KR-20260065769-A - IoT-based electric two-wheeler battery real-time diagnosis and anomaly prediction system

Abstract

The present invention relates to a system that combines Internet of Things (IoT) communication technology with machine learning algorithms to diagnose the condition of a battery pack mounted on an electric two-wheeled vehicle in real time and predict abnormalities and failures in advance. The system of the present invention includes a multi-channel sensing unit that collects cell voltage, pack current, temperature, and electrochemical impedance of a battery pack in real time; an edge computing diagnostic module equipped with an SOC estimation algorithm combining Coulomb counting and Kalman filtering and an Autoencoder-based anomaly detection model lightweight to 100KB or less; a BLE 5.0 and LTE-M/NB-IoT dual wireless communication module; a cloud server including a battery health prediction engine that estimates battery SOH within ±3% error and predicts remaining useful life (RUL) using an LSTM neural network-based time series prediction model; and a mobile application that provides battery status information visualization and anomaly detection push notifications. The edge computing diagnostic module independently performs primary anomaly detection while mounted on the vehicle even in the event of a communication failure, and classifies four types of anomalies—cell voltage imbalance, rapid increase in internal resistance, temperature anomalies, and overcurrent—and reports them to the cloud server. The battery health prediction engine on the cloud server uses charge-discharge cycle time series data, SOC-OCV curve derivative voltage analysis results, and impedance spectrum parameters as input features to output the estimated time to reach 80% SOH as the remaining effective life. According to the present invention, battery abnormalities can be predicted at least 2 to 4 weeks in advance to prevent fire and explosion accidents, and the battery status of multiple vehicles can be centrally monitored through a fleet management dashboard, thereby reducing preventive maintenance costs.

Inventors

• 이상호

Assignees

• 주식회사 핸디라이프

Dates

Publication Date

20260511

Application Date

20260416

Claims (1)

1. An IoT-based system for diagnosing the condition of a battery pack mounted on an electric two-wheeled vehicle in real time and predicting abnormalities comprises: a multi-channel sensing unit (100) including a cell voltage measurement channel for individually measuring the voltage of each cell of the battery pack, a Hall effect current sensor (101) for measuring the total current of the pack, NTC thermistor-based temperature sensors (102) installed at three or more locations within the battery pack, and an impedance measurement circuit (103) for measuring the internal resistance and diffusion impedance of the battery by applying AC impedance spectroscopy (EIS); and an edge computing diagnostic module (200) including an MCU of ARM Cortex-M4 or higher specification that receives data from the multi-channel sensing unit (100) and runs an SOC estimation algorithm combining Coulomb counting and a Kalman filter, and an Autoencoder-based anomaly detection model lightweighted to a model size of 100KB or less. An IoT-based electric two-wheeler battery real-time diagnosis and anomaly prediction system characterized by comprising: a dual-mode wireless communication module (300) that communicates near-field with a user smartphone using the BLE 5.0 protocol and wide-area with a cloud server (400) using the LTE-M or NB-IoT protocol; a cloud server (400) that includes a battery health prediction engine (410) that stores battery data received through the wireless communication module (300), runs an LSTM neural network-based time series prediction model to estimate battery SOH within ±3% error, and predicts the remaining effective life (RUL); and a mobile application (500) that visualizes battery SOC, SOH, RUL, real-time voltage, current, and temperature in conjunction with the cloud server (400), and sends a push notification when an anomaly is detected.

Description

IoT-based electric two-wheeler battery real-time diagnosis and anomaly prediction system The present invention relates to a system that combines Internet of Things (IoT) communication technology and data analysis algorithms to diagnose the condition of a battery pack mounted on an electric two-wheeled vehicle in real time and to predict degradation, abnormalities, and failures in advance. More specifically, the invention relates to an end-to-end battery diagnostic platform that integrates a vehicle-mounted sensing unit, an edge computing-based anomaly detection module, a wireless communication module, a cloud server-based battery life prediction engine, and a user mobile application. With the expansion of electric two-wheeled vehicles, there is an increasing demand for safety management and lifespan prediction of battery packs. The capacity and output of batteries gradually decrease due to stress factors such as electrochemical degradation from repeated charge-discharge cycles, exposure to high and low temperature environments, overcharging and over-discharging, and high-rate discharge. If battery abnormalities are discovered retrospectively, the vehicle may already be inoperable, or in the worst case, lead to a battery fire or explosion; therefore, proactive diagnosis and anomaly prediction are of great importance. Existing BMSs are reactive systems that only perform protection operations when threshold values such as overvoltage, overcurrent, and overtemperature are exceeded, and often lack the ability to quantitatively analyze battery degradation trends or predict remaining effective lifespan. Furthermore, existing battery diagnostic systems are independent systems confined to the vehicle interior and lack integration with cloud servers or remote monitoring capabilities, making it difficult for fleet operators or battery service providers to centrally manage multiple vehicles. While machine learning-based battery state of health (SOH) estimation and risk of injury (RUL) prediction technologies are garnering attention, technologies regarding lightweight AI models capable of real-time operation in resource-constrained environments on two-wheeled vehicles and edge-cloud collaboration architectures have not yet been sufficiently disclosed. FIG. 1 is a block diagram of the overall 4-layer architecture of an IoT-based battery diagnostic system according to the present invention. Embodiments of the present invention will be described in detail with reference to the attached drawings. 1. System Architecture The system of the present invention is composed of a four-layer architecture consisting of a Vehicle Layer, a Communication Layer, a Cloud Layer, and a User Layer. The vehicle layer is equipped with a multi-channel sensing unit (100), an edge computing diagnostic module (200), and a wireless communication module (300). The edge computing diagnostic module is designed to independently perform primary anomaly detection functions even in the event of a communication failure or cloud server failure. The communication layer supports dual communication of BLE and LTE-M/NB-IoT. BLE 5.0 communicates directly with the user's smartphone with a maximum communication distance of 40m and a transmission speed of 2Mbps, and is used for checking real-time data and changing settings near the vehicle. LTE-M (Cat-M1) provides a stable cloud connection even while on the move, and its low power consumption minimizes the impact on battery life. 2. Multichannel sensing unit (100) Cell voltage measurement is performed by using a battery cell monitoring IC (e.g., Texas Instruments BQ76940 or equivalent) to collect the voltage of each cell with 1mV resolution and a 100ms measurement cycle. Up to 15 series cells can be measured with a single IC, and if there are more cells, additional ICs are connected in a daisy-chain manner. The impedance measurement circuit (103) consists of a voltage-controlled current source (VCCS) and a high-precision AD converter (ADC). It performs EIS at seven major frequencies in the range of 0.1 Hz to 1 kHz, and the measurement time is approximately 60 seconds. Impedance measurement is performed automatically when the battery is stopped (after 10 minutes have passed since driving ended). 3. AI model of edge computing diagnostic module (200) The Autoencoder model used has a symmetric structure with 20 nodes in the input layer (15 for each cell voltage + 1 for current + 3 for temperature + 1 for impedance), 10→5 nodes in the encoder hidden layer, 5→10 nodes in the decoder hidden layer, and 20 nodes in the output layer. The activation function used is ReLU, and it is pre-trained with normal data for more than 3,000 cycles. The model is converted and quantized (INT8 Quantization) using TensorFlow Lite to reduce its size to 82KB and inference time to 30ms or less, and then loaded onto the MCU. 4. LSTM model of battery health prediction engine (410) The LSTM model running on a cloud server uses an input sequence le