Paul Albert Busch

Battery diagnostic models play an essential role in understanding the internal state of batteries. They allow for the online estimation of the cell's states and parameters, such as capacity and internal resistance, without the need to open the battery cells or do extensive offline testing. However, many existing models in the literature are developed and validated based on laboratory datasets, which, therefore, struggle with dynamic load profiles or fluctuating environmental conditions, limiting their application in real-world scenarios. This work addresses these limitations by introducing an advanced battery diagnostic model designed for lithium iron phosphate (LFP) cells. The model is developed for field data analysis to handle dynamic load profiles and can be applied to field data with very low logging frequencies of one sample per minute. It combines an equivalent circuit model of the cell with a hysteresis component and an open-circuit voltage reconstruction model. We validate the model on an extensive field dataset from residential home storage systems. This dataset includes over a thousand battery systems operating under real-world conditions. The validation shows that the model can accurately simulate cell voltage across various load profiles, logging frequencies, and data quality levels. The model demonstrates reliability even with noisy or incomplete data. To analyse the degradation of the batteries, we use the proposed diagnostic model to estimate capacity and power loss from the field data. The results reveal that battery ageing is highly dependent on customer usage patterns. These insights are crucial for understanding how operational behaviours influence long-term battery performance. To extend the impact of this work, we apply deep learning techniques to forecast battery ageing trajectories. The forecasting model uses historical data to predict future capacity loss and power loss. It accounts for differences in load profiles, environmental conditions, and logging quality. By training on diverse degradation paths, the model provides reliable predictions that reflect real-world variability. This ageing prognosis approach allows stakeholders to proactively anticipate maintenance needs and plan replacements. Battery operators can optimise usage to extend battery life. The ability to predict ageing also helps reduce unexpected failures, improving system reliability. These advantages are particularly relevant for residential storage systems, where cost efficiency and long-term performance are critical. In conclusion, this work contributes to a better understanding of LFP battery degradation in residential applications. It provides a robust modelling framework that works across varying load profiles, logging frequencies, and data qualities. Since data logging frequency correlates with data storage costs, this study shows the implications of changing the logging frequency of the battery management system. Additionally, the integration of forecasting adds a forward-looking dimension to battery diagnostics, enabling more informed decision-making.