermal runaway. On January 16, 2025, fire destroyed the Phase I building of Vistra's Moss Landing Energy Storage Facility in California — a 300 MW / 1,200 MWh installation that was the world's largest battery when it entered service in 2020, part of a complex that had grown to 750 MW / 3,000 MWh by 2023 — forcing the closure of Highway 1 and the evacuation of about 1,200 residents. A single 1 GWh battery installation contains stored energy comparable to hundreds of tons of TNT. Yet the battery management systems (BMS) that monitor these installations rely on cell-level models derived from lab testing that systematically miss the emergent failure modes of large-scale deployment.

Why this matters

Global energy storage installations are projected to exceed 1 TWh cumulative by 2030 (BloombergNEF) — a roughly 15-fold increase over 2020. Real-world efficiency falls well short of cell-level lab measurements: the U.S. utility-scale battery fleet returns about 80% of the electricity it stores (EIA), and analyses accounting for inverter losses, thermal management, control systems, and auxiliary loads put field round-trip efficiency closer to 70%, versus the 85%-plus manufacturers cite from controlled testing (NCEA, 2025). Annual degradation rates of 3–7% at grid scale (NCEA, 2025) further erode the economics. If the industry cannot predict how batteries will actually behave at scale, it cannot accurately price storage contracts, dimension warranty terms, or — most critically — prevent safety incidents that could undermine public acceptance of the entire technology.

What’s been tried and why it hasn’t worked

Standard battery characterization protocols (IEC 62660, UL 9540A) test cells and modules under controlled temperature, humidity, and cycling conditions that don't replicate the thermal gradients, vibration, and uneven current distribution of a shipping-container-scale battery. Multi-physics simulation tools attempt to bridge this gap but require accurate parameterization that varies with manufacturing batch, age, and operating history. Field-deployed battery management systems monitor voltage, current, and temperature at the module level but cannot directly observe the internal electrochemical processes (SEI layer growth, lithium plating, transition metal dissolution) that drive degradation and thermal runaway. Post-mortem analysis of failed cells reveals the mechanisms, but by then the damage is done. Machine learning approaches for degradation prediction have shown promise in lab settings but struggle to generalize across different cell chemistries, manufacturers, and operating environments due to the lack of large, standardized field degradation datasets.

What would unlock progress

A breakthrough in non-invasive, real-time sensing of internal battery state at the cell level — beyond voltage, current, and external temperature — would transform grid battery management. Promising approaches include embedded optical fiber sensors for internal temperature and strain, ultrasonic probes for detecting lithium plating and gas formation, and electrochemical impedance spectroscopy adapted for continuous in-situ monitoring during operation. Equally important is the development of standardized, open-source field degradation datasets from real grid deployments (analogous to weather station networks) that could train more robust predictive models. The University of Sheffield's CREESA program is pioneering real-world diagnostics on a grid-connected 2 MW / 1 MWh testbed, but such instrumented, research-grade grid installations remain rare.

Entry points for student teams

A student team could build a small-scale battery pack (4S or larger lithium-ion configuration) and systematically compare degradation and thermal behavior under idealized cycling (constant temperature, uniform current) versus "realistic" cycling (variable loads, temperature fluctuations, uneven cell balancing). Measuring how quickly the pack's behavior diverges from single-cell predictions would quantify the scale-up problem in miniature. Alternatively, a team could analyze publicly available incident reports (from NFPA, DOE Global Energy Storage Database) to identify common precursors to grid battery failures and assess whether existing BMS monitoring would have detected them. Skills in electrochemistry, thermal engineering, data science, and systems engineering would be most relevant.