[Paper Review] Data-driven Thermal Anomaly Detection for Batteries using Unsupervised Shape Clustering

Xiaojun Li, Jianwei Li+2 more|arXiv (Cornell University)|Mar 15, 2021

Advanced Battery Technologies Research14 references30 citations

TL;DR

This paper proposes an unsupervised, data-driven thermal anomaly detection method for batteries using K-shape clustering to compare shape-similarity of temperature signals. By detecting deviations in shape patterns across sensor measurements—without requiring continuous data, labeled examples, or prior knowledge of pack configuration—the method enables early detection of thermal anomalies, outperforming traditional BMS by identifying over-temperature faults over 90 minutes earlier in real-world EV data.

ABSTRACT

For electric vehicles (EV) and energy storage (ES) batteries, thermal runaway is a critical issue as it can lead to uncontrollable fires or even explosions. Thermal anomaly detection can identify problematic battery packs that may eventually undergo thermal runaway. However, there are common challenges like data unavailability, environment and configuration variations, and battery aging. We propose a data-driven method to detect battery thermal anomaly based on comparing shape-similarity between thermal measurements. Based on their shapes, the measurements are continuously being grouped into different clusters. Anomaly is detected by monitoring deviations within the clusters. Unlike model-based or other data-driven methods, the proposed method is robust to data loss and requires minimal reference data for different pack configurations. As the initial experimental results show, the method not only can be more accurate than the onboard BMS and but also can detect unforeseen anomalies at the early stage.

Motivation & Objective

• To address challenges in cloud-based battery thermal anomaly detection, including data loss, sensor asynchrony, battery aging, and environmental variations.
• To develop a model-free, data-driven method that requires minimal reference data and no labeled training examples.
• To enable early detection of thermal anomalies, especially those developing gradually, before threshold-based systems trigger alarms.
• To improve robustness against configuration-specific variations such as sensor placement and pack design differences.
• To provide a scalable solution suitable for large-scale fleet and energy storage systems using unsupervised learning.

Proposed method

• Employs the K-shape clustering algorithm to group temperature measurements based on shape-similarity using a shape-based distance (SBD) metric.
• The SBD metric is defined as 1 minus the maximum normalized cross-correlation between two time series, making it invariant to scaling and time shifts.
• Clusters are updated iteratively via centroid re-estimation and assignment based on SBD, with convergence monitored over iterations.
• Anomalies are detected by monitoring deviations in shape similarity within clusters, using a total distance function combining intra-cluster and inter-cluster changes.
• A confidence level for anomaly detection is computed using a weighted combination of shape deviation and cluster consistency metrics.
• The method is applied to short segments (e.g., 4.1 minutes) of temperature data, enabling real-time or near-real-time monitoring.

Experimental results

Research questions

• RQ1 Can shape-based clustering detect thermal anomalies earlier than conventional threshold-based BMS in real-world EV data?
• RQ2 How robust is the proposed method to intermittent or long-term data loss and sensor asynchrony?
• RQ3 Can the method detect anomalies without prior knowledge of battery configuration, sensor layout, or labeled fault data?
• RQ4 How does the method perform under battery aging and environmental variations?
• RQ5 Can the method distinguish between normal configuration-induced thermal differences and actual thermal faults?

Key findings

• In Case I, the proposed method detected an over-temperature fault 90 minutes before the onboard BMS triggered an alarm, based on early shape deviation in sensor #13.
• The method successfully identified the anomaly at the onset of shape deviation, even when temperature differences were still small, demonstrating early warning capability.
• In Case II, the method correctly classified two distinct thermal clusters corresponding to different battery packs, avoiding false alarms triggered by the BMS due to design-induced temperature differences.
• The method remained robust to intermittent and long-term data loss, as shape similarity is preserved despite missing data points.
• The approach detected anomalies without requiring labeled data, model training, or prior knowledge of sensor locations or pack configurations.
• The method achieved higher accuracy than the BMS in detecting real thermal faults while reducing false positives in normal operational variations.