Machine Learning for Thermal Runaway Prevention in EV Batteries

Intelligent battery management system that can detect and mitigate battery faults in real-time while the battery is in use. The system uses machine learning to identify anomalies in battery behavior, extracts data containing those anomalies, modifies the training data, and retrains the AI model to better recognize and manage the faulty battery operations. This allows the system to continuously learn and adapt to battery faults as they occur, improving fault detection accuracy compared to offline methods.

Source

Identifying root causes of premature battery aging in vehicles and recommending mitigation strategies based on confidence levels. The technique involves monitoring battery and vehicle data to identify causes like excessive discharge during parking, high temperature operation, high charge throughput, infrequent driving, and primary power source saturation. A supervised learning algorithm compares metrics to known aging categories and identifies the most similar root causes. Mitigation actions like deactivating electronics or dimming lights are taken based on confidence levels.

Source

Modular machine learning structure for estimating battery state of health (SoH) in electric vehicles. The technique involves using graph neural networks (GNNs) to accurately estimate battery SoH by capturing cell-to-cell interactions. Raw battery sensor data from the cells and modules is fed into the GNN nodes representing each cell. The node states are updated and used to estimate cell-specific SoH indicators. Aggregated node states from modules are used to estimate overall SoH for the pack. This allows more accurate SoH estimation compared to relying solely on cell-level models.

Source

Battery diagnosis method using AI to accurately ascertain the state of a battery exhibiting non-linearity by considering time series data like voltage, current, and temperature along with non-time series data like battery impedance. The battery state information is predicted by inputting both types of data into a trained battery prediction model. This allows more accurate battery state determination compared to just using basic measurements like current, voltage, and temperature.

Source

Smart battery management system (SBMS) that predicts and prevents battery failures in advance using sensors and machine learning. The SBMS monitors metrics like pressure, temperature, voltage, current, and capacitance from cells. It predicts failures using a trained neural network. If a cell failure is predicted, the SBMS disconnects the cell to prevent damage. This allows load balancing and disconnecting cells before thermal runaway or other failures occur. The SBMS can also provide visual representations of SoH, temperatures, pressures, etc. throughout a pack.

Source

Early battery failure prediction using AI to evaluate the remaining useful life (RUL) of a battery based on minimal cycling data. The method involves identifying parameters related to battery composition during charging/discharging cycles. It determines variation patterns in voltage, current, temperature, and resistance until failure. An AI model is trained on these patterns and compositions to predict RUL. This allows early detection of battery degradation and failure without needing large amounts of cycling data.

Source

Big-data-based battery diagnostic system for electric vehicles that learns from mass-produced vehicles to accurately diagnose battery cell safety in all scenarios. The system collects state data from vehicles, learns reference voltages for each vehicle, compares new data to the reference, and diagnoses cell health using a diagnostic range. It continuously learns from vehicles and updates reference voltages. This allows customized, consistent safety diagnosis based on actual field data instead of passive formulas.

Source

Machine learning forecasting of battery health to predict battery failure trajectories instead of just failure times. The forecasting is done by extracting time-based features from small segments of charging and discharging data using convolutional neural networks. These features capture temporal evolution of voltage, current, etc. A trained Gaussian process regression model then predicts future battery state-of-health based on the features. This allows forecasting a path to failure rather than just when it will happen.

Source

Determining battery degradation in-situ without invasive cell testing. The method involves monitoring battery temperature, load power, and environmental conditions during normal operation. Machine learning models are trained on historical data to correlate changes in these parameters with battery degradation. By tracking the changes in real-time, the models can predict degradation levels without needing separate cell-level tests.

Source

Predicting the state of health (SOH) of an electrical energy storage device like a battery in an electric vehicle, even during periods of inactivity, using a data-based model trained from operating variable data. The model assigns a predicted SOH to the battery based on characteristics like charge/discharge cycles, temperature, and aging factors. This allows estimating the battery's remaining capacity and life even when sensors aren't actively measuring during idle periods.

Source