Soft Short-Circuit Fault Diagnosis for Vehicular Battery Packs With Interpretable Full-Dimensional Statistical Analytics

Severe thermal runaway is induced in the majority by short-circuit faults inside battery packs in vehicular applications. In this article, the challenging issue of interpretable soft short-circuit fault diagnosis for lithium-ion battery packs is investigated comprehensively. By using the streaming differential matrices from improved virtual voltage measurement, a full-dimensional statistical analysis scheme considering the random occurrence of a soft short-circuit is proposed to characterize the fault signatures. On this basis, an eigenvalue-decomposition-based fault detection indicator is designed, and the quantitative threshold correlation between the designed fault indicator and the original fault signature is derived such that the interpretable fault detection control limit self-commissioning can be achieved. Furthermore, the detected fault can be localized by evaluating the eigenvalue contributions to the designed fault detection indicator. Finally, a test platform with commercial lithium-ion phosphate cells in a compact battery pack is constructed to validate the approach. Experiments conducted with the World Light Vehicle Test Cycle confirm that millivolt-level short-circuit fault diagnosis can be achieved within minutes, and the fault detection and localization procedures are complementary such that the false alarm rate is minimized. It is robust against both the momentary and cumulative short-circuit conditions even with a maximum 2-C current within almost 83% state-of-charge range.