Lithium-ion batteries (LIBs) are now an integral part of our energy system. Their high-performance characteristics make them suitable for automotive, marine and stationary applications. However, safety concerns exist for LIBs related to thermal runaway (TR) which presents significant fire, explosion and toxicity hazards. Computational modelling has been used to predict cell TR [1] and thermal runaway propagation (TRP) in battery modules [2]. However, our previous work [3] has shown that it is important to consider the variation in kinetic parameters of the model to predict the probability, as well as the severity, of cell TR under an abuse scenario. Our previous work is extended herein to consider the effects of cell variations on the predictions of module TRP behaviour. From this, we aim to determine the value and confidence in predicted TRP prevention methods.A 0-dimensional thermal resistive network model for a battery stack of 6 prismatic cells was studied to evaluate the value of heat transfer that is required to prevent TRP. The cells are assumed to be electrically connected, with heat transfer between cell surfaces and tab connections, and heat loss to the environment by convection and radiation. TR heat generation in each cell is modelled by Arrhenius equations for the four major decomposition reactions. The initiation of TR is modelled by a large internal short circuit in the first cell. The TRP behaviour of NMC and LFP cell battery stacks are compared. Uncertainty analysis is performed via Monte Carlo analysis.The model, not considering parameter uncertainty, is validated against NMC data and shown to predict TR behaviour accurately. Both the NMC and LFP cell stack undergo TRP at ambient conditions. Considering parameter uncertainty, it is shown that the predictions of maximum temperature, time to TR and time to TRP become more uncertain as TRP progresses. Further, while the uncertainty in maximum cell temperatures is similar between chemistries, for time to TRP and TRP it is greater for the LFP stack.Considering a lumped heat dissipation coefficient to account for radiation and convection, it is found that TRP is completely prevented in the NMC and LFP stacks at values of 344W/m2K and 150W/m2K, respectively (see Figure 1). However, when considering parameter uncertainty, TRP is only prevented 40% and 45% of the time for the NMC and LFP stack, respectively. To reach a median probability of 99% for the prevention of TRP the heat dissipation coefficient has to be increased to 950W/m2K and 300W/m2K for the NMC and LFP stack, respectively (see Figure 1). As such, with a heat dissipation coefficient up to 500W/m2K and 1000W/m2K for heat pipe cooling and liquid immersion cooling, respectively, LFP cell stack TR can be mitigated by thermal management systems discussed in the literature. However, NMC cell stacks requiring a heat dissipation coefficient over 950W/m2K would entail more complex forced convection or condensing vapour technique.Using a thermal resistive network model of a LIB stack to evaluate TRP, it is shown that the predicted heat transfer coefficient to prevent TRP is 2-3 times larger (depending on cell chemistry) when considering parameter uncertainty compared to typical modelling techniques. This highlights the importance of considering uncertainty in TR modelling when determining safety-critical parameters.Acknowledgements: This work was supported by the Faraday Institution [grant number FIRG061].
Read full abstract