Optimized Embedded Architectures for Model Predictive Control Algorithms for Battery Cell Management Systems in Electric Vehicles
Year: 2020
Type: article
Abstract: With the ever-growing concerns about carbon emissions and air pollution throughout the world, electric vehicles (EVs) and hybrid electric vehicles (HEVs) are some of the most viable options for clean transportation. EVs are typically powered by battery packs created from a large number of individual cells such as lithium-ion cells. In order to enhance the durability and prolong the useful life of the battery pack, it is imperative to monitor and control the battery packs at the cell level. The best techniques to provide maximum cell performance, longevity and safety are also the techniques that are the most computationally expensive and least suited towards the constrained resources of embedded systems. Advances in the power and computing capability of both microprocessors and field programmable gate arrays (FPGA) make it possible to bring the computationally expensive algorithms and methods into embedded systems. Model predictive control is an adaptive control approach that uses models to predict the outcome of a control move and then optimize the control move based on the system constraints. This control approach incorporates prediction, optimization and constraints to determine the best input to achieve a desired output and is computationally expensive. The ability to incorporate constraints allows reduction of the safety margins, increasing the capacity or productivity of a system. This approach is well established in industrial applications and is very desirable for use in battery management systems. Currently in MPC for battery cell management there are two popular model approaches for battery cell management, the Equivalent Circuit model (ECM) and the physics-based model (PBM). ECM has the advantage of simplicity and uses an empirical approach to describing battery cell interactions. The disadvantage to ECM is the limit on the operating range of the model, though the model operates very well within the boundaries set by the data collection parameters ECM does not perform well outside the boundaries. The ECM also does not model the internal electrochemical reactions of the battery that lead to battery degradation. The PBM, being based on first order principals that describe the electrochemical reactions of the battery, is more flexible and can handle a wider range of operating conditions and supports monitoring and control of the internal reactions that lead to degradation. The challenge of PBM is reducing the complexity of the algorithm so that it can be implemented in a resource constrained environment while maintaining the ability to accurately model the battery. As it is difficult to physically measure the internal electrochemical reactions of the batteries while in use, an observer is required to estimate the state of the reactions in support of the control algorithm. The non-linear nature of the PBM adds complexity to both the MPC approach and the observer. MPC using PBM has long been considered too complex for resource constrained systems. Our goal in this research is to investigate and architect embedded systems software and hardware co-designs that support the computationally expensive model predictive control algorithms, models, and approaches that maximize both performance and longevity of battery packs at the cell level previously considered too computationally complex for embedded systems applications. The first contribution of this work introduces one software and two hardware embedded architectures that address the computational complexity of ECM based MPC for the at-rest fast charging of an HEV battery. The second contribution introduces novel and efficient software and hardware embedded architectures for a non-linear observer based on the Extended Kalman filter using a physics-based models based on first order principals. These architectures demonstrate the utility and accuracy of the state space PBM developed using the discrete realization algorithm (DRA). Experiments are run on the embedded systems that show the capability to maintain the advantages of PB approaches while using minimal resources. Results are provided that show the resource constrained architectures retain the accuracy and flexibility advantage of the physics-based models using the minimum resources required for mobile battery management. The final contributions are embedded architectures for a unique adaption of model predictive control to create a smart sensor for battery cell management that focuses on preventing cell degradation. These architectures incorporate the PB EKF and a PB MPC on a single system. Two architectures are presented, a software architecture on a 32-bit 128KB microcontroller and a hardware architecture on an FPGA. Experiments run on the actual embedded systems demonstrate that the current design approaches provide excellent performance and easily support the required control intervals. In each case the experiments are run on the actual embedded systems and demonstrate that the current state of research for physics-based battery cell management can be achieved and are becoming suitable for resource constrained embedded systems.
Author: Madsen, Anne Kulbitski
Language: English
Subfield: Automotive Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W3095139270