Predictive battery health detection from pack soaking behavior

1. A computer-implemented method for predicting a quality of a battery having multiple cells, the computer-implemented method comprising:

receiving a first battery measurement data from a plurality of sensors coupled with the battery for a first duration of time, wherein the first duration of time is shorter than a total duration of soaking the battery;

computing a plurality of machine learning features based on the first battery measurement data;

predicting, based on the plurality of machine learning features, a state of the battery during a second duration of time, the plurality of machine learning features being defined by the first battery measurement data;

receiving a second battery measurement data from the plurality of sensors for the second duration of time, the second duration of time being subsequent to the first duration of time, and the second duration of time being within the total duration of soaking the battery;

comparing the predicted state of the battery during the second duration of time to the received second battery measurement data; and

outputting suitability of the quality of the battery based on the state of the battery as predicted.

2. The computer-implemented method of claim 1 , wherein the second duration of time is shorter than or equal to a difference between the first duration of time and the total duration of soaking the battery, further comprising predicting a second battery measurement data for a second duration, wherein the plurality of machine learning features are computed based on the first battery measurement data and the second battery measurement data.

3. The computer-implemented method of claim 2 , wherein predicting the second battery measurement data is based on the first battery measurement data using a time-series prediction.

4. The computer-implemented method of claim 3 , wherein the time-series prediction is performed using a Bayesian neural network.

5. The computer-implemented method of claim 1 , wherein each measurement data comprises a voltage of each cell, an identifier of each cell, and a timestamp corresponding to the measurement data.

6. The computer-implemented method of claim 5 , wherein the plurality of machine learning features comprises a first value A that represents an individual cell voltage drop rates and/or voltage drops across the battery.

7. The computer-implemented method of claim 6 , wherein the plurality of machine learning features comprises a second value B that represents a value of the first value shifted by an average of the first value across all cells in the battery.

8. The computer-implemented method of claim 7 , wherein the plurality of machine learning features comprises a third value C that represents a value of the first value shifted by an average of the first value for a same cell in the battery over the first duration and a second duration.

9. The computer-implemented method of claim 8 , wherein the plurality of machine learning features comprises a fourth value D that represents is computed based on the first value, the second value, and the third value.

10. The computer-implemented method of claim 1 , wherein the battery is installed in a vehicle based on the suitability of the quality of the battery.

11. The computer-implemented method of claim 1 , wherein predicting the state of the battery after completion of the soaking comprises predicting a set of outlier cells in the battery based on the features.

12. The computer-implemented method of claim 11 , wherein the suitability of the quality of the battery is based on the set of outlier cells predicted.

13. A system for predicting a quality of a battery having multiple cells, the system comprising:

a processor comprising a plurality of processing cores;

a plurality of sensors coupled with the processor and the battery, wherein the processor is configured to receive a first battery measurement data from the plurality of sensors for a first duration of time during a soaking of the battery, and predicting the quality of the battery after completion of the soaking by performing a method comprising:

computing a plurality of machine learning features based on the first battery measurement data;

predicting, based on the plurality of machine learning features, during a second duration of time, the plurality of machine learning features being defined by the first battery measurement data;

receiving a second battery measurement data from the plurality of sensors for the second duration of time, the second duration of time being subsequent to the first duration of time, and the second duration of time being within the total duration of soaking the battery;

comparing the predicted state of the battery during the second duration of time to the received second battery measurement data; and

outputting suitability of the quality of the battery based on the state of the battery as predicted.

14. The system of claim 13 , wherein the method further comprises predicting a second battery measurement data for a second duration, wherein the plurality of machine learning features are computed based on the first battery measurement data and the second battery measurement data, and wherein predicting the second battery measurement data is based on the first battery measurement data using a time-series prediction.

15. The system of claim 14 , wherein each measurement data in the first battery measurement data and the second battery measurement data comprises a voltage of each cell, an identifier of each cell, and a timestamp corresponding to the measurement data.

16. The system of claim 13 , wherein the battery is installed in a vehicle based on the suitability of the quality of the battery.

17. The system of claim 13 , wherein predicting the state of the battery after completion of the soaking comprises predicting a set of outlier cells in the battery based on the features.

18. A computer program product comprising a non-transitory memory device that has one or more computer executable instructions stored thereon, and which are executable by one or more processing units to perform a method for predicting a quality of a battery, the method comprising:

receiving a first battery measurement data from a plurality of sensors coupled with the battery for a first duration of time, wherein the first duration of time is shorter than a total duration of soaking the battery;

computing a plurality of machine learning features based on the first battery measurement data;

during a second duration of time, the plurality of machine learning features being defined by the first battery measurement data;

receiving a second battery measurement data from the plurality of sensors for the second duration of time, the second duration of time being subsequent to the first duration of time, and the second duration of time being within the total duration of soaking the battery;

comparing the predicted state of the battery during the second duration of time to the received second battery measurement data; and

outputting suitability of the quality of the battery based on the state of the battery as predicted.

19. The computer program product of claim 18 , wherein the method further comprises predicting a second battery measurement data for a second duration, the plurality of machine learning features are computed based on the first battery measurement data and the second battery measurement data, and wherein predicting the second battery measurement data is based on the first battery measurement data using a time-series prediction.

20. The computer program product of claim 18 , wherein predicting the state of the battery after completion of the soaking comprises predicting a set of outlier cells in the battery based on the features, and wherein the suitability of the quality of the battery is based on the set of outliers predicted.