Abstract: In one aspect, based on at least a received first state of health of a battery pack and an initial state of charge of the battery pack, a method may include determining, by a state of charge estimator of a digital twin battery model, states of charges for the battery pack. Based on the states of charges for the battery pack, the method may include determining, by a voltage predictor of the digital twin battery model, predicted battery voltages. Based on the predicted battery voltages, the method may include determining, by a state of charge corrector of the digital twin battery model, a voltage difference between the predicted battery voltages and measured voltages. Based on the voltage difference, the method may include correcting, by the state of charge corrector, the states of charges to generate corrected states of charges for the battery pack.

Abstract: In one aspect, based on at least a received first state of health of a battery pack and an initial state of charge of the battery pack, a method may include determining, by a state of charge estimator of a digital twin battery model, states of charges for the battery pack. Based on the states of charges for the battery pack, the method may include determining, by a voltage predictor of the digital twin battery model, predicted battery voltages. Based on the predicted battery voltages, the method may include determining, by a state of charge corrector of the digital twin battery model, a voltage difference between the predicted battery voltages and measured voltages. Based on the voltage difference, the method may include correcting, by the state of charge corrector, the states of charges to generate corrected states of charges for the battery pack.

Abstract: In one aspect, while a battery pack is charging, a computer-implemented method may include predicting an anode potential of the battery pack, and determining whether the anode potential satisfies a threshold condition. Responsive to determining that the anode potential satisfies the threshold condition, the method may include modifying a charging policy of a battery pack to adjust an anode potential offset, and controlling, based on the charging policy, charging of the battery pack to adjust the anode potential offset.

Abstract: In one aspect, computer-implemented method may include receiving, from one or more sensors associated with a battery pack, one or more measurements pertaining to voltage, temperature, or both. The method may include determining, based on the one or more measurements, a voltage score and a temperature score, and predicting, based on the voltage score and the temperature score, whether the battery pack is experiencing a fault condition. The prediction is performed by an artificial intelligence engine. Responsive to predicting the battery pack is experiencing the fault condition, the method may include performing one or more preventative actions.

Abstract: In one aspect, computer-implemented method may include, while a battery pack is charging, receiving, from sensors, measurements associated with the battery pack. The battery pack includes cells. The method may include separating the measurements into separate profiles for the cells, wherein the separate profiles include data pertaining to current, voltage, temperature, or some combination thereof. The method may include identifying, using the separate profiles, features, generating a training dataset by reducing the features based on a mean-comparison technique, a minority scaling technique, or both, and generating a trained machine learning model using the training dataset including the reduced features as labeled input and true lithium plating occurrence statuses as labeled output. The method may include predicting, using the trained machine learning model, an occurrence of lithium plating by inputting subsequently received data into the trained machine learning model.

Abstract: In one aspect, computer-implemented method may include receiving, from one or more sensors associated with a battery pack, one or more measurements pertaining to voltage, temperature, or both. The method may include transforming the one or more measurements into a time-series sequential window format, determining, based on the time-series sequential window format of the one or more measurements, a voltage score and a temperature score, and predicting, based on the voltage score and the temperature score, whether the battery pack is experiencing a fault condition. The prediction is performed by one or more trained machine learning models. Responsive to predicting the battery pack is experiencing the fault condition, the method may include performing one or more preventative actions.

Abstract: In one aspect, computer-implemented method may include receiving, at a cloud-based computing system, a set of measurements over a certain period of time. The set of measurements are received from a set of vehicles and pertain to voltages and temperatures of a set of battery packs associated with the set of vehicles. The method may include training, using the set of measurements, one or more machine learning models to predict a battery pack fault condition. The one or more machine learning models include a set of parameters that are modified during the training. The method may include transmitting the set of parameters to the set of vehicles to enable the set of vehicles to update, based on the set of parameters, one or more respective in-vehicle machine learning models configured to predict the battery pack fault condition.

Abstract: In one aspect, a method may include generating an aging battery dataset including information pertaining to age histories of electric vehicles, cell manufacturer specifications, and laboratory tests. Based on the information and battery health metrics associated with battery pack modules, the method may include predicting a first state of health of the battery pack modules. The method may include receiving a second state of health of the battery pack modules, wherein the second state of health is determined based on determined states of charges for the battery pack modules. The method may include determining an uncertainty for the second state of health using a rest duration after and before charging of the battery pack modules. Based on the uncertainty, the first state of health, and the second state of health, the method may include determining a confidence score for the second state of health of the battery pack modules.

Abstract: In one aspect, computer-implemented method may include, while a battery pack is charging, receiving, from sensors, measurements associated with the battery pack. The battery pack includes cells. The method may include separating the measurements into separate profiles for the cells, wherein the separate profiles include data pertaining to current, voltage, temperature, or some combination thereof. The method may include identifying, using the separate profiles, features, generating a training dataset by reducing the features based on a mean-comparison technique, a minority scaling technique, or both, and generating a trained machine learning model using the training dataset including the reduced features as labeled input and true lithium plating occurrence statuses as labeled output. The method may include predicting, using the trained machine learning model, an occurrence of lithium plating by inputting subsequently received data into the trained machine learning model.

Abstract: In one aspect, a computer-implemented method may include receiving charge cycle data pertaining to a battery pack. The method may include determining, based on the charge cycle data, whether a noise level of a battery management system exceeds a first threshold. In response to determining the noise level exceeds the first threshold, the method may include determining an initial state of charge of the battery pack using coulomb counting by reversing the charge cycle data. In response to determining the noise level does not exceed the first threshold, the method may include determining whether a rest time before charge cycle exceeds a second threshold. In response to determining the rest time before charge cycle does not exceed the second threshold, the method may include determining the initial state of charge of the battery pack using coulomb counting by reversing the charge cycle data.

Abstract: In one aspect, computer-implemented method may include receiving, from computing devices, fleet data pertaining to battery packs each including first cells. The fleet data includes false positive images of lithium plating affecting at least a first cell, true positive images of the lithium plating affecting at least a second cell, or both. The method may include training, using at least the fleet data, machine learning models to predict occurrences of the lithium plating, receiving, from sensors associated with second cells, measurements pertaining to current, voltage, temperature, or some combination thereof, and inputting the measurements into the machine learning models to predict the occurrences of the lithium plating for the second cells.

Abstract: In one aspect, computer-implemented method may include receiving, from a cloud-based computing system, one or more machine learning model parameters that are configured to enable predicting a remaining useful life of each cell of a battery pack of a vehicle. The method may include loading, into memory of a processing device at the vehicle, the one or more machine learning model parameters, receiving data comprising one or more measurements and one or more user battery usage profiles, and based on the data, executing a trained machine learning model with the one or more parameters to input the data and to output the remaining useful life of each cell of the battery pack.

Abstract: In one aspect, computer-implemented method may include receiving, from a cloud-based computing system, one or more machine learning model parameters that are configured to enable predicting a remaining useful life of each cell of a battery pack of a vehicle. The method may include loading, into memory of a processing device at the vehicle, the one or more machine learning model parameters, receiving data comprising one or more measurements and one or more user battery usage profiles, and based on the data, executing a trained machine learning model with the one or more parameters to input the data and to output the remaining useful life of each cell of the battery pack.

Abstract: In one aspect, a computer-implemented method for a cloud-based computing system may include receiving data pertaining to a battery pack of a vehicle, wherein the data is measured by one or more sensors associated with the vehicle. The method may include determining, based on the data, whether a sum of a physics-based model estimated terminal voltage minus an actual voltage of the battery pack satisfies a threshold. Responsive to determining the threshold is satisfied, the method may include determining a trigger event has occurred and calibrate one or more parameters of the physics-based model, a machine learning model, or both, wherein the physics-based model outputs one or more properties pertaining to the battery pack of the vehicle, and the machine learning model uses the one or more properties to predict a remaining useful life of each cell of the battery pack.

Abstract: In one aspect, a method for a cloud-based computing system may include training, using test data, machine learning models to predict a remaining useful life of each cell of a battery pack of a vehicle. The method may include using a rule-based evaluator to determine first scores for the machine learning models, using a machine learning based metric evaluator to determine second scores for the machine learning models, using a model selection inference engine to select, based on the first and second scores for the machine learning models, a machine learning model to use to predict the remaining useful life of each cell of the battery pack of the vehicle, and transmitting, to a processing device of the vehicle, the selected machine learning model and parameters to predict the remaining useful life of each cell of the battery pack of the vehicle.

Abstract: In one aspect, a method comprises receiving first data pertaining to a battery pack of a vehicle, wherein the first data is received from sensors associated with the vehicle, and the first data pertains to a battery pack current, a cell voltage, a cell current, a cell temperature, or some combination thereof; receiving second data pertaining to simulation of the battery pack; receiving third data from a manufacturer; receiving historical data on a fleet of vehicles that use the battery pack; predicting a remaining useful life of the battery pack of the vehicle by using a hybrid model comprising a physics-based model that receives the based on the first, second, third, and historical data, and generates properties pertaining to the battery pack; a machine learning model that uses the properties to predict the remaining useful life of each cell of the battery pack; and transmitting the remaining useful life.