The widespread adoption of lithium-ion technology in energy storage systems necessitates robust and intelligent management solutions. The inherent characteristics of an energy storage cell—such as its sensitivity to over-charge, over-discharge, excessive current, and temperature extremes—demand continuous monitoring and control to ensure safety, longevity, and optimal performance. A Battery Management System (BMS) is therefore an indispensable component, acting as the central nervous system for the battery. My work has focused on designing a comprehensive BMS architecture that addresses the limitations of conventional systems, particularly in terms of communication efficiency, granular management capabilities, and ease of deployment and maintenance. The proposed system is built around a three-tier architecture with a corresponding three-level network, utilizing the LTC6813 as the core Analog Front-End (AFE) for precise cell parameter acquisition.
The foundational design philosophy centers on a clear hierarchical structure. The physical energy storage cell organization is mirrored by the management hardware. The first tier consists of individual battery modules and their dedicated Module Management Units (MMUs). Multiple series-connected modules form a battery cluster, which is managed by the second-tier Cluster Management Unit (CMU). Finally, several parallel-connected clusters constitute a battery stack, overseen by the third-tier Stack Management Unit (SMU). This logical partitioning allows for localized, fine-grained control at the module level while enabling coordinated management at the cluster and stack levels.
To facilitate efficient data exchange across this hierarchy, a dedicated three-level communication network was implemented. The first-level network connects MMUs to their parent CMU using a daisy-chained topology via RJ45 interfaces, simplifying wiring within a cluster. The second-level network interconnects all CMUs with the SMU. Crucially, this connection employs a dual Ethernet design (Switch A and B) rather than the conventional CAN bus. This architectural choice enables advanced functionalities like cluster-level fault waveform recording, which is essential for precise post-event analysis—a capability often missing in standard systems. The third-level network connects the SMU to external systems like the Energy Management System (EMS) and the Power Conversion System (PCS). The communication with the PCS uses high-speed GOOSE messaging over direct Ethernet links, reducing response times from seconds to the order of 10 milliseconds. This entire network framework, predominantly using standard RJ45 connectors, supports a “plug-and-play” approach, significantly easing installation and replacement procedures without the need for manual address configuration.
The hardware realization of this architecture involves three distinct device types. The Module Management Unit (MMU) is the most numerous component, attached to each battery module. Its core function is to interface directly with the energy storage cells. Based on the LTC6813 AFE chip, each MMU is configured to monitor the voltage of 1 to 18 series-connected cells and supports passive balancing for each channel. It also acquires temperature data from up to 16 NTC thermistors placed within the module. A significant design simplification is that the MMU operates without an onboard microcontroller (MCU). Its configuration and position in the daisy-chain are automatically recognized, allowing for replacement units to be installed without any software downloads or address switches, truly achieving “install-and-forget” operation.
The Cluster Management Unit (CMU) acts as the cluster controller. It aggregates data from all MMUs in its cluster, estimates State of Charge (SOC) and State of Health (SOH) for both individual cells and the entire cluster, and monitors critical cluster-level parameters. These include total cluster voltage and current, as well as the insulation resistance of the positive and negative DC buses. The CMU is physically housed within a High-Voltage Control Box (HVCB), which it directly controls. The HVCB contains the main contactors, pre-charge circuit, fuses, and maintenance switches that physically connect the battery cluster to the common DC bus. The CMU’s logic manages the sequential control of this power path, including orderly connection/disconnection, pre-charging, and in-rush current limitation.
The Stack Management Unit (SMU), along with its accompanying Local Display Unit (LDU), resides at the top of the hierarchy. The SMU is responsible for coordinating all clusters within the stack, communicating with external systems (EMS/PCS), logging historical data and events, and presenting a unified system view. The LDU provides a local human-machine interface (HMI) for operators to view real-time data, check system status, and, with proper authentication, issue control commands like connect/disconnect to individual clusters. This local interface enhances operational convenience for on-site maintenance personnel.
The efficacy of any BMS hinges on the accuracy and reliability of its core measurement functions. For temperature sensing, a high-precision model was developed. The relationship between the resistance of the NTC thermistor ($R_T$) and temperature ($T$ in Kelvin) is traditionally expressed with an exponential model involving the B-constant. However, to achieve higher accuracy across a wide temperature range, a third-order exponential model was fitted to the sensor’s characteristic curve:
$$ R_T = R_{25} \cdot e^{(A + \frac{B}{T} + \frac{C}{T^2} + \frac{D}{T^3})} $$
For the specific 10 kΩ NTC used ($R_{25}=10,000 \Omega$), the fitted constants are $A = -12.89228$, $B = 4245.148$, $C = -87493$, and $D = -9588114$. The measurement circuit uses the LTC6813’s GPIO pin to measure the voltage ($V_{GPIO}$) across a voltage divider formed by a precision reference resistor $R$ (also 10 kΩ) and the NTC. The governing equation is:
$$ R_T = \frac{R \cdot V_{GPIO}}{V_{REF} – V_{GPIO}} $$
where $V_{REF}$ is the 3V reference. Combining these equations allows for the calculation of temperature from the ADC reading. Converting to Celsius ($t$):
$$ t = T – 273.15 $$
This model was validated against a calibrated thermal chamber. The results from four sample channels demonstrate the accuracy of the approach, with all measured values well within ±1°C of the setpoint across a range from -40°C to 70°C.
| Test Temperature (°C) | NTC1 (°C) | NTC2 (°C) | NTC3 (°C) | NTC4 (°C) |
|---|---|---|---|---|
| -40 | -38.2 | -38.3 | -38.1 | -39.0 |
| -25 | -24.5 | -24.4 | -24.2 | -24.4 |
| -20 | -19.4 | -20.1 | -20.0 | -20.5 |
| 0 | -0.2 | -0.2 | 0.1 | -0.1 |
| 20 | 20.1 | 19.4 | 19.2 | 19.2 |
| 25 | 24.7 | 24.6 | 24.9 | 24.8 |
| 40 | 40.1 | 39.6 | 39.4 | 39.4 |
| 45 | 44.4 | 44.3 | 44.6 | 44.5 |
| 60 | 59.7 | 59.6 | 59.4 | 59.5 |
| 70 | 69.7 | 69.8 | 69.7 | 69.7 |
Insulation monitoring is another critical safety feature. Conventional methods like balanced or unbalanced bridge techniques have limitations, such as the inability to detect symmetrical insulation degradation or the introduction of permanent imbalance. The designed system employs a novel “balanced-unbalanced bridge” method. The key innovation is the use of switched resistor networks ($R_{p1}$, $R_{p2}$, $R_{n1}$, $R_{n2}$) and a connection switch ($K_3$) to a functional ground (FGND). By taking voltage measurements under different switch states (e.g., with $K_3$ closed, and then with $K_1$ & $K_3$ closed), the positive ($R_p$) and negative ($R_n$) insulation resistances to FGND can be calculated independently and accurately, even when both degrade simultaneously. After measurement, switch $K_3$ is opened, disconnecting the measurement circuit from FGND and thus not permanently affecting the system’s insulation. The derived formulas for the resistance calculations are:
$$ R_{p^-} = \frac{U_s (U_n – U_{n1}) R_{p1} (R_{p1}+R_{p2})}{U_s U_{n1} R_{p1} + U_n U_{n1} R_{p2} – U_s U_n (R_{p1}+R_{p2})} $$
$$ R_{n^-} = \frac{U_s (U_p – U_{p1}) R_{p1} (R_{p1}+R_{p2})}{U_s U_{p1} R_{p1} + U_p U_{p1} R_{p2} – U_s U_p (R_{p1}+R_{p2})} $$
where $U_s$ is the stack voltage, $U_n$ and $U_p$ are the negative and positive pole voltages relative to FGND with only $K_3$ closed, and $U_{n1}$ and $U_{p1}$ are the corresponding voltages with both $K_1$ and $K_3$ closed. Testing with simulated insulation faults demonstrated the high precision of this method.
| Insulation Simulation | Rp Test Value (kΩ) | Rp Error (%) | Rn Test Value (kΩ) | Rn Error (%) |
|---|---|---|---|---|
| Rn=40 kΩ, Rp=1300 kΩ | 1307.64 | 0.59 | 39.52 | -1.20 |
| Rn=200 kΩ, Rp=1300 kΩ | 1329.33 | 2.26 | 202.11 | 1.06 |
| Rn=400 kΩ, Rp=1300 kΩ | 1298.52 | -0.11 | 398.23 | -0.44 |
| Rp=40 kΩ, Rn=1300 kΩ | 39.91 | -0.22 | 1309.50 | 0.73 |
| Rp=1300 kΩ, Rn=1300 kΩ | 1299.80 | -0.02 | 1300.71 | 0.05 |
The fundamental task of monitoring each energy storage cell‘s voltage is entrusted to the LTC6813 AFE in each MMU. This chip provides highly accurate (±1.2mV) and finely resolved cell voltage measurements through its multi-channel, 16-bit delta-sigma ADC with integrated noise filtering. Data is transmitted digitally via an isolated SPI (isoSPI) interface to the CMU, ensuring integrity in the noisy high-voltage environment of a battery cluster. Validation tests on a 1P12S module showed excellent agreement with a high-precision digital multimeter, with per-cell errors within 2mV and total module voltage error under 0.2%. This level of accuracy is paramount for reliable SOC estimation and effective cell balancing, which are essential for maximizing the lifespan of every energy storage cell in the system.
| Channel | Reference Value (V) | System Test Value (V) | Error (V) |
|---|---|---|---|
| V1 | 3.2965 | 3.295 | 0.0015 |
| V2 | 3.2965 | 3.295 | 0.0015 |
| V3 | 3.2960 | 3.295 | 0.0010 |
| V4 | 3.2960 | 3.295 | 0.0010 |
| V5 | 3.2970 | 3.295 | 0.0020 |
| V6 | 3.2960 | 3.295 | 0.0010 |
| V7 | 3.2960 | 3.295 | 0.0010 |
| V8 | 3.2970 | 3.295 | 0.0020 |
The complete BMS has been successfully deployed in operational energy storage projects. During on-site commissioning, full charge and discharge cycles were conducted. The system reliably recorded the voltage trajectory of every energy storage cell within a module. The data curves from a representative 12-cell module show coherent behavior across all cells during constant-current, constant-voltage charge, open-circuit rest, discharge, and final rest periods. The close alignment of the twelve voltage profiles indicates good initial cell consistency within the module, which the passive balancing function of the MMU will help maintain over the system’s operational life. The system correctly implemented protective current limiting as cell voltages approached the upper safety threshold during charging. This successful field validation confirms the system’s stability, measurement accuracy, and ability to execute complete control sequences, proving the practical viability and robustness of the designed three-tier BMS architecture for managing large-scale energy storage cell assemblies.
In summary, the designed battery management system presents a holistic solution for modern lithium-ion energy storage. Its three-tier hardware and network architecture provide clear scalability, fault localization, and maintenance advantages. The use of advanced AFE chips like the LTC6813 ensures high-fidelity data acquisition at the individual energy storage cell level. Sophisticated algorithms for temperature measurement and insulation monitoring overcome the limitations of traditional methods, delivering superior accuracy and safety. The integration of high-speed Ethernet communications enables advanced diagnostics and rapid control responses. The system’s design, from the plug-and-play MMUs to the local display interface, prioritizes not only performance but also practical deployability and operational ease. This combination of precise measurement, intelligent control, robust communication, and user-centric design makes it a capable and reliable guardian for the health and performance of large-scale battery energy storage systems.