Battery Fault Prediction Using Trace Graph Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for monitoring device batteries, such as those in electric vehicles, rely on rule-based anomaly detection which only identifies faults after threshold values are exceeded, failing to predict imminent failures like thermal runaway or total battery failure, and are inaccurate due to model deviations in physical ageing state models.
Innovation Solution
A method using a trace graph model to predict anomalies by evaluating temporal operating variable curves, assigning operating feature points to nodes with transition probabilities, and calculating the overall probability of fault occurrence, combined with a hybrid ageing state model that corrects physical models with data-based corrections for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If rule-based anomaly detection using threshold values is used, then fault detection is simple and fast, but prediction capability for imminent failures is lost and accuracy is insufficient
Solution Approach 1:
The patent applies preliminary action by using trace graph models to predict future battery states and anomalies before they occur. The model forecasts the probability of reaching critical states (like thermal runaway) based on current operating conditions and historical patterns, enabling proactive intervention before actual failure happens. This transforms detection from reactive threshold-based to predictive forward-looking analysis.
Solution Approach 2:
The trace graph model acts as an intermediary between raw operating data and failure prediction. It processes operating variable curves through a structured model that captures temporal patterns and transitions, converting complex time-series data into probabilistic predictions about future battery states without direct threshold comparison.
2Measurement precision
If physical electrochemical battery models based on differential equations are used, then detailed battery state modeling is achieved, but model deviations occur and accuracy is reduced
Solution Approach 1:
The patent implements feedback by continuously comparing model predictions with actual operating data and using this information to refine the trace graph model. The model learns from real battery behavior patterns, adjusting transition probabilities and state representations to better match actual degradation paths, thereby reducing model deviations over time.
Solution Approach 2:
The approach changes parameters from fixed physical model constants to adaptive probabilistic parameters. Instead of relying on fixed differential equation parameters that may not capture real behavior, the trace graph uses learned transition probabilities that adapt to specific battery instances and operating conditions, improving reliability without sacrificing modeling detail.
3Measurement precision
If high temporal resolution sampling (1-100 Hz) is used for operating variable data, then accurate battery state ascertainment is achieved, but data processing complexity and computational load increase
Solution Approach 1:
The patent extracts only the essential information from high-frequency operating variable curves by identifying and processing only the critical transitions and state changes relevant to degradation. The trace graph model processes simplified representations of operating patterns rather than raw high-frequency data, reducing computational complexity while maintaining prediction accuracy.
Solution Approach 2:
The approach segments the continuous operating variable curves into discrete temporal segments and state transitions that are meaningful for degradation analysis. By dividing the time series into relevant intervals and focusing on transition points rather than processing every data point, the system maintains precision while reducing processing burden.
4Reliability
If continuous monitoring and evaluation of operating variable curves is performed, then real-time fault detection is achieved, but computational resources and processing time are consumed
Solution Approach 1:
The trace graph model performs preliminary computation by pre-calculating transition probabilities and degradation paths during off-peak times or in advance. This allows the system to make rapid predictions during actual monitoring without performing computationally intensive real-time analysis, reducing processing time while maintaining continuous monitoring capability.
Data Source
AI summary
A monitoring method includes ascertaining from a temporal operating variable curve of operating variables of a battery an operating feature point characterizing a battery state and/or an operating history in a time period between a most recent and a second most recent change point time in the curve, providing a trace graph model comprising nodes with respective characteristic operating feature points connected via directed transitions with respective transition probabilities. One node is an anomaly node associated with an operating feature point corresponding to a particular fault of the battery. The ascertained operating feature point is assigned to one of the nodes as a monitoring node. An overall probability of occurrence of a fault is ascertained as a sum of all path probabilities from the monitoring node to the anomaly node. The path probability is a product of the respective transition probabilities along the nodes of the respective path.