In the context of modern infrastructure, the reliance on battery energy storage systems as critical backup power sources has grown exponentially. Facilitating the safety and operational integrity of these energy storage systems during emergency scenarios remains a paramount challenge. Based on my recent research and development work, I have designed a low-cost, high-reliability monitoring and protection system specifically for battery energy storage systems, which addresses the limitations of traditional approaches. This system utilizes the STM32F103RCT6 microcontroller and a multi-sensor fusion architecture that operates independently from conventional battery management systems (BMS).
The core objective of my work was to provide a robust safety net for battery energy storage systems, which are prevalent in hospitals, data centers, and telecommunication base stations. The thermal runaway issue in lithium-ion batteries within these battery energy storage systems can lead to catastrophic fires, making real-time monitoring and rapid response not just beneficial but essential. By integrating temperature, current, and smoke sensors, this system forms a comprehensive monitoring network that enhances detection accuracy and reduces false alarms. The system’s independent design ensures deployment flexibility across various configurations of battery energy storage systems without the constraints of commercial BMS protocols.
System Architecture and Hardware Design
The hardware architecture of my monitoring and protection system for battery energy storage systems is centered around the STM32F103RCT6 processor. This chip provides sufficient computational power for real-time data processing while maintaining low power consumption and cost. The system’s design philosophy was to minimize dependencies on external communication networks and to ensure fail-safe operation even in degraded modes.
I selected a combination of sensors to monitor the key parameters of the battery energy storage system. For temperature monitoring, I employed multiple DS18B20 digital temperature sensors. These sensors offer a compact footprint and high accuracy, which is crucial for detecting hot spots in the battery pack. A dedicated current sensor, the ACS712, was integrated to monitor charge and discharge currents, providing context for thermal models. Finally, the MQ-2 smoke and gas sensor was included to detect the early signs of electrolyte vapor release, a precursor to thermal runaway in many battery energy storage systems.
To illustrate the key components, the following table summarizes the primary hardware elements used in the system:
| Component | Model/Type | Function in Battery Energy Storage System |
|---|---|---|
| Microcontroller | STM32F103RCT6 | Central processing and control logic for the monitoring system |
| Temperature Sensor | DS18B20 | High-precision temperature monitoring within the battery enclosure |
| Current Sensor | ACS712 | Real-time current measurement for power and thermal estimation |
| Smoke/Gas Sensor | MQ-2 | Detection of smoke or combustible gases indicative of thermal runaway |
| Actuator | Solenoid Valve | Controls the flow of fire suppressant agent (e.g., CO2 or Novec 1230) |
| Alarm Module | Buzzer + LED | Provides audible and visual warnings during pre-alarm and alarm states |
The sensor data acquisition strategy was designed to ensure redundancy. I placed multiple temperature sensors in a strategic array around the battery modules within the battery energy storage system. This configuration allowed for spatial temperature mapping and the detection of localized heating. The following table details the typical sensor arrangement.
| Sensor Type | Quantity | Placement Location | Measurement Purpose |
|---|---|---|---|
| DS18B20 | 6 | On top of each battery module terminal | Direct cell temperature and hot spot detection |
| DS18B20 | 2 | Ambient air inlet and outlet | Ambient and exhaust air temperature for thermal balance |
| MQ-2 | 2 | Inside the battery enclosure top and bottom | Early detection of electrolyte vapor and smoke |
| ACS712 | 1 | Main power bus bar | Total current flow into/out of the battery energy storage system |
Software Architecture and Multi-Sensor Fusion Algorithm
The software running on the STM32F103RCT6 is the brain of the protection system. I developed a control strategy that involves multi-threshold detection and a hierarchical decision-making process. The raw data from the sensors are first pre-processed to filter out noise and transient spikes. I employed a moving average filter for the temperature data, which provides a smoothed value for stable decision-making. The formula for this averaged temperature is as follows:
$$ T_{avg}(n) = \frac{1}{N} \sum_{i=0}^{N-1} T_{raw}(n – i) $$
where \(T_{raw}\) is the raw temperature reading and \(N\) is the sample window size, typically set to 10 samples for a 1-second sampling interval in our battery energy storage system monitor.
The core of the algorithm is the multi-sensor fusion logic, which combines inputs from the temperature, current, and smoke sensors to assess fire risk. The system calculates a risk score based on weighted inputs. The fusion algorithm can be expressed as a composite function. The primary equation for the hazard level \(H\) is:
$$ H = w_1 \cdot f_{temp}(T_{max}) + w_2 \cdot f_{smoke}(S_{conc}) + w_3 \cdot f_{current}(I_{rate}) $$
In this equation, \(T_{max}\) is the maximum temperature reading from the sensor array, \(S_{conc}\) is the smoke concentration, and \(I_{rate}\) is the rate of current change. The functions \(f_{temp}\), \(f_{smoke}\), and \(f_{current}\) are sigmoid normalization functions that map input values to a scale of 0 to 1. The weights \(w_1, w_2, w_3\) are pre-calibrated constants that prioritize temperature and smoke signals over current for fire hazard detection in stationary battery energy storage systems.
The hierarchical control strategy is structured in three primary states: Normal, Pre-Alarm, and Fire. A summary of the state machine is provided in the table below.
| State | Trigger Condition | Action |
|---|---|---|
| Normal | All parameters within safe limits | No action; continuous monitoring |
| Pre-Alarm | Temperature > 90°C for 3s OR Smoke detection > threshold | Activate buzzer and LED alarm; wait for operator confirmation |
| Fire (Auto) | Temperature > 130°C for 3s without operator intervention | Automatically open solenoid valve for fire suppression |
| Fire (Manual) | Operator presses manual button during Pre-Alarm | Override auto delay; open solenoid valve immediately |
This logic ensures that the system avoids unnecessary activation while guaranteeing a response when a hazard is confirmed. The 3-second hold condition on temperature thresholds prevents false triggers caused by instantaneous sensor noise. The manual confirmation step in the Pre-Alarm state allows for human-in-the-loop decision-making, which is critical in scenarios where non-fire related heat sources might exist near the battery energy storage system.
Experimental Results and Validation
I conducted a series of experiments to validate the accuracy and response time of the proposed system. The first set of tests focused on the temperature measurement accuracy of the DS18B20 sensors within the battery energy storage system environment. I compared the readings from the system’s sensors against a calibrated industrial thermometer for 20 different test points at various temperatures. The results demonstrate that the monitoring error is maintained below 4% across the entire operational range, with a mean absolute error of less than 0.2°C.
The experimental data for temperature accuracy is summarized in the following table:
| Test Point | Reference Temperature (°C) | Measured Temperature (°C) | Absolute Error (°C) | Relative Error (%) |
|---|---|---|---|---|
| 1 | 48.68 | 48.60 | -0.08 | 0.16 |
| 2 | 31.52 | 31.54 | 0.02 | 0.06 |
| 3 | 29.38 | 29.43 | 0.05 | 0.17 |
| 4 | 48.95 | 49.02 | 0.07 | 0.14 |
| 5 | 48.52 | 48.46 | -0.06 | 0.12 |
| 6 | 49.63 | 49.55 | -0.08 | 0.16 |
| 7 | 41.62 | 41.52 | -0.10 | 0.24 |
| 8 | 37.98 | 37.91 | -0.07 | 0.18 |
| 9 | 37.47 | 37.52 | 0.05 | 0.13 |
| 10 | 29.87 | 29.94 | 0.07 | 0.23 |
| 11 | 33.82 | 33.91 | 0.09 | 0.27 |
| 12 | 49.56 | 49.62 | 0.06 | 0.12 |
For the control strategy simulation, I developed a digital twin of the battery energy storage system monitoring environment using MATLAB Simulink. This allowed me to simulate various fault scenarios without risking actual hardware. I tested two primary scenarios: a temperature ramp leading to thermal runaway and a sudden smoke event. The simulation results consistently validated the system’s logic.
In the temperature simulation test, I injected a signal representing a thermal runaway event. The temperature exceeded the 90°C threshold at 4.2 seconds. According to my control logic, the system held this condition for 3 seconds before triggering the Pre-Alarm at 7.2 seconds. Since the operator did not press the manual confirm button in this scenario, the temperature continued to rise. At 10.2 seconds, the temperature reached the critical threshold of 130°C. The system then verified that the high temperature persisted for 3 seconds (until 13.2 seconds) and automatically issued the fire suppression command, activating the solenoid valve. This entire sequence is described by the following timing logic:
$$ t_{alarm} = t(T > T_{pre}) + \Delta t_{hold} $$
$$ t_{fire} = t(T > T_{crit}) + \Delta t_{hold} $$
where \(T_{pre} = 90^\circ \text{C}\), \(T_{crit} = 130^\circ \text{C}\), and \(\Delta t_{hold} = 3 \text{s}\).
The smoke simulation test validated the manual override function. In this test, the smoke simulator triggered the threshold at 1.2 seconds. The system applied the 3-second confirm delay, activating the Pre-Alarm at 4.2 seconds. At this point, an operator simulated a manual intervention by pressing the fire suppression button. After a 0.2-second software debounce delay, the system responded. The operator held the button down, effectively confirming the hazard, and the controller sent the drive command to the solenoid valve at 7.4 seconds. This demonstrates the system’s flexibility in allowing both automatic and manual operation paths, which is crucial for real-world deployment in critical infrastructure battery energy storage systems.
I also compiled a summary of the key performance metrics obtained from the simulation and accuracy tests. These metrics confirm the system’s suitability for protecting battery energy storage systems.
| Performance Indicator | Measured Value | Requirement |
|---|---|---|
| Temperature Measurement Accuracy | < 4% (Relative Error) | ≤ 5% |
| Pre-Alarm Response Time (from threshold) | 3.0 s (configurable) | ≤ 5 s |
| Fire Suppression Activation Time (Auto) | 3.0 s after critical threshold | ≤ 5 s |
| Manual Button Debounce Delay | 0.2 s | ≤ 0.5 s |
| False Alarm Rate (in 100 test runs) | 0 | Low |
Conclusion and Future Directions
In conclusion, the monitoring and protection system I have designed provides a reliable, low-cost solution for the safety management of battery energy storage systems. By adopting a multi-sensor fusion architecture independent of traditional BMS, the system achieves a high level of fault tolerance and ease of integration. The experimental results confirm that the system’s temperature monitoring error is within acceptable limits (below 4%), and the control logic successfully implements a comprehensive “monitor-judge-respond” protective framework. The system can trigger alerts within 3 seconds of detecting an anomaly and automatically activate fire suppression if the condition worsens without human intervention.
Looking forward, I plan to expand on this work in two primary directions. First, I aim to incorporate advanced signal processing techniques such as Kalman filters and neural networks into the data fusion algorithm. This would allow for even more precise state-of-health estimation of the battery cells within the battery energy storage system and improve the system’s ability to predict thermal runaway events before they occur. The Kalman filter, for instance, can be modeled for state estimation of the battery’s internal temperature, which is often not directly measurable:
$$ \hat{x}_{k|k} = \hat{x}_{k|k-1} + K_k (z_k – H \hat{x}_{k|k-1}) $$
where \(\hat{x}\) is the state estimate, \(K_k\) is the Kalman gain, \(z_k\) is the measurement, and \(H\) is the observation model.
Second, I intend to expand the monitored parameter set to include voltage and internal resistance measurements of individual cells. These additional parameters will provide a more comprehensive view of the battery’s electrochemical state, enabling more accurate thermal runaway prediction models. By integrating these dimensions, the monitoring system can evolve from a reactive safety device into a proactive predictive maintenance tool for battery energy storage systems, thereby enhancing the overall reliability and safety of critical power backup infrastructure.