Battery SOH Trajectory Prediction Using Fleet Data Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining the state of health (SOH) of device batteries in battery-operated machines, such as electric vehicles, are often inaccurate due to the lack of reliable models for unknown battery types and are heavily dependent on usage profiles, making it difficult to predict the remaining life of batteries.
Innovation Solution
A method that uses a central processing unit to evaluate time characteristics of operating variables from multiple batteries of unknown type, employing outlier elimination and domain knowledge to determine a state of health trajectory, allowing for accurate assessment and prediction of battery aging, even for batteries without electrochemical parameterization, by leveraging both domain and data-based knowledge.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physical aging models are used to determine state of health based on historical operating variables, then the state of health can be estimated, but the prediction accuracy is highly inaccurate
Solution Approach 1:
The patent combines multiple data sources including operating variables, state of charge information, temperature data, and aging time characteristics from a fleet of batteries to create a comprehensive state of health model. This merging of multiple data streams and modeling approaches (Coulomb counting, impedance-based methods, and machine learning) significantly improves prediction accuracy compared to single-model approaches.
Solution Approach 2:
The patent introduces intermediate processing steps including data filtering, outlier elimination, and feature extraction between raw operating variables and the final state of health prediction. These intermediary processes clean and prepare the data, removing noise and inconsistencies that would otherwise degrade model accuracy and reliability.
2Adaptability or versatility
If battery types are unknown without extensive pre-startup measurement, then batteries can be deployed flexibly, but no state of health models are available for prediction
Solution Approach 1:
The system enables unknown battery types to self-characterize by collecting and analyzing their own operating data during normal use. Through continuous monitoring of voltage, current, temperature, and state of charge, the battery effectively performs its own identification and modeling, allowing the system to adapt to new battery types without pre-existing models while maintaining assessment accuracy.
Solution Approach 2:
The patent dynamically adjusts modeling parameters and data collection strategies based on the amount of available data for unknown battery types. As more operating data accumulates, the system refines its understanding of the battery's characteristics, transitioning from generic models to battery-specific predictions, thereby maintaining accuracy despite initial lack of type information.
3Quantity of substance
If fleet data from multiple batteries with different usage profiles is used, then more data points are available for modeling, but data quality varies due to different load profiles and user behaviors
Solution Approach 1:
The patent extracts and removes outlier data points and inconsistent measurements from the fleet data through statistical analysis and validation rules. By identifying and eliminating anomalous data resulting from extreme usage conditions or measurement errors, the system maintains high data quality while utilizing the large quantity of available fleet data for robust model training.
Solution Approach 2:
The system segments the fleet data into meaningful groups based on operating conditions, battery types, and usage patterns. This segmentation allows the model to learn from diverse data sources while accounting for variations in load profiles and user behaviors, thereby maintaining precision across different operating scenarios.
Data Source
AI summary
A method for determining a state of health trajectory of a device battery is based on state of health values of device batteries of an identical battery type. The method includes providing time characteristics of operating variables of a multiplicity of device batteries of battery-operated machines in a central processing unit, and determining one or more state of health values of one or more of the multiplicity of device batteries by evaluating a respective characteristic of the operating variables within an evaluation period. The one or more state of health values with the applicable aging times each indicate a data point for the relevant device battery. The method further includes eliminating data points from the determined data points to obtain a set of cleaned-up data points, and ascertaining the state of health trajectory with an accuracy statement for each trajectory point based on the set of cleaned-up data points.