Deep learning-based breakthrough for next-generation energy systems: applications in battery diagnosis and hydrogen production optimization

Abstract

In recent years, the world has witnessed a surge in the frequency and intensity of extreme weather events, drawing increased attention to global warming and greenhouse gases. At the forefront of this climate crisis stands carbon dioxide, which has been rapidly escalating since the Industrial Revolution. Accordingly, to mitigate the severe impacts of climate change, humanity faces the imperative mission of achieving carbon neutrality. For this purpose, innovations in energy systems are essential, particularly electrification which is transitioning from fossil fuel-based energy to carbon-free electricity. In alignment with this trend, the prevalence of electric vehicles continues to rise, offering a promising solution for eco-conscious mobility. However, this shift has brought the maintenance and management of vehicle batteries into the spotlight for extending lifespan, ensuring safe operation, and implementing recycling processes. As ongoing research for accurately assessing the health status of lithium-ion batteries is actively studied, the current technologies, grounded in the phenomenological estimation of battery health through the analysis of measurable cycling profiles, still remain challenges for general applications. This limitation arises from the diverse patterns found in cycling profiles, influenced by factors such as cathode compositions, battery configurations, and cycling conditions. In each case, the requirement for the individually tailored system, trained on vast cycling data, leads to inefficiencies in terms of both time and cost. Furthermore, the necessity for fully cycling profiles poses an additional challenge to real- world applications, as electric devices frequently undergo partial charging and discharging. To address this challenge, this paper introduces a novel aimed at establishing the generally applicable framework for accurate battery health estimation. The core of this approach is the adoption of spatio-temporal methodology, which involves spatializing cycling profiles. This spatio-temporal technique, combined with deep learning-based regression and generative algorithms, has demonstrated remarkable performance, irrespective of variations in cathode compositions, such as LiFePO4, LiNiCoAlO2, and LiNiCoMnO2, battery configurations, including 14S7P pack and 7P modules, and cycling conditions encompassing environmental temperatures of 15°C, 25°C, and 35°C. Furthermore, this approach has been leveraged to construct the one-step battery reusability verification system that can non-invasively reveal the health of individual modules. This system is expected to offer the advantage of estimating the status of internal modules based on the overall battery pack health, without necessitating pack disassembly and module cycling. Batteries will undoubtedly have a substantial impact on reducing carbon dioxide emissions. Nevertheless, it is important to note that batteries, by themselves, cannot achieve complete carbon neutrality. This limitation stems from the fact that carbon dioxide is also generated during the electricity generation used for the charging process. Hence, there is a need to develop advanced systems focused on enhancing the environmental sustainability of energy production, particularly by effectively reducing CO2 emissions. From this perspective, this study introduces an innovative hydrogen production system that employs intensified series-connected reactors and membranes, with the aim of enhancing energy efficiency and safety by compacting the process units through the integration of multiple well- established technologies. Furthermore, a machine learning-based framework has been developed to investigate the most influential process parameters, providing precise estimation of process feasibilities, including H2 production rate, H2 production cost, and CO2 emissions. It also recommends the optimal conditions that lead to desired performance in the early stages of process design. With these advantages, the proposed hydrogen production system along with the feasibility estimation framework are expected to offer valuable insights to advance the development of eco-friendly energy systems.