Comparison of battery internal temperature using electrochemical impedance spectroscopy

Abstract

In a battery storage system, a reliable temperature monitoring mechanism to completely eliminate the occurrence of thermal runaway accidents is a primary concern. At present, temperature sensors are used to measure the battery surface temperature. However, since battery heat is generated by internal electrochemical reactions, measuring the battery surface temperature can result in significant difference and heat transfer delay. This discrepancy can result in delayed detection of over-temperature conditions, potentially leading to safety hazards or even fire. To address these issues, electrochemical impedance spectroscopy (EIS) has been used to estimate the battery internal temperature (BIT). In this paper, EIS-based BIT estimation methods are reviewed and classified into four methods based on the EIS parameters used: the magnitude of impedance (|Z|), the phase shift of impedance (θ), the real part of impedance (Re(Z)), and the imaginary part of impedance (Im(Z)). Guidelines for selecting the appropriate frequency for more accurate BIT estimation are provided. Furthermore, a deep neural network (DNN) was adapted to construct an EIS-based DNN (EIS-DNN) to improve BIT estimation accuracy. EIS measurements were conducted on 28 brand-new batteries at different BITs and states of charge (SoC). The experimental results indicate that the conventional |Z|-based, θ-based, Re(Z)-based, and Im(Z)-based BIT estimation methods are feasible, with average temperature estimation errors (TETavg) ranging from 1.04°C to 3.24°C. However, significant differences were observed in the maximum temperature estimation errors (TETmax). The Im(Z)-based method performed the best, with a TETmax of 3.42°C, whereas the Re(Z)-based method exhibited the worst performance, with a TETmax of 18.59°C. In contrast, the proposed EIS-DNN method achieved significantly improved accuracy, with a TETavg of approximately 0.112 °C and a TETmax of 0.98 °C. Compared to traditional methods, the TETmax is improved by 71.3 %, and the TETavg is improved by 89.2 %.