Battery Pack RUL Prediction Using Hybrid Physics-AI Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for predicting the remaining useful life (RUL) of batteries and assets are inadequate, failing to account for electrochemical dynamics, user usage profiles, temperature variations, physical structure properties, and manufacturer data, leading to inaccurate predictions and increased maintenance and recycling demands.
Innovation Solution
A hybrid model combining physics-based and machine learning models, utilizing sensor data, manufacturer information, and historical data to predict RUL, with a model selection engine for optimal performance and continuous calibration through event-based triggers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional methods are used for predicting RUL, then the prediction process is simple, but the prediction accuracy is low
Solution Approach 1:
The patent combines physics-based models with machine learning models to create a hybrid prediction system. The physics-based model incorporates electrochemical dynamics, temperature variations, and battery structure properties, while the machine learning model learns from sensor data and historical information. This merging allows the system to achieve high prediction accuracy by leveraging both theoretical understanding and data-driven patterns, resolving the contradiction between simple methods and accurate predictions.
Solution Approach 2:
The prediction system uses a composite approach by integrating multiple data sources (sensor data, manufacturer data, historical data) and multiple modeling approaches (physics-based and machine learning). This composite structure enables the system to capture complex battery degradation mechanisms that single methods cannot, thereby improving prediction accuracy without relying on overly simplistic models.
2Measurement precision
If comprehensive data collection is performed, then prediction accuracy improves, but data processing complexity increases
Solution Approach 1:
The patent segments the comprehensive data collection into distinct categories: sensor data (current, voltage, temperature), manufacturer data (battery specifications, chemistry), and historical data (usage patterns, degradation history). Each segment is processed by specialized components within the hybrid model, allowing manageable handling of complex data while maintaining high prediction accuracy through targeted analysis of each data type.
Solution Approach 2:
The physics-based model acts as an intermediary that translates complex multi-source data into meaningful degradation indicators. It processes sensor data, manufacturer data, and historical data through electrochemical principles, converting raw comprehensive data into interpretable RUL predictions. This intermediary approach simplifies the overall data processing complexity while preserving prediction accuracy.
3Reliability
If physics-based models are used, then electrochemical dynamics are captured, but computational requirements increase
Solution Approach 1:
The patent implements a partial physics-based approach where only the most critical electrochemical dynamics are modeled explicitly, while less significant effects are captured through machine learning components. This partial action allows the system to maintain electrochemical accuracy for dominant degradation mechanisms while reducing overall computational power requirements compared to a complete physics-based model.
Solution Approach 2:
The patent substitutes complex physics-based calculations with machine learning models for certain aspects of degradation prediction. The machine learning components learn patterns from data and can predict degradation trends without requiring detailed physics-based computations, thereby reducing computational power requirements while maintaining reliability through the complementary physics-based model.
4Measurement precision
If continuous monitoring is implemented, then RUL prediction is improved, but energy consumption increases
Solution Approach 1:
The patent implements periodic monitoring and prediction updates rather than continuous real-time processing. The hybrid model processes sensor data and updates RUL predictions at predetermined intervals or when significant changes occur in battery behavior. This periodic action maintains accurate RUL prediction by capturing essential degradation trends while significantly reducing energy consumption compared to continuous monitoring and processing.
Data Source
AI summary
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; 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 first 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 for presentation.