Non-Invasive Early Detection of Lithium-Ion Battery Heating Using CAN–BMS Temperature Forecasting

Abstract

Temperature monitoring is critical for lithium-ion battery packs during transportation and warehousing, yet common surface-based methods (e.g., thermocouples or infrared thermography) suffer from thermal lag and may indicate internal heating too late. Modern battery management systems (BMS) already measure pack temperature using embedded sensors and broadcast the value over the Controller Area Network (CAN) bus. This paper presents a non-invasive early-warning approach that reads BMS-reported pack temperature from CAN (the arithmetic mean of three factory-installed NTC sensors between modules), decodes it using a manufacturer-provided DBC description, and forecasts temperature over 120 s, 300 s, and 600 s horizons. Three lightweight predictors are compared: a Kalman-based trend model, linear extrapolation, and a persistence baseline. Forecast accuracy is evaluated using one experimental charge–discharge cycling test, and threshold-anticipation performance (lead time relative to 40∘C and 60∘C ) is evaluated using two controlled heating simulations. The experimental dataset is further used to compare CAN–BMS temperature with pack-surface thermocouples and to quantify internal-to-surface thermal lag. Results show that CAN–BMS temperature responds substantially earlier than the pack surface, and that multi-horizon forecasting can provide several minutes of additional reaction time for threshold-based warnings in the simulated heating scenarios. The findings support CAN-accessible temperature forecasting as a practical, non-invasive safety monitoring layer for lithium-ion battery transport, warehousing, and recycling operations.