Faulty Sensor Signal Reconstruction in Li-Ion Battery Packs

Early detection and isolation of faults in battery packs are critical to improving performance and ensuring safety. Sensor-related faults such as noisy measurements, sensor bias, sensor drift, and loose connection are typically not safety issues but they could mislead the battery management system (BMS) to take erroneous control actions. Thus, we propose an effective fault tolerance approach to correct faulty voltage and temperature measurements associated with otherwise normal cells by using the measurements from other cells within a group of cells with similar thermal conditions and the same current (e.g., cells in series within a module). This article reconstructs one or more faulty voltage and temperature measurements using other measurements from a series string. Principal component analysis (PCA) applied to median-based voltage and temperature residuals captures cell-to-cell relationships in a series string. This learned cell-to-cell relationship is leveraged to reconstruct the faulty voltage and temperature signals using the remaining nominal voltage and temperature measurements, respectively. Faulty sensor signal reconstruction is validated using data from a battery electric locomotive. The voltage and temperature reconstructions are accurate within 0.51 mV and 0.028 C-degrees, respectively.