Abstract
Although the battery surface temperature is commonly measured, the core temperature of a cell may be much higher hence more critical than the surface temperature. The core temperature of a battery, though usually unmeasured in commercial applications, can be estimated by an observer, based on a lumped-parameter battery thermal model and the measurement of the current and the surface temperature. Even with a closed loop observer based on the measured surface temperature, the accuracy of the core temperature estimation depends on the model parameters. For such purpose, an online parameterization methodology and an adaptive observer are designed for a cylindrical battery. The single cell thermal model is then scaled up to create a battery cluster model to investigate the temperature pattern of the cluster. The modeled thermal interconnections between cells include cell to cell heat conduction and convection to the surrounding coolant flow. An observability analysis is performed on the cluster before designing a closed loop observer for the pack. Based on the analysis, guidelines for determining the minimum number of required sensors and their exact locations are derived that guarantee the observability of all temperature states.
Introduction
Lithium-ion batteries are highly sensitive to thermal conditions, with optimal performance typically restricted to a narrow window between -10°C and 50°C. While surface temperature is easily monitored, the internal core temperature—where heat is generated—is the primary driver of degradation and safety risks. In commercial electric vehicles (EVs), placing sensors inside every cell is impractical. This research develops a framework to estimate these hidden core temperatures using only surface measurements and current data, addressing the challenge of parameter drift as batteries age.
Methodology
Lumped-Parameter Modeling
The foundation of the approach is a second-order thermal model for a single cylindrical cell. The cell is treated as two distinct nodes: the core ($T_c$) and the surface ($T_s$). This model accounts for internal heat generation ($Q$), thermal resistance between the core and surface ($R_c$), and convective cooling to the fluid ($T_f$) through resistance $R_u$.
Online Parameterization and Adaptive Observer
To account for battery aging and manufacturing variability, the system utilizes Recursive Least Squares (RLS) to identify thermal parameters online. This allows the model to adapt to the battery's evolving thermal properties in real-time. A closed-loop adaptive observer then uses these identified parameters to estimate the core temperature, correcting its predictions based on the error between measured and predicted surface temperatures.
Scalable Cluster and Observability Analysis
The single-cell model is scaled to a 1D string of cells, incorporating cell-to-cell conduction and coolant-driven convection. A critical contribution of this work is the observability analysis, which determines the minimum number of sensors required to reconstruct the temperature of every cell in a cluster. The analysis reveals that coolant flow creates a unidirectional dependency, which is a key factor in system observability.
Validation and Results
The methodology was validated using high-fidelity simulations of A123 32157 LiFePO4/graphite cells. To ensure "persistent excitation" for the RLS algorithm, the Urban Assault Cycle (UAC) was employed, featuring high-power excursions up to 20C discharge.
Key Findings
- Rapid Convergence: Online parameter identification converged within approximately 10 minutes, while the core temperature estimate stabilized in about 5 minutes.
- The 1/3 Rule: For a 12-cell string, only 4 sensors (one for every three cells) are mathematically required to achieve full observability of all 24 thermal states, provided they are distributed correctly.
- Sensor Placement: The study proved that sensor location is as vital as quantity; for instance, placing sensors at the start and end of a 5-cell string provides full state knowledge, whereas placing them on the first two cells does not.
Conclusion
This research provides a robust framework for reducing the sensor count in battery packs while maintaining high-accuracy thermal monitoring. Future work will explore extending this 1D analysis to complex 3D pack structures and accounting for non-uniform degradation across cell clusters.
Summary updated