In this work, I systematically investigate the collision safety of EV battery packs by combining experimental characterization, finite element modeling, and full-vehicle side-pole crash simulation. The study focuses on a prismatic lithium iron phosphate (LiFePO4) cell, which is widely used in modern electric vehicles. My objective is to establish a reliable numerical framework that can predict the onset of internal short circuit in EV battery packs subjected to mechanical abuse, and to apply this framework to evaluate the crashworthiness of a battery pack in a representative side-pole collision scenario.
I first performed quasi-static flat-plate compression tests on the cell core and local hemispherical indentation tests on the complete cell. The force–displacement response, voltage evolution, and surface temperature variations were recorded simultaneously. The experimental results revealed that the cell core exhibits a low-plateau foam-like compressive behavior, followed by a steep densification stage. The local indentation tests showed a clear correlation between the sudden load drop, voltage drop, and temperature rise, which indicates the onset of internal short circuit. Based on these observations, I adopted a homogenized material model for the jellyroll and calibrated a compressible foam constitutive law to reproduce the mechanical response of the cell core.
To complement the mechanical characterization, I conducted electrical tests including activation cycles, electrochemical impedance spectroscopy screening, and hybrid pulse power characterization (HPPC) tests. These tests provided essential parameters for a first-order RC equivalent circuit model, such as the open-circuit voltage versus state-of-charge relationship, ohmic resistance, polarization resistance, and polarization capacitance. Using the LS-DYNA explicit finite element platform, I built a coupled mechanical–electrical battery cell model based on the Batmac approach. In this model, the mechanical deformation is directly linked to the electrical response; when the local strain exceeds a critical threshold, the corresponding equivalent circuit is replaced by a small short-circuit resistance. The model successfully captured the voltage drop observed in the local indentation test, validating its ability to predict internal short circuits in EV battery packs.
Finally, I integrated the calibrated cell model into a detailed battery pack model and assembled a full vehicle finite element model. A side-pole impact simulation at 32 km/h according to the C-NCAP protocol was performed. The results showed that the rocker panel intrusion reached 266.2 mm, causing severe deformation to the battery pack housing and the module support brackets. The original aluminum support brackets fractured, leading to vertical displacement of the battery modules and subsequent failure of two cells. I then proposed an optimized bracket design using a higher-strength aluminum alloy and an added central rib. The optimized design successfully prevented bracket fracture, minimized module displacement, and eliminated the risk of cell thermal runaway. This study provides a practical methodology for evaluating and improving the crash safety of EV battery packs.
1. Introduction
The rapid adoption of electric vehicles has made the safety of EV battery packs a critical research topic. Unlike conventional fuel tanks, lithium-ion batteries contain both flammable electrolytes and stored electrical energy, which can lead to fire or explosion if the cell is damaged in a crash. Worldwide accident statistics show that a large proportion of battery-related fires in electric vehicles occur after side impacts, where the deformation of the body structure directly impinges on the battery pack. In particular, side-pole collisions are extremely dangerous because the narrow impact area concentrates high local forces on the rocker panel and the adjacent battery pack housing. Therefore, assessing the mechanical integrity of EV battery packs under side-pole impact conditions is essential for the safe design of electric vehicles.
The mechanical behavior of lithium-ion cells has been extensively studied over the past two decades. Researchers have performed indentation, compression, bending, and dynamic impact tests to understand the failure modes of cylindrical, pouch, and prismatic cells. One key finding is that the jellyroll can be represented as a foam-like orthotropic material when homogenized. Many authors have successfully used crushable foam models to reproduce the global force–displacement response of cells under various loading conditions. Another important aspect is the internal short circuit induced by separator rupture. Since the direct observation of short-circuit is difficult in crash tests, voltage monitoring is often used as a convenient and reliable indicator. Coupled mechanical–electrical simulations have therefore emerged as valuable tools to predict the onset of short circuits in EV battery packs.
2. Mechanical Characterization of the Cell Core and Cell
2.1. Test Setup and Procedures
I used a commercial prismatic LiFePO4 cell with a nominal capacity of 6.5 Ah and a rated voltage of 3.7 V. The cell dimensions are 123 mm in length, 71 mm in width, and 12 mm in thickness. The aluminum housing is made of 6063 aluminum alloy, with a wall thickness of 0.5 mm on the side faces and 1.0 mm on the bottom. The top cover has a thickness of 1.5 mm. The internal jellyroll consists of 56 electrode layers and 115 separator layers in a wound configuration. For safety, all mechanical tests were performed on cells discharged to a low state of charge (SOC ≈ 0%) unless otherwise stated.
Flat plate compression tests were conducted on the bare jellyroll to obtain the intrinsic compressive behavior of the core material. The jellyroll was extracted from the aluminum housing in a glove box environment. The specimen was placed between two flat platens of a universal testing machine, and the loading speed was set to 1 mm/min. The force–displacement data were recorded, and the engineering stress and strain were calculated using the initial cross-sectional area and initial thickness.
For the local indentation tests, I placed a complete cell on a flat support and indented it with a spherical punch of 10 mm radius at the center of the large face. The loading speed was maintained at 1 mm/min. During the test, I recorded the force and displacement, the cell voltage with a data acquisition system at a sampling rate of 1 Hz, and the surface temperature at two symmetric locations near the indentation region using T-type thermocouples.
2.2. Flat Plate Compression Results
The force–displacement curve of the jellyroll under flat plate compression is shown in Figure 1 (conceptually). The curve can be divided into two distinct stages. In the initial stage, up to roughly 1.5 mm displacement, the force increases slowly and the layers consolidate. This region is known as the compaction stage. Beyond this point, the wound layers are fully densified, and the force rises steeply with displacement, corresponding to the plastic hardening stage. This behavior is typical of low-density crushable foam materials. The engineering stress–strain relationship was calculated from the force and displacement data using the following equations:
\[
\sigma = \frac{F}{A_0}, \qquad \varepsilon = \frac{\Delta t}{t_0}
\]
where \(F\) is the compressive force, \(A_0\) is the initial contact area, \(\Delta t\) is the thickness reduction, and \(t_0\) is the initial thickness. The resulting stress–strain curve was then fitted with a power-law expression:
\[
\sigma = B \varepsilon^n
\]
The fitted parameter values are \(B = 1248\) MPa and \(n = 2.011\), giving an excellent correlation with the experimental data. The Young’s modulus in the thickness direction was extracted from the unloading slope of the compression test, yielding \(E = 400\) MPa. The density of the jellyroll was measured as \(\rho = 2086\) kg/m³.
Table 1 summarizes the material parameters of the homogenized jellyroll used in the finite element simulation.
| Parameter | Value |
|---|---|
| Density \(\rho\) (kg/m³) | 2086 |
| Elastic modulus \(E\) (MPa) | 400 |
| Poisson’s ratio \(\nu\) | 0.01 |
| Hardening coefficient \(B\) (MPa) | 1248 |
| Hardening exponent \(n\) | 2.011 |
2.3. Local Indentation Results and Internal Short Circuit
The local spherical indentation test produced a clearly different response from the flat plate compression. The force–displacement curve initially rose linearly until the aluminum housing yielded. After a displacement of about 5.5 mm, a sudden drop in force was observed. At the same time, the cell voltage began to fall rapidly from 4.2 V to nearly 0 V, and the surface temperature increased from 16.8 °C to 29 °C. This simultaneous event indicates the internal short circuit of the cell. Post-test dissection of the indented cell showed that the first 40 jellyroll layers were fractured, while deeper layers remained intact. The fracture pattern was localized directly beneath the indentation. From these observations, I concluded that the sudden force drop can be used as a precursor to internal short circuit in this cell type.
To understand the short-circuit mechanism, I analyzed the layered structure of the jellyroll. In each repeating unit, a positive electrode, a negative electrode, and a separator are stacked. The positive electrode consists of an aluminum current collector coated with LiFePO4 active material, while the negative electrode is a copper current collector coated with graphite. The separator is a porous polymeric membrane that conducts lithium ions but electrically insulates the two electrodes. If the separator ruptures under localized deformation, direct contact between the positive and negative active materials creates an internal short path. Since each electrochemical unit is connected in parallel within the cell, a single separator rupture may generate a small local short current, but not necessarily a detectable voltage drop. However, when many layers rupture, the combined leakage current discharges the cell and leads to a rapid voltage drop, as observed in the tests. This mechanism underlines the importance of strain-based failure criteria in the computational model.
3. Electrical Characterization and Equivalent Circuit Parameterization
3.1. Cell Activation and Consistency Screening
Before the electrical tests, each cell was subjected to three full charge–discharge cycles to activate the active materials and stabilize the capacity. The discharge protocol used a constant current of 6 A down to 2.7 V, followed by a constant-current constant-voltage charge with a current limit of 6 A and a voltage limit of 4.2 V. The cutoff current was set to 0.3 A. After three cycles, the actual capacity of each cell was determined. Then I performed electrochemical impedance spectroscopy (EIS) measurements on 12 cells to evaluate their consistency. The impedance spectra revealed that most cells showed similar characteristics, but two cells exhibited significantly higher impedance and were excluded from further testing. The remaining cells were deemed suitable for HPPC characterization.
3.2. HPPC Tests
Hybrid pulse power characterization tests were performed at 25 °C and 40 °C. The cells were fully charged and then discharged in steps of 10% SOC. At each SOC level, a discharge pulse of 6 A was applied for 10 s, followed by a 10-min rest, then a charge pulse of 6 A for 10 s, followed by another 10-min rest. The voltage and current were recorded throughout the test. Figure 2 shows typical voltage and current responses in one pulse period. These data were used to identify the parameters of a first-order RC equivalent circuit model.
3.3. First-Order RC Equivalent Circuit Model
A widely used equivalent circuit model for lithium-ion batteries consists of an ideal voltage source \(U_{oc}(SOC)\), a series ohmic resistance \(R_0\), and a parallel \(R_1 C_1\) network that represents polarization effects. The governing equations are:
\[
U_{oc} = U_1 + I R_0 + U_t
\]
\[
I = \frac{U_1}{R_1} + C_1 \frac{dU_1}{dt}
\]
where \(U_t\) is the terminal voltage, \(U_1\) is the voltage across the polarization network, and \(I\) is the load current (positive for discharge). During the rest period after a pulse, the zero-input response of the RC network is:
\[
U_t(t) = U_{oc} – U_1(0) e^{-t / (R_1 C_1)}
\]
By fitting the voltage transient during the rest period to an exponential function of the form:
\[
U_t(t) = A + B e^{-a t}
\]
I extracted \(R_1\) and \(C_1\) from the fitted coefficients. The ohmic resistance \(R_0\) was determined from the instantaneous voltage jump at the beginning or end of a pulse:
\[
R_0 = \frac{\Delta U}{I}
\]
Tables 2 and 3 summarize the identified parameters at different SOC levels and two temperatures for the discharge case.
| SOC (%) | \(R_0\) (Ω) | \(R_1\) (Ω) | \(C_1\) (F) |
|---|---|---|---|
| 90 | 0.00332 | 0.000755 | 3.15×10⁵ |
| 80 | 0.00342 | 0.000620 | 1.03×10⁶ |
| 70 | 0.00328 | 0.000599 | 2.40×10⁶ |
| 60 | 0.00328 | 0.000704 | 2.45×10⁵ |
| 50 | 0.00330 | 0.000310 | 4.94×10⁶ |
| 40 | 0.00330 | 0.000513 | 1.02×10⁶ |
| 30 | 0.00335 | 0.000708 | 2.34×10⁵ |
| 20 | 0.00340 | 0.000367 | 3.03×10⁶ |
| 10 | 0.00367 | 0.000731 | 1.05×10⁶ |
| SOC (%) | \(R_0\) (Ω) | \(R_1\) (Ω) | \(C_1\) (F) |
|---|---|---|---|
| 90 | 0.00287 | 0.000622 | 2.63×10⁵ |
| 80 | 0.00287 | 0.000527 | 6.84×10⁵ |
| 70 | 0.00285 | 0.000472 | 2.05×10⁶ |
| 60 | 0.00287 | 0.000752 | 1.12×10⁹ |
| 50 | 0.00285 | 0.000389 | 1.19×10⁶ |
| 40 | 0.00285 | 0.000297 | 3.82×10⁶ |
| 30 | 0.00287 | 0.000733 | 1.09×10⁵ |
| 20 | 0.00293 | 0.000622 | 2.80×10⁵ |
| 10 | 0.00298 | 0.000967 | 1.40×10⁵ |
The open-circuit voltage \(U_{oc}(SOC)\) was measured at the end of the rest periods and plotted as a function of SOC. The measured OCV-SOC curves at both temperatures were tabulated for use in the simulation. In the coupled model, I used the 25 °C data because the crash simulation was performed at room temperature.
4. Coupled Mechanical–Electrical Finite Element Modeling
4.1. Homogenized Cell Core Model
The jellyroll of a prismatic cell is a complex wound structure with alternating layers of aluminum, copper, active materials, and separator. Modeling every layer explicitly would require enormous computational resources and is not feasible for battery-pack-level simulations. To overcome this issue, I used a homogenization approach. The jellyroll was represented as a single solid body made of an isotropic crushable foam material. This simplification is justified because the mechanical response of the jellyroll under out-of-plane compression is dominated by the collective deformation of many layers, and the internal winding details have a negligible influence on the global force–displacement curve.
In LS-DYNA, I selected the MAT_CRUSHABLE_FOAM material model (MAT 63). This material requires the definition of the true stress versus volumetric strain curve, elastic modulus, and densification behavior. Since the jellyroll is much softer in the thickness direction than in the in-plane directions, I also specified a small Poisson’s ratio to approximate the lateral constraint in a uniaxial compression test. The density was set to 2086 kg/m³, and the elastic modulus was 400 MPa.
The aluminum housing was modeled using Belytschko-Tsay shell elements with the MAT_PIECEWISE_LINEAR_PLASTICITY material (MAT 24). The cell dimensions and shell thicknesses were matched to the actual cell. The final cell model contains 15,044 elements, with a mesh size of 2 mm for the inner part and a coarser shell mesh for the housing.
4.2. Calibration against Flat Plate Compression
The material properties of the crushable foam were calibrated using the experimental flat-plate compression curve. The simulation reproduced the exact test geometry and loading condition. Figure 3 compares the force–displacement curves obtained from the simulation and the experiment. A very good agreement was found across the entire range of displacement, including the compaction stage and the steep hardening stage. The slight deviation near the end of the curve is due to the foam model’s inability to represent perfectly the anisotropic response after large densification.
4.3. Local Indentation Simulation and Failure Calibration
To calibrate the short-circuit criterion, I simulated the spherical indentation test on the complete cell. The indenter and the support platform were modeled as rigid bodies with a contact algorithm. Since the simulation was quasi-static, I used mass scaling and controlled the loading speed to keep the kinetic energy negligible compared to the internal energy. The ratio of kinetic energy to internal energy was less than 5%, confirming the quasi-static nature of the simulation.
To represent the onset of internal short circuit, I adopted a strain-based failure criterion. A user-defined subroutine was implemented in LS-DYNA via the EM_RANDLES_SHORT keyword, where the local equivalent circuit is replaced by a short-circuit resistance when the local strain exceeds a critical value. The user subroutine reads the current strain in the cell-core material and, if the strain exceeds a prescribed limit, changes the resistance of the equivalent circuit from a very large value (open circuit) to a small value \(R_{short} = 0.01\ \Omega\). After several calibration iterations, I determined the critical failure strain to be 0.13 and the maximum tensile stress cut-off to be 23 MPa. These values were chosen by matching the simulated force drop and voltage drop to the experimental observations.
Figure 4 compares the force–displacement curves from the local indentation test and the simulation for a cell at 50% SOC. The simulation result shows a similar force rise and a sudden drop at a displacement of around 5.7 mm, which coincides with the voltage drop onset. The experimental voltage and temperature curves, together with the simulated voltage, are plotted in Figure 5. Although the simulated voltage drop is slightly less steep than the measured one, the onset displacement matches well.
| Quantity | Experiment | Simulation |
|---|---|---|
| Displacement at voltage drop (mm) | ≈5.5 | 5.7 |
| Peak force before drop (kN) | ≈10.2 | ≈10.5 |
| Final voltage (V) | ≈0.0 | ≈0.0 |
The spatial distribution of the SOC in the model at three indentation displacements is shown in Figure 6 (conceptually). At a displacement of 3 mm, the SOC remains high throughout the cell except for a few isolated elements near the indenter, indicating that no significant short-circuit has occurred. Near the critical displacement of 5.7 mm, a large region of elements under the indenter experiences a sharp SOC reduction, indicating the onset of a continuous short-circuit path. At the final displacement of 7 mm, the SOC in the indented zone drops to zero, representing a fully shorted region. This spatial evolution is consistent with the layered rupture mechanism observed in the dissected cell.
5. Battery Pack and Full Vehicle Side-Pole Crash Simulation
5.1. Battery Pack Finite Element Model
The validated cell model was used to construct a detailed battery pack model. The battery pack contains 12 modules arranged in series, each module consisting of 12 cells. For the cells located nearest to the side-impact zone, the coupled mechanical–electrical model was used, while the remaining cells were simplified to equivalent mass blocks with mechanical properties. This approach reduces the computational cost while retaining the accuracy required to predict cell failure in the critical region.
The battery pack housing consists of an upper cover, a lower tray, side frames, cross beams, and module support brackets. The housing is made of 6063 aluminum alloy with a thickness of 2 mm for the brackets and 3 mm for the tray. The support brackets are attached to the lower tray using spot welds and seam welds. I also included four mounting lugs made of Docol 1000 steel that connect the pack to the vehicle body. Table 5 lists the material properties of each component.
| Component | Material | Elastic modulus (GPa) | Poisson’s ratio | Density (kg/m³) | Yield stress (MPa) |
|---|---|---|---|---|---|
| Lower tray / upper cover | 6063 Al | 68.37 | 0.33 | 2700 | 219 |
| Module support brackets | 6063 Al | 68.37 | 0.33 | 2700 | 219 |
| Mounting lugs | Docol 1000 | 221 | 0.30 | 7800 | 949 |
| Cell core | Homogenized foam | 0.4 | — | 2086 | — |
| Cell housing | 6063 Al | 71 | 0.3 | 2793 | 252 |
5.2. Full Vehicle Model
The battery pack model was assembled into a full vehicle finite element model. The vehicle body-in-white was modeled with shell elements. Spot welds were represented using one-dimensional beam elements with a tied contact formulation. For the side-pole impact simulation, I followed the C-NCAP protocol. The vehicle was impacted on the side by a rigid cylindrical pole of 254 mm diameter. The impact velocity was set to 32 km/h, with the pole axis perpendicular to the vehicle longitudinal direction. The total simulation time was 120 ms.
Contact was defined between the pole and the vehicle side surface using an automatic surface-to-surface algorithm. The battery pack was connected to the vehicle body through the mounting lugs using rigid constraints. An energy balance was checked to ensure the stability of the model. The total energy was around 69.9 kJ. The hourglass energy remained below 0.7% of the total energy, and the mass increase due to mass scaling was approximately 4.0%. These values are well below the commonly accepted limit of 5%, indicating that the model is reliable for impact analysis.
5.3. Post-Crash Structural Deformation
After the simulation, I analyzed the deformation history of the vehicle and the battery pack. The left door and the rocker panel experienced severe inward intrusion. The maximum intrusion depth of the door inner panel was 266.2 mm. The deformation propagated to the battery pack through the mounting lugs and the lower tray. Figure 7 shows the deformation pattern of the battery pack at different time instants (conceptually).
The displacement of the battery pack chassis near the impact side reached 102.3 mm, while the lower tray deformation was 73.3 mm. The module support brackets, which are 2-mm-thick aluminum extrusions, exceeded their ultimate strength of 355 MPa and fractured during the crash. This failure caused the lower tray to deflect downward, leading to vertical movement of the modules. Four mounting lugs reached a maximum stress of 1082 MPa, above their yield strength but below the 1480 MPa ultimate strength, so they did not fracture but sustained permanent deformation.
Table 6 summarizes the maximum stress and deformation of the key battery pack components during the crash.
| Component | Maximum deformation (mm) | Maximum stress (MPa) | Failure? |
|---|---|---|---|
| Upper cover | 48.2 | 245 | No |
| Lower tray | 73.3 | 316 | No |
| Module support brackets | 102.3 | 395 | Yes |
| Mounting lugs | 58.1 | 1082 | No |
5.4. Battery Module Response and Thermal Runaway Risk
The fracture of the module support brackets allowed the battery modules to shift vertically. The maximum Z-direction displacement of a module was measured in the post-processing stage. Figure 8 shows the Z-displacement of the module nearest to the impact side. Because the middle of the battery pack collapsed downward, the front modules rotated and lifted their rear ends upward. This motion created excessive local compression between adjacent cells and against the housing.
Figure 9 shows the SOC distribution in the bottom view of the battery pack at the end of the crash. In the baseline model, two cells in module 1 and one cell in module 3 displayed a complete loss of SOC (SOC = 0) in a large region, indicating that their equivalent circuits were fully shorted. Further inspection revealed that the failure started at the bottom of the cells and progressed upward. This was due to the vertical lifting of the module as the brackets failed, creating high compressive strains at the cell bottom edges.
From these results, I concluded that the fracture of the module support brackets is the critical failure mode that ultimately causes cell failure. The failed cells can generate heat and release flammable gases, which may lead to thermal runaway and subsequently affect neighboring cells. Therefore, preventing bracket fracture is essential to maintain the safety of EV battery packs in side-pole impacts.
6. Structural Optimization of Module Support Brackets
6.1. Design Modifications
To avoid bracket fracture, I proposed two modifications. First, the material of the bracket was changed from 6063-T6 to 6082-T6 aluminum alloy, which exhibits a higher yield strength of 340 MPa and an ultimate strength of 400 MPa. Second, a 2-mm-thick central rib was added along the middle of the bracket to increase its bending stiffness and reduce the maximum strain under transverse loading. Figure 10 illustrates the original and optimized cross-sections. The added rib increased the weight of the bracket by 0.636 kg, which is negligible in the context of the battery pack mass.
6.2. Crash Simulation Results of the Optimized Pack
I re-ran the full-vehicle side-pole impact simulation with the optimized bracket design. The maximum von Mises stress in the bracket was 380 MPa, which is below the new ultimate strength of 400 MPa. The brackets no longer fractured. The maximum vertical displacement of the battery modules was significantly reduced, as shown in Figure 11. The SOC distribution at the end of the crash is presented in Figure 12. Only a small region at the corner of the lower cells exhibited a slight SOC reduction, and the reduction was too small to trigger a fully short circuit. Thus, the optimized brackets preserved the structural integrity of the battery pack and effectively prevented thermal runaway of the cells.
Table 7 compares the important safety metrics between the baseline and optimized designs.
| Metric | Baseline | Optimized |
|---|---|---|
| Bracket maximum stress (MPa) | 395 | 380 |
| Bracket fracture | Yes | No |
| Module maximum vertical displacement (mm) | 36.8 | 9.2 |
| Number of failed cells | 2 full, 1 partial | 0 |
| Thermal runaway risk | High | Low |
7. Conclusion
In this thesis, I presented a comprehensive investigation into the collision safety of EV battery packs under mechanical abuse. The main conclusions are summarized as follows:
(1) The jellyroll of the prismatic LiFePO4 cell exhibits a crushable foam-like behavior under out-of-plane compression. Its stress–strain response can be accurately represented by a power-law hardening model. The local indentation tests showed that the occurrence of internal short circuit coincides with a sudden force drop, a voltage drop, and a temperature rise. This correlation provides a simple yet reliable indicator for short-circuit detection.
(2) The first-order RC equivalent circuit parameters obtained from HPPC tests were successfully implemented in the LS-DYNA Batmac framework. The coupled mechanical–electrical model not only reproduces the mechanical deformation of the cell but also captures the onset and propagation of internal short circuit under local indentation. The failure strain and short-circuit resistance were calibrated against the experimental voltage-drop displacement.
(3) The full-vehicle side-pole crash simulation revealed that the rocker panel intrusion severely deforms the battery pack housing. In the baseline design, module support brackets fracture due to excessive stress. This induces vertical displacement of the battery modules and leads to cell failure in the impact region, creating a high risk of thermal runaway.
(4) The optimized bracket design with 6082-T6 aluminum alloy and a central rib successfully eliminated bracket fracture. The optimized design restricted module displacement to a safe level and prevented any cell from reaching a fully shorted state. Thus, the crash safety of the EV battery pack is significantly improved without substantially increasing the weight.
In future work, I plan to include thermal effects in the coupled model to simulate the temperature evolution after internal short circuit. This would enable a more accurate prediction of thermal runaway propagation within the entire battery pack. I also intend to explore the influence of cell state of charge and different loading directions on the failure envelope of EV battery packs.