The traction battery used in railway systems faces significant challenges in estimating the state of a large-scale battery system due to limited computing power and memory. Additionally, bidirectional communication between the battery system and the cloud is often unavailable for various reasons. Moreover, state estimation algorithms require numerous parameters to cover the entire battery lifecycle. To address these challenges, multiple steps are necessary to develop algorithms that ensure highly accurate state estimation for customers.
In the first step, the performance, safety, and benchmarking of battery cells are tested. At this stage, key characteristics and parameters such as the OCV-SoC curve, impedance, and aging behavior are extracted. Given the time-consuming nature of these tests and limited laboratory resources, generative AI is employed to create synthetic test data based on realistic cell tests. This accelerates and optimizes the battery testing process. Using the results from this stage, suitable algorithms for state estimation (e.g., SoC and SoH) are developed, and their parameters are fine-tuned.
Within the second step, the algorithms developed in the laboratory are implemented in the Battery Management System (BMS). To enhance the accuracy of these algorithms, direct measurement methods such as OCV comparison and onboard Electrochemical Impedance Spectroscopy (Onboard EIS) are incorporated and integrated into the estimation model. These methods are executed at the cell/module level and aggregated to provide estimations at the string and pack levels. Ultimately, customers receive a single representative state value for the battery system.
In the final step, field data generated by the BMS is sent to the cloud. This data is used to simulate and validate the models in the cloud by comparing the results with the state estimation provided by the BMS. Complex algorithms, such as machine learning models that cannot be implemented directly in the BMS due to computational limitations, are executed in the cloud. These advanced algorithms enable monitoring of the battery system’s behavior, detecting abnormal states, and preparing for rapid service response as soon as possible.
By providing a comprehensive overview of the algorithm development cycle, this study contributes to the advancement and robustness of BMS development. Furthermore, the research highlights the synergy between laboratory testing, generative AI, field implementation, and cloud simulation to overcome the challenges posed by limited computing power in BMS. This approach not only improves the accuracy of state estimation but also enables early defect detection and rapid response to potential issues.