Battery Anomaly Degradation Detection System Using Autoencoders

Early Detection of Abnormal Battery Degradation Using Bayesian Convolutional Autoencoders and Unsupervised Clustering

This technology converts battery data into patterned images and inputs them into a Bayesian Convolutional Autoencoder (BCAE) that utilizes Laplace and Student's t-mixture distributions as priors to generate a Virtual Health Index (VHI) with quantified uncertainty. The resulting index is then used for unsupervised clustering to detect abnormal degradation.

Conventional threshold-based or simple deep learning methods require large amounts of labeled data. These approaches often underperform in data-constrained environments and struggle to model complex degradation patterns and data uncertainty, making it difficult to identify early-stage abnormal degradation.

This technology optimizes Bayesian autoencoder network parameters using a modified Bayes by Backprop algorithm that integrates Monte Carlo sampling with a Gumbel-Softmax module. By precisely modeling diverse degradation patterns through mixture distributions, it generates a VHI even with limited labels and automatically identifies anomalies via clustering methods like GMM or DBSCAN. It can be applied to EV fleet management servers, battery cell inspection, and second-life battery grading systems, enabling early detection of defective cells without the cost of labeling.

‍

‍

Key Features:

• A database storing a first set of data containing multiple battery data received from an input unit
• A data preprocessing unit that converts the first set of data into patterned images while maintaining temporal and spatial correlations
• A feature learning unit that extracts features from the patterned images and trains an AI model to generate a Virtual Health Index for the battery
• An anomaly detection unit that detects abnormal battery degradation by clustering the Virtual Health Index generated from a second set of data input by the user

‍

This invention was developed with support from the Ministry of Trade, Industry and Energy’s project for the development and demonstration of a microgrid platform for energy conversion of industrial complex waste resources.

‍