1. Introduction
Lithium-ion batteries have become the dominant energy storage carrier due to their high energy density, low self-discharge rate, and long cycle life. Among them, lithium iron phosphate (LFP) batteries are extensively deployed in electrochemical energy storage power stations owing to their superior safety characteristics and low cost. However, the development of battery safety technologies has not kept pace with the rapid advancement of battery performance, resulting in increasingly prominent safety concerns. In recent years, fire accidents have occurred frequently, particularly in large-scale energy storage stations, severely hindering the healthy growth of the energy storage industry. Consequently, conducting research on efficient and reliable ultra-early warning methods for thermal runaway in energy storage batteries is of paramount importance.
Previous extensive experimental studies have demonstrated that gas generation within the battery during thermal runaway leads to casing expansion and deformation. This deformation is a macroscopic manifestation of internal pressure accumulation, which is directly correlated with the progression of internal side reactions. Based on this understanding, this thesis focuses on the strain characteristics exhibited by batteries during thermal runaway. The research begins with a theoretical analysis of the strain generation mechanism. Following this, a dedicated thermal runaway experimental platform is constructed to systematically investigate the thermal runaway characteristics and strain parameter evolution of LFP batteries under various operating conditions, including different states of charge (SOC) and micro-overcharge states. Finally, a full charge-discharge cycle thermal runaway ultra-early warning model, based on dynamic strain signals, is developed to enhance the safety monitoring capability of energy storage systems.
2. Experimental Platform and Equipment
To support the experimental research, a multifunctional battery thermal runaway experimental platform was established. The platform provides a controlled, safe, and highly monitorable environment for conducting experiments on LFP batteries.
2.1 Battery Specifications
The battery samples used in this study were prismatic LFP batteries with a nominal capacity of 100 Ah. The key specifications of the battery samples are summarized in the following table.
| Parameter | Specification |
|---|---|
| Nominal Capacity (Ah) | 100 |
| Maximum Cut-off Voltage (V) | 3.6 |
| Minimum Cut-off Voltage (V) | 2.5 |
| Dimensions (H×L×W, cm) | 21.6×13.5×3.5 |
| Mass (kg) | 1.9±0.1 |
| Mass Energy Density (Wh/kg) | 173.7 |
2.2 Experimental Setup
The experimental platform comprises a stainless steel combustion chamber with dimensions of 2 m × 2 m × 2 m, a fume hood, ventilation ducts, and a fan system. The combustion chamber is equipped with a high-temperature-resistant explosion-proof observation window, allowing researchers to visually observe critical phenomena such as battery deformation, gas generation, and safety valve opening without compromising experimental safety.
Key equipment utilized in the experiments includes:
| Equipment | Model | Accuracy |
|---|---|---|
| Strain Data Acquisition Module | CMCU-08 | 0.01% |
| Force Sensor | PLD204DP | 0.25% |
| Thermocouple | K-type | ±1.5 ℃ |
| Battery Tester | Neware 10V-100A | ±0.1% |
| High-precision Electronic Balance | Custom | 0.1 g |
| SONY Camera | 30X, 1080P | 50 fps |
A heating plate (15 cm × 20 cm × 2 mm, 500 W, 220 V) was used for thermal abuse tests. Temperature was monitored using K-type armored thermocouples connected to a multi-channel data acquisition instrument. Strain was measured using high-temperature resistance strain gauges (BA120-3AA, resistance 120.0±0.3 Ω, sensitivity coefficient 2.0±1) bonded to the battery casing with CH-31 adhesive. Additionally, a scanning electron microscope (SEM, TESCAN VEGA Compact) was employed for microstructural characterization of the battery casing before and after thermal runaway.
3. Strain Generation Mechanism and Behavioral Characteristics
3.1 Theoretical Analysis of Strain Generation
Under single-sided heating conditions, the heating plate serves as the primary heat source that triggers internal side reactions. The heat transfer inside the battery can be approximated as a one-dimensional heat conduction process. Materials near the heating side first undergo side reactions, including the decomposition of the solid electrolyte interphase (SEI) film and the reaction between the negative electrode and electrolyte, generating gases. Additionally, some electrolyte vaporizes due to heating. As the heating plate continues to supply heat and side reactions release heat, the internal temperature rises continuously, causing the gas to expand and further increasing the internal pressure. Therefore, the total internal pressure of the battery can be expressed as:
$$P_{int} = P_r + P_{eva} + P_{exp}$$
where \(P_{int}\) is the total internal pressure of the battery, \(P_r\) is the pressure due to gas generation from side reactions, \(P_{eva}\) is the pressure due to electrolyte evaporation, and \(P_{exp}\) is the pressure increase due to thermal expansion of internal gases.
The increase in internal pressure leads to a rise in the stress on the battery casing, manifesting externally as battery expansion and strain increase. Assuming the ideal gas law applies to the internal gases:
$$P_{int}V = nRT$$
where \(n\) is the amount of gas substance, \(V\) is the internal gas volume, \(R\) is the universal gas constant, and \(T\) is temperature.
In practical applications, the large faces and bottom of the battery are typically constrained, so expansion is mainly concentrated in the small face direction. The battery can be simplified to a thin-walled sealed container model. A stress analysis on the small face of the battery yields the following relation:
$$2\sigma_{\phi} s l \sin\phi = P_{int} l^2$$
where \(\sigma_{\phi}\) is the tangential stress, \(s\) is the casing thickness, \(l\) is the side length of the selected square region, and \(\phi\) is the bending angle. Combining these equations yields:
$$\sigma_{\phi} = \frac{nRTl}{2sV\sin\phi}$$
During thermal runaway, the battery casing experiences a complex stress-strain state, including elastic and plastic deformation. Since the stress is in a continuous loading state, the strain increases monotonically with stress:
$$\varepsilon = f(\sigma_{\phi})$$
For a specific battery, the increase in internal gas generation is the fundamental cause of casing strain. Since the strain response to gas generation is instantaneous, strain can effectively reflect the internal state of the battery in real time, making it a more efficient and reliable indicator for early warning before the safety valve opens.
3.2 Spatial Characteristics of Strain Behavior
The battery casing is a thin plate with a thickness-to-minimum-dimension ratio of 1/35. The side shell of the battery can be modeled as a thin plate with fixed boundaries. The boundary conditions are:
$$\omega(0,y) = 0, \quad \omega(a,y) = 0, \quad \omega(x,0) = 0, \quad \omega(x,b) = 0$$
Using the inverse method, the deflection is assumed to be a trigonometric series:
$$\omega = \sum_{m=1}^{\infty}\sum_{n=1}^{\infty} A_{mn}\sin\frac{m\pi x}{a}\sin\frac{n\pi y}{b}$$
Substituting into the basic differential equation of thin plate bending \(D\nabla^2\nabla^2\omega = q\), the deflection can be obtained as:
$$\omega = \frac{16q_0}{\pi^6 D}\sum_{m=1,3,5,\ldots}\sum_{n=1,3,5,\ldots}\frac{\sin\frac{m\pi x}{a}\sin\frac{n\pi y}{b}}{mn\left(\frac{m^2}{a^2}+\frac{n^2}{b^2}\right)^2}$$
The deflection distribution indicates that at the center of the battery casing (\(x = a/2\), \(y = b/2\)), the deflection reaches its maximum value, and therefore the strain is also greatest at this location.
To experimentally verify this, three strain gauges (\(ε_1\), \(ε_2\), and \(ε_3\)) were attached along the vertical direction on the side of a 100% SOC battery. The results showed that the peak strain at the center (\(ε_2\)) reached 8137.6 µε, which was approximately 20% and 30% higher than \(ε_3\) and \(ε_1\), respectively, confirming that the center of the battery side is the strain-sensitive region and the optimal location for sensor placement.
3.3 Typical Strain Evolution Process During Thermal Runaway
Fig. 3-9 in the original thesis illustrates the typical strain evolution of a 100% SOC LFP battery under single-sided heating. The experimental results demonstrate that the process can be divided into five characteristic stages:
| Stage | Time Period | Characteristics |
|---|---|---|
| Stage I | t < tabn | Slow strain increase due to local gas generation |
| Stage II | tabn < t < tεp | Strain rate reaches 10 µε/s, marking the acceleration period |
| Stage III | tεp < t < tV | Peak strain reaching 8137.6 µε, followed by a curved decrease |
| Stage IV | tV < t < tTR | Strain drops linearly after safety valve opens |
| Stage V | t > tTR | Strain rises again during the thermal runaway eruption |
The turning point of the strain rise rate at 10 µε/s coincided with the turning point of the expansion force signal at 7.5 N/s. The strain peaked at 8137.6 µε before the safety valve opened, demonstrating a more significant abrupt change characteristic in the early stage of thermal runaway compared to other parameters such as temperature and expansion force.
SEM Analysis
SEM analysis of the battery casing before and after thermal runaway provided direct evidence of the strain mechanism. The normal battery casing exhibited a smooth, uniformly structured surface. After thermal runaway, the casing surface showed significant wrinkling, visible stretching marks, and even fracture signs at the center region. The fracture edges exhibited distinct tensile tear morphology, confirming that the casing deformation during thermal runaway is essentially a process of stretching, displacement, and eventual structural failure under internal pressure.
3.4 Strain Characteristics Under Normal Charge-Discharge Conditions
To comprehensively understand the strain behavior of the energy storage battery, normal charging and discharging tests were conducted. A full charge-discharge cycle consisted of constant current (CC) charging at 0.5 C, constant voltage (CV) charging to full SOC, and constant current discharging at 0.5 C to 0% SOC.
During the first cycle, the strain evolution showed three distinct phases:
| Phase | Strain Behavior | Peak Value |
|---|---|---|
| CC Charging | Gradual increase, reaching a peak at the end | ~600 µε |
| CV Charging | Linear decrease with a relatively stable rate | Decreasing |
| CC Discharging | Slow increase initially, then rapid increase at the end | ~670 µε |
In subsequent cycles, the strain variation patterns showed a high degree of repeatability. The strain change rate curves for multiple cycles exhibited consistent trends, with fluctuations always remaining below ±0.6 µε/s. Three repeated experiments confirmed that under normal charge-discharge conditions, the peak strain values were consistently below 1000 µε, and the peak strain rates remained stable in the range of 0.3–0.5 µε/s.
| Experiment | Cycle 1 Strain Peak (µε) | Cycle 1 Strain Rate Peak (µε/s) | Cycle 2 Strain Peak (µε) | Cycle 2 Strain Rate Peak (µε/s) |
|---|---|---|---|---|
| Replicate 1 | 670.4 | 0.5 | 667.5 | 0.4 |
| Replicate 2 | 818.2 | 0.3 | 812.9 | 0.3 |
| Replicate 3 | 443.9 | 0.3 | 438.6 | 0.3 |
These results clearly demonstrate that the strain behavior of the energy storage battery under normal operating conditions is markedly different from that during thermal runaway, providing a solid basis for determining warning thresholds.
4. Strain Characteristics Under Key Influencing Factors
4.1 Effect of State of Charge on Strain Characteristics
To investigate the influence of SOC on the thermal runaway strain characteristics of the energy storage battery, thermal runaway experiments were conducted on batteries with SOC levels of 0%, 20%, 40%, 60%, 80%, and 100%. The batteries were heated with a 500 W heating plate until thermal runaway was triggered.
The experimental results revealed that batteries with low SOC (0%, 20%, 40%) only experienced the first four typical stages of thermal runaway without entering the violent thermal runaway phase, while high SOC batteries (60%, 80%, 100%) further developed into full thermal runaway. A crucial finding was that for all SOC levels, the peak strain occurred before the safety valve opened, confirming that the strain peak can serve as a key indicator for early warning.
The strain and expansion force showed an exponential relationship during the early stage of thermal runaway:
$$\varepsilon \propto e^{kF_p}$$
where \(F_p\) is the expansion force. This relationship suggests that the strain can sensitively reflect the internal chemical reaction progress even when temperature changes are minimal.
Fig. 4-4 in the original thesis shows the rate of change curves for strain, expansion force, and temperature. The strain rate exhibited three key characteristic nodes before the safety valve opened: \(t_{abn}\) (when \(dε/dt = 10\) µε/s), \(t_{εrp}\) (when the strain rise rate peaked), and \(t_{εp}\) (when the strain peaked). These nodes were consistently observed across all SOC levels.
| SOC (%) | tabn (s) | tεrp (s) | tεp (s) | tV (s) | tTR (s) | ∆t1 (s) | ∆t2 (s) | ∆t3 (s) |
|---|---|---|---|---|---|---|---|---|
| 0% | 680 | 881 | 999 | 1201 | / | 521 | 320 | 202 |
| 20% | 576 | 787 | 869 | 1017 | / | 441 | 230 | 148 |
| 40% | 534 | 761 | 810 | 979 | / | 445 | 218 | 169 |
| 60% | 646 | 845 | 919 | 1015 | 1235 | 369 | 170 | 96 |
| 80% | 616 | 883 | 954 | 1056 | 1234 | 440 | 173 | 102 |
| 100% | 528 | 768 | 798 | 941 | 1068 | 413 | 173 | 143 |
On average, the three strain-based characteristic nodes preceded the safety valve opening by approximately 438 s, 214 s, and 143 s, respectively. Lower SOC batteries provided longer warning windows before thermal runaway.
4.2 Thermal Runaway Characteristics Under Continuous Overcharging
To establish the experimental basis for defining micro-overcharge states, a continuous overcharging test was conducted. The battery was charged at 50 A continuously until the voltage reached the safety limit of 10 V. The overcharging process could be divided into three stages:
| Stage | Time Period | Characteristics |
|---|---|---|
| Stage I | 0–500 s | Voltage rises linearly from 3.4 V to 4.8 V, then stabilizes; strain and temperature rise slowly |
| Stage II | 500–1200 s | Strain rate reaches peak of 10 µε/s at ~108% SOC; expansion force begins to rise linearly |
| Stage III | After safety valve opens | Voltage rises exponentially to 10 V cut-off; no full thermal runaway triggered |
The continuous overcharging experiment showed that the battery safety valve opened at approximately 116.6% SOC. The strain peak during overcharging (4258.2 µε) was significantly lower than that during thermal abuse, indicating different internal reaction mechanisms. The expansion force peak reached 12416.6 N. Since the safety valve opened but full thermal runaway was not triggered under continuous overcharging, micro-overcharge states of 102%, 104%, 106%, 108%, 110%, and 112% SOC were selected for subsequent thermal runaway experiments.
4.3 Strain Characteristics During Micro-Overcharging
Fig. 4-8 in the original thesis shows the characteristic parameter variations during micro-overcharging to different SOC levels. The key finding was that when the overcharge level did not exceed 108% SOC, the strain exhibited a smooth linear increase with a stable rate of approximately 2 µε/s, and the peak strain remained below 1500 µε. However, when the overcharge level exceeded 108% SOC, the strain rise rate exhibited a turning point, and the expansion force began to rise significantly.
| Overcharge Range (SOC) | Δε (µε) | ΔF (N) | ΔT (℃) | ΔU (V) |
|---|---|---|---|---|
| 100%~102% | 97.8 | 9.8 | 2 | 1.33 |
| 100%~104% | 826.9 | 49 | 5 | 1.43 |
| 100%~106% | 1228.8 | 98 | 9 | 1.69 |
| 100%~108% | 1489.6 | 196 | 12 | 2.01 |
| 100%~110% | 3353.9 | 1960 | 19 | 2.09 |
| 100%~112% | 4762.3 | 3371.2 | 24 | 2.04 |
These results demonstrate that strain is highly sensitive to micro-overcharge levels, while temperature and voltage changes remain relatively moderate. The 108% SOC threshold represents a critical point beyond which the internal reactions intensify dramatically, making the energy storage battery significantly more hazardous.
4.4 Thermal Runaway Characteristics After Micro-Overcharging
Following the micro-overcharging procedure, the batteries were immediately subjected to external heating to study their thermal runaway characteristics. The results showed significant differences among batteries with varying micro-overcharge levels:
Although the micro-overcharge levels differed by only 2% SOC increments, the thermal runaway behaviors exhibited substantial variations. Batteries with higher overcharge levels had significantly earlier safety valve opening times. For example, the 110% SOC battery reached safety valve opening at 658 s, whereas the 102% SOC battery reached it at 1100 s. However, the thermal runaway trigger time for the 112% SOC battery (1120 s) was unexpectedly longer than that of the 106% SOC (1007 s) and 110% SOC (764 s) batteries, possibly due to different internal reaction pathways influenced by the excessive initial gas accumulation.
The strain rise rate during thermal runaway of micro-overcharged batteries displayed two distinctive peaks. The first weak peak occurred immediately after heating began, caused by the thermal expansion of the casing materials. The intensity of this weak peak increased with the overcharge level, reaching 30 µε/s for the 112% SOC battery. Below 108% SOC, the second strong peak of the strain rise rate occurred after a period of decline and re-ascent, while above 108% SOC, the turning point appeared at 20 µε/s, indicating a more dangerous thermal runaway process.
| SOC (%) | tabn (s) | tεrp (s) | tεp (s) | tV (s) | tTR (s) | ∆t1 (s) | ∆t2 (s) | ∆t3 (s) |
|---|---|---|---|---|---|---|---|---|
| 102% | 648 | 880 | 886 | 1100 | 1180 | 452 | 220 | 214 |
| 104% | 556 | 784 | 814 | 1026 | 1115 | 470 | 242 | 212 |
| 106% | 425 | 630 | 678 | 893 | 1007 | 468 | 263 | 215 |
| 108% | 32 | 478 | 598 | 821 | 1116 | 789 | 343 | 223 |
| 110% | 189 | 496 | 564 | 658 | 764 | 469 | 162 | 94 |
| 112% | 10 | 369 | 519 | 519 | 1120 | 509 | 150 | 0 |
For batteries with SOC levels between 102% and 106%, the strain-based early warning nodes provided average advance warnings of approximately 460 s (\(t_{abn}\)), 250 s (\(t_{εrp}\)), and 213 s (\(t_{εp}\)) before safety valve opening. However, when the SOC exceeded 108%, the time intervals became more volatile, with the \(t_{εp}\) of the 112% SOC battery coinciding exactly with the safety valve opening time, indicating an extremely dangerous condition.
5. Ultra-Early Warning Model for Thermal Runaway
5.1 Comparing Normal Operation and Thermal Runaway Strain Characteristics
The comparative analysis of strain behavior between normal charge-discharge conditions and thermal runaway conditions reveals fundamental differences that form the basis of the early warning strategy:
| Characteristic | Normal Charge-Discharge | Thermal Runaway |
|---|---|---|
| Strain Peak | < 1000 µε | 6000–16000 µε |
| Strain Rate | 0.3–0.5 µε/s | > 10 µε/s |
| Evolution Pattern | Cyclic, reversible, predictable | Abrupt, monotonic, irreversible |
| Underlying Mechanism | Li-ion intercalation/deintercalation, thermal expansion | Gas generation, internal pressure accumulation |
| Variability | Stable across cycles (±0.6 µε/s) | Exponential increase |
5.2 Three-Level Early Warning Strategy
Based on the extensive experimental data and analysis, a three-level early warning strategy for thermal runaway was developed, applicable to batteries with SOC ranging from 0% to 108%. The warning model is summarized as follows:
| Warning Level | Judgment Criterion | Threshold |
|---|---|---|
| Level I | Strain rise rate (\(dε/dt\)) | 10 µε/s |
| Level II | Peak strain rise rate | \(ε_{rps}\) |
| Level III | Strain peak | \(ε_{ps}\) |
Level I Warning: The strain rise rate in normal charge-discharge environments is consistently below 1 µε/s. In contrast, under thermal abuse conditions, the strain rise rate exhibits a turning point at 10 µε/s across different SOC levels of the energy storage battery. This threshold marks the onset of significant internal gas generation dominating the strain response, corresponding to the widespread decomposition of the SEI film within the battery.
Level II Warning: When the strain rise rate of the battery reaches its peak, a Level II warning is triggered. To ensure reliable triggering across different SOC levels, a 10% safety margin was applied to the peak values. The relationship between SOC and the adjusted strain rise rate peak (\(ε_{rps}\)) was established through empirical equations.
For SOC range 0%–100%:
$$\varepsilon_{rps} \approx 19.13 + 0.64 \cdot SOC – 0.01 \cdot SOC^2 + 5.53 \times 10^2 \cdot SOC^3$$
For SOC range 100%–108%:
$$\varepsilon_{rps} = 174095.712 – 5062.92696 \cdot SOC + 49.05964 \cdot SOC^2 – 0.15844 \cdot SOC^3$$
Level III Warning: The experimental results consistently demonstrated that the battery strain reaches its peak before the safety valve opens. The Level III warning threshold (\(ε_{ps}\)) was established based on the strain peak with a 10% safety margin to account for manufacturing tolerances and sensor measurement uncertainties. The thresholds were calibrated at 20% SOC intervals.
| SOC (%) | 20 | 40 | 60 | 80 | 100 | 102 | 104 | 106 | 108 |
|---|---|---|---|---|---|---|---|---|---|
| εp (µε) | 6278.2 | 7633.7 | 8001.8 | 10818.4 | 8137.6 | 8602.7 | 11461.3 | 11136 | 15730.6 |
| εps (µε) | 5650.38 | 6870.33 | 7201.62 | 9736.56 | 7323.84 | 7742.43 | 10315.17 | 10022.4 | 14157.54 |
For SOC levels between the calibrated intervals (e.g., 30%, 50%), the thresholds from the nearest lower calibrated SOC level are used to ensure conservative warning decisions, assuming similar strain characteristics of the energy storage battery at proximate SOC levels.
5.3 Warning Performance Analysis
The evaluation of the early warning model indicates that:
- Level I warning is activated on average approximately 450 s before safety valve opening for various SOC levels of the energy storage battery.
- Level II warning provides early warning approximately 200 s before safety valve opening.
- Level III warning could provide early warning approximately 150 s before safety valve opening for SOC levels below 108%.
The strain-based warning strategy demonstrates superior early detection capability compared to traditional temperature-based and voltage-based methods. By the time the safety valve opens, significant fire risk has already developed, whereas the strain-based method provides critical lead time for early intervention and emergency response measures, thereby substantially enhancing the safety of energy storage systems.
6. Conclusions
This thesis systematically investigated the thermal runaway strain characteristics of LFP energy storage batteries, leading to the development of an ultra-early warning model. The main conclusions are as follows:
(1) The strain generation mechanism of the battery casing is primarily driven by internal gas pressure, which consists of gas generation from side reactions, electrolyte evaporation, and gas thermal expansion. Theoretical analysis revealed that the strain increases monotonically with internal pressure during the continuous loading state of thermal runaway. SEM characterization confirmed that the casing deformation involves stretching, displacement, and eventual material failure. Both theoretical and experimental results confirmed that the center of the battery side is the strain-sensitive region, exhibiting peak strain values 20–30% higher than the edges.
(2) The typical thermal runaway process of the energy storage battery can be divided into five characteristic stages based on strain evolution: slow strain increase in the initial stage, abrupt strain increase at 10 µε/s in the acceleration stage, strain peaking before safety valve opening, strain decrease after valve opening, and secondary strain rise during thermal runaway eruption. The strain behavior shows excellent correlation with internal gas generation and expansion force changes.
(3) Under normal charge-discharge conditions, the strain of the energy storage battery exhibits smooth periodic fluctuations with peak values consistently below 1000 µε and strain rates remaining within ±0.6 µε/s across cycles, demonstrating high repeatability and stability. These values provide a reliable reference for determining warning thresholds.
(4) The SOC significantly influences the thermal runaway characteristics of the energy storage battery. Batteries with low SOC levels (0%, 20%, 40%) do not undergo full thermal runaway, while high SOC batteries (60%, 80%, 100%) do. For all SOC levels, the strain rise rate and strain peak occur before safety valve opening, making them key indicators for early warning. The continuous overcharging experiments revealed three characteristic stages, with the 108% SOC level identified as the critical threshold beyond which the battery exhibits significantly higher hazard levels.
(5) The three-level early warning strategy based on dynamic strain signals was established, using the strain rise rate threshold of 10 µε/s, the strain rise rate peak, and the strain peak as the warning criteria. This strategy enables early warning on average 450 s before safety valve opening across the full charge-discharge cycle of the energy storage battery, providing sufficient response time for early intervention and emergency measures in thermal runaway disaster prevention.
The innovation of this research lies in the theoretical analysis of the strain generation mechanism, the systematic investigation of strain characteristics under both normal and thermal runaway conditions, the elucidation of the internal coupling relationships between strain and other characteristic parameters under different SOC and micro-overcharge conditions, and the construction of a three-level early warning model based on dynamic strain signals, which effectively improves the timeliness of thermal runaway early warning for energy storage batteries.
Future research directions include investigating thermal runaway characteristics under constant voltage overcharging conditions, incorporating additional key influencing factors such as battery aging, ambient temperature, trigger methods, and electrode materials into the early warning model, and validating the applicability of the model to batteries from different manufacturers with various capacities.