A Diagnostic Approach for Backup Time Estimation and Early Detection of Abnormal Cells in UPS Systems

Abstract

UPS (Uninterruptible Power Supply) systems operate predominantly in a standby state, making it difficult to obtain real usage data, and impedance-based diagnostic techniques suffer from inherent limitations—not only restrictive measurement conditions but also user-dependent fitting during analysis. For these reasons, the lifetime of UPS batteries is often evaluated based on the period of use rather than their actual performance. Considering that the primary role of a UPS is to deliver stable power for a required duration during outages, the backup time should serve as a direct and meaningful indicator of battery health. Furthermore, even a single abnormal cell can degrade the overall pack performance and compromise system reliability, highlighting the necessity of a non-invasive method capable of detecting abnormal cells at an early stage. In this study, we introduce a new diagnostic framework tailored for UPS operating environments by employing a low-current pulse test that does not induce battery degradation. Using the amplitude indicator extracted from the pulse response, we demonstrate that an SVR-based regression model can accurately estimate the backup time without requiring a full discharge test. In addition, an SVM-based classification model was constructed to identify abnormal cells using only amplitude values obtained from the first three pulse-test loops, and the incorporation of data augmentation significantly enhanced both accuracy and recall. The proposed method overcomes the diagnostic limitations of conventional UPS environments and provides an integrated approach capable of simultaneously estimating backup time and detecting abnormal cells using only the pulse amplitude, thereby offering substantial practical value for real-world UPS battery management.