Abstract
The Battery Management System (BMS) is crucial for understanding battery performance under extreme conditions, such as high rate testing. This study proposes a novel BMS for continuous monitoring, transmission, and storage of key parameters such as voltage, current, and temperature of NCA 18650 4S lithium-ion battery packs during high rate testing. This BMS combines deep learning technology to predict the health status of the battery (measured by discharge capacity) by monitoring external battery parameters. Two experiments were conducted: static experiment to verify the BMS function, and actual working condition experiment (high magnification abuse test with vibration on electric propulsion vehicles) to evaluate its actual performance. The results showed that the peak surface temperature of the battery during actual flight conditions reached 55 ° C, which was higher than the static test; The deep learning capacity estimation algorithm detected an average capacity deviation of 0.04 Ah, demonstrating accurate health status by predicting battery capacity. This BMS demonstrates effective data collection and prediction capabilities, reflecting the actual situation in abuse testing.
1. Introduction
The importance of lithium-ion batteries (LIBs) and their related technologies: LIBs are crucial in the current technological field and widely used in electric vehicles, drones, and portable electronic devices. Compared with traditional battery technology, LIBs have advantages such as high energy density and long cycle life, but their widespread application also brings challenges of battery aging. Therefore, State of Health (SOH) is an important parameter for measuring battery aging. Accurately estimating SOH faces many challenges, and battery management systems (BMSs) are crucial for accurately monitoring battery parameters.
Design and related indicators of BMS: The design of BMS is usually related to specific applications, and in addition to SOH, State of Charge (SOC) and Remaining Useful Life (RUL) are also common indicators of battery health. The data for these indicators usually comes from proprietary data acquisition (DAQ) settings, which can provide data for deep learning (DL), but have limitations such as large size, high cost, and targeting specific batteries. Integrating DL technology with advanced BMS is a promising approach that can overcome the challenges of data collection methods and provide scalability.
Important parameters and related research requirements of LIBs: SOH, SOC, RUL, and C-rate are all important parameters for LIBs performance. An increase in C-rate will lead to a decrease in battery capacity and performance. The current data collection methods face challenges such as complex settings and unclear sensor accuracy, thus requiring the development of innovative portable BMS frameworks for data collection in various application scenarios such as high magnification. This study proposes a BMS comprehensive framework that integrates the previously developed Capacity Degradation Network (CD Net) DL model, which can meet the real-time monitoring requirements of electric propulsion. Its data collection and model integration capabilities have been verified through experiments.
2. Development of BMS
Overview of BMS development: A printed circuit board was developed for the proposed BMS to collect key data required for battery health prediction and take action based on the prediction. The developed BMS uses actual operating condition data (including battery voltage, current, and temperature) for battery health prediction, and the collected data is input into a deep learning (DL) model for real-time prediction.
Sensor measurement
Current and voltage measurement: The high side current sensor INA219 is used to measure the current by inserting a shunt resistor. To adapt to high current scenarios, the default 0.1 Ω shunt resistor is replaced with a 0.01 Ω resistor, allowing the current measurement range to reach ± 32A.
Temperature measurement: PT100 Adafruit MAX31865 sensor is selected to measure temperature, which has low power consumption, high accuracy, and stability. The Callendar van Dusen equation is used to obtain the relationship between temperature and resistance. Five sensors are used to measure the surface temperature and ambient temperature of four batteries respectively, and calibration is carried out.
Device control: Choose the Arduino Uno Rev 2 Wi Fi circuit board as the controller, which has data recording, preprocessing, and transmission capabilities. It provides power to the sensor network through its internal 5V regulator and connects the sensor and microcontroller using SPI protocol.
| Type | Connection | Use |
| Power and Serial | USB | 5 V power supply as well as serial communication to the main CPU |
| Ground | GND | Common star ground for every component in the sensor network |
| Shunt V+ | Current Sensor V+ | Positive Kelvin connection from current shunt to battery positive |
| Shunt V- | Current Sensor V- | Negative Kelvin connection from current shunt to load positive |
| Arduino Pins | ||
| SCLK | Arduino Pin 13 | Clock line for SPI |
| SDO | Arduino Pin 12 | Serial Data Output for SPI |
| SDI | Arduino Pin 11 | Serial Data Input for SPI |
| CS1 | Arduino Pin 10 | Chip Select Temperature Sensor 1 |
| CS2 | Arduino Pin 9 | Chip Select Temperature Sensor 2 |
| CS3 | Arduino Pin 8 | Chip Select Temperature Sensor 3 |
| CS4 | Arduino Pin 7 | Chip Select Temperature Sensor 4 |
| CS5 | Arduino Pin 6 | Chip Select Temperature Sensor 5 |
| SCLK | Arduino Pin SCLK | Serial Data Clock for I2C |
| SDO | Arduino Pin SDO | Serial Data Address for I2C |
| Purpose | Sensors Used | Operating Voltage | Maximum Supply Current |
| Pack voltage and current sensor | Adafruit INA219 | 3.0 - 5.5 V | 1 mA |
| Battery surface temperature sensors | Adafruit PT100 MAX31865 | 3.0 - 3.6 V | 3 mA |
| Ambient temperature sensor | Adafruit PT100 MAX31865 | 3.0 - 3.6 V | 3 mA |
SOH estimation: The voltage and temperature data collected by BMS are transmitted to the computer, and the current data is used to calculate the state of charge (SOC) of the battery through Coulomb counting method. SOC, along with the nominal capacity and chemical composition of the battery, is input into the CD Net model to predict the discharge capacity of the battery and calculate SOH. The CD Net model uses a combination of neural networks with specific structures, which are optimized and processed for prediction.
| Characteristic | Value |
| Cell chemistry | NCA |
| Cell form factor | 18650 |
| Nominal capacity | 3120 mAh |
| Nominal voltage | 3.6 V |
| Standard charge | CCCV, 1 C, 4.2 V |
| Standard discharge | Constant charge, 1 C, 2.5 V |
| Weight | 46.4 ± 1.5 g |
3. Experimental setup
Overview of experimental setup: A battery pack consisting of four 18650 Sony VTC 6 batteries connected in series was tested using NCA chemical system. The relevant specifications of individual batteries and battery packs were introduced.
Ground testing: The purpose is to analyze the performance of the newly developed BMS before applying it to electric propulsion vehicles. Use the NEWARE Powerwall CT-4004-20V20A system as the load, connect the BMS to the battery pack and load, monitor the voltage, current, and surface temperature of each battery pack, and record the data with the BMS. Place an RTD sensor at the center of the battery and record the direction of the battery pack to ensure consistency in temperature recording. Perform 42 charge and discharge cycles, with a cycle pattern similar to NASA's certification experiment for small satellites.
Aerial testing: Conducted after ground testing, using an electric aircraft (FLYWOO Explorer drone) for 20 charge discharge cycles to collect data under high rate discharge conditions. Introduction to the relevant parameters of the aircraft, the battery pack is installed on a 3D printed bracket, the BMS is on the top, the charging protocol is consistent with ground testing, but the discharge cycle is random. The aircraft is about 1 foot above the ground during discharge. When the BMS displays that the battery voltage reaches 10V, the discharge stops and the vehicle rests for 0.167 hours before charging.
4. Results and Discussion
Summary of Results Discussion: Introduces the results and key findings obtained using the newly developed BMS (including CD Net model prediction). The newly developed BMS collects data from static and dynamic battery loads and seamlessly integrates with DL models such as CD Net, providing flexibility to adapt to technological developments.
Ground test results
Voltage curve measurement: The voltage time data of 42 charge and discharge cycles of BMS and testing equipment (BAn) are similar. Although the BMS data has initial lag, it eventually converges, with an average deviation of 0.2V between the two. The constant voltage part of the discharge curve can be used to study the relationship with SOH, and the deviation is due to differences in data transmission rate and internal clock.
Current curve measurement: The current data of BMS and BAn are matched as a whole, and the current changes during the discharge and charging stages follow the rules. However, there is a delay in BMS reading data during the current conversion stage, resulting in some deviations. After removing large deviation data points, the average deviation is smaller, and the current reading is more accurate than the voltage reading.
Temperature curve measurement: Monitor the surface temperature of four batteries and find that the temperature gradually increases during the charging and discharging process, reaching its highest point at the end of constant current charging. Then, the temperature changes during constant voltage charging and discharging. The fourth battery in the battery pack has a relatively high temperature, and the BMS can detect and display the temperature differences of each battery. Temperature anomalies can be used for battery management.
Aerial test results
Measurement of current and voltage curves: During the flight of the drone, the current randomly changes, and the BMS can accurately record the high rate discharge current. The discharge current increases with the number of tests, and the voltage decreases from 16.8V to 10V during the discharge process. The current and voltage are affected by the adjustment of the flight attitude during the flight.
Temperature curve measurement: During aerial testing, the surface temperature of the battery is higher than that during ground testing, with a maximum temperature of about 55 ° C. During the charging process, the temperature tends to decrease, and during the discharging process, the temperature gradually increases. There are differences in temperature among different batteries, and temperature fluctuations are affected by flight attitude adjustments.
Deep learning ensemble results: In ground testing, the battery capacity gradually decreased, and the CD Net model predicted the capacity from the 5th cycle, which was similar to the Coulomb count capacity recorded by BMS. The model's prediction was relatively accurate; The Coulomb counting capacity was unstable during aerial testing, but the model was still able to predict with an average difference of 0.046Ah. By comparing the capacity measured by Coulomb counting method with the predicted capacity of the model, the successful integration of BMS and DL models was verified, which can be used to predict the State of Health (SOH) of batteries.
5. Summary
Summary of research findings: The actual working condition data collection of electric propulsion vehicles requires a portable BMS that can operate under harsh conditions, such as high rate discharge of LIBs. This research proposes a novel BMS architecture, which uses edge and cloud frameworks to record, transmit and receive data, and can handle high rate discharge, replacing the traditional methods based on CAN bus and edge computers.
Summary of test results: Ground and air tests were conducted, and the surface temperature of the battery was highest at the end of the discharge constant current stage. The surface temperature of the battery was even higher in the air test, reaching a maximum of nearly 55 ° C. The temperature increase may be caused by factors such as battery aging and SOC changes, and excessive temperature may lead to battery failure. During 42 cycles on the ground and 20 cycles in the air, the surface temperature gradually increased.
Model prediction results: Using the collected data, the CD Net model predicts the State of Health (SOH) of the battery under actual operating conditions. The model's capacity prediction for the next cycle in ground testing is relatively accurate, with an average deviation of -0.026 Ah; Although there were capacity fluctuations during aerial testing, the predicted average difference was 0.046 Ah, and BMS was able to effectively collect data within the threshold range of current voltage sensors.