Artificial intelligence (AI) represents a potent tool for tackling problems where mathematical modeling is challenging. By utilizing machine learning techniques with chemical process operation data, it is possible to forecast the performance of chemical processes and identify optimal operating conditions. This approach has the potential to enhance process efficiency, reduce costs, and minimize environmental impact.
State of health (SOH) and remaining useful life (RUL) estimation for Lithium Ion Batteries (LIBs) using machine learning
Lithium ion batteries (LIBs) are gaining importance due to their widespread use in consumer electronics and electric vehicles. As part of an on-going effort involving smart manufacturing and diagnosis of LIBs, we aim to predict the state-of-health (SOH) and remaining useful life (RUL) of LIBs under various charging/discharging conditions combining data and first-principles based models. Degradation factors of LIBs include SEI (Solid Electrolyte Interphase) formation, dendrite, and lithium plating. However, physical models that can fully and comprehensively explain all these factors do not currently exist. Additionally, these models are too complex to use in actual LIB operations.
Therefore, data-based models, which are simpler and easier to build, are essential for the diagnosis (SOH) and regression (RUL) of LIBs’ lifespan. We select degradation variables reflecting the process knowledge and perform degradation modeling based on experimental and real operation data. Furthermore, in order to build a model that is valid under a wide range of operation patterns, we seem to combine the advantages of physical models and data-based models into a hybrid model using machine learning techniques.
Associated members: Jaewook Lee
Active learning for efficient identification of optimal operating condition of SMB process
The simulated moving bed (SMB) process is a chromatographic process that enables continuous separation of mixtures. The continuous separation is made possible by arranging several, i.e., four or more, columns and periodically switching the positions of the inlet ports of the feed and desorbent, and the outlet ports of the products, thus giving the effect equivalent to rotating the solid phase. Compared to conventional batch chromatography, the SMB process is preferred for industrial-scale separations as it reduces the quantity of the stationary phase and desorbent used while achieving high purity and yield.
In industrial operation, it important to quickly find the optimal operating conditions for a new separation task, because off-spec products are produced during the test runs, which can be expensive. On the other hand, the SMB process takes a long time to reach a cyclic steady state when operating conditions are changed. Therefore, methods such as random search or grid search to identify optimal or even feasible operating conditions are not appropriate. In this case, it can be helpful to use a first principles-based model, but a high-fidelity model of the SMB process can take a long time to simulate, hours or even days, making it difficult to use in practice.
Therefore, we aim to develop an automated algorithm that efficiently recommends new operating conditions to try based on the results of previous tests, so that the optimal operating condition may be found quickly. Bayesian optimization, one of the representative active learning methods, is further developed and tailored to this purpose.
Associated members: Woohyun Jeong