This Small Business Technology Transfer Research (STTR) Phase I project aims to improve predictive battery models and control systems for grid-scale energy storage applications. A typical grid-scale battery system is composed of thousands of individual batteries that may have initial material and manufacturing variations that tend to increase over time and reduce overall string efficiency, with the result that energy storage installations must be significantly over-specified, making them too expensive for many customers. There is a great opportunity to develop an integrated predictive battery modeling and control system that can determine the exact performance of each battery in operation and optimize string function. The research objectives of this SBIR project are to develop Neural Network (NN) based predictive models of battery performance using information gathered early in the life of each cell like Electrochemical Impedance Spectroscopy measurements in addition to current, voltage, and temperature measurements that can be taken throughout the life of the battery, in order to accurately estimate the state of charge and state of health of each battery in the battery string. The NN models will be incorporated into the control strategy to operate the battery string safely but aggressively, thereby decreasing the total system cost and required volume.
The broader impact/commercial potential of this project is to enable grid-scale energy storage by reducing system costs. The main barrier to the adoption of energy storage on the grid is its high cost. Recent advancements in energy storage technology have resulted in lower cost, longer-life batteries capable of meeting grid requirements, though there have not been the analogous transformative improvements to battery management systems to optimize system efficiency and cost-effectiveness. The innovations supported by this SBIR will enhance scientific and technical understanding of battery function and failure modes, resulting in improved battery performance and lifetimes. The addition of energy storage to the grid will have an enormous societal impact, as storage is required to firm zero-carbon renewable sources such as wind and solar and can reduce energy prices by time-shifting energy loads. While the market for these types of stationary battery systems is currently less than $5 billion, this sector is expected to surge to approximately $100 billion in the next ten years. The dominant battery management system technology for these systems has not yet been established. Commercial advanced battery controls are the key to unlocking this market and represent the next step toward a lower carbon, more sustainable energy future.
Urban Electric Power (UEP) is developing safe, low-cost zinc-manganese dioxide (Zn-MnO2) rechargeable batteries (See Figure A) originally developed at the CUNY Energy Institute (CUNY-EI) for use in grid-scale electricity storage systems and uninterruptible power supplies. Cost-effective energy storage is the missing link that can make otherwise intermittent renewable energy sources available when they are most needed, significantly increasing their value and speeding their adoption. Stationary energy storage is price-competitive with other forms of electricity generation when the cost of storage is below $0.05-$0.10/kWh-cycle. This grant has enabled UEP and the CUNY-EI to begin developing state of health and failure prediction algorithms using electrochemical techniques that measure transient responses of batteries, physical models, and Motivation: The ability to predict state of health, and potential battery failures ahead of their happening is very important for the successful commercialization of our Zn-MnO2 battery chemistry for several reasons: 1) Speeding Research and Development Time: cycle life for batteries are necessarily estimates, no startup has time to test cells for 10 years even if they are expected to have a useful life that long. Determining quickly which formulations under test have the most potential to reach cycle life targets is important for getting to market quickly. 2) Predicting Battery Failure to Reduce Unscheduled Maintenance: eventual installations of the Zn-MnO2 batteries will have up to 576 cells connected in series, if one dies, the entire series rack is down. Being able to predict that a cell is nearing failure ahead of it actually failing will allow it to be replaced during off hours, at the convenience of the customer. 3) Automating State of Health Assessment: reliable predictive state of health measurements can also allow for adaptive control algorithms that proactively increase the life of underperforming cells, reducing failures. Transient techniques for in situ electrochemical cell testing Electrochemical power sources are cells with two electrodes: an anode and a cathode. At each electrode an electrochemical reaction takes place: oxidation at the anode and reduction at the cathode. Each of these reactions has inherent kinetics, and therefore proceeds at a given rate. The rates of the two electrodes must balance, and therefore the cell is limited by the slower electrode. This limitation manifests as an electrode polarization. Both electrodes are polarized some amount from their equilibrium potential, and the cell potential is thus determined by electrode polarization: Ecell = E0c - E0a - ηc + ηa, where E0c and E0a are the cathode and anode equilibrium potentials and ηc and ηa are the corresponding overpotentials. The overpotentials (polarizations) are caused by the many physical and chemical phenomena occurring in the electrodes. These phenomena include the electrochemical reaction kinetics as well as mass transport, electric migration of ions, double layer charging, etc. As these phenomena act in concert, only the most kinetically limiting of them will be obvious in steady state potential-current (E-i) data from the cell (See Figure B). However, cell to cell variations in these phenomena indicate which cells are flawed, and changes in these parameters in each cell as a function of time indicate which cells are overly stressed and prone to failure. Two transient techniques which can be used to detect phenomena at many time constants are electrochemical impedance spectroscopy (EIS) (See Figure C) and current interrupt techniques (CIT) (See Figure D). Ideally these allow all separate ohmic, kinetic, and mass transport phenomena to be observed, distinguished by time constant. When a battery fails due to internal physical degradation, this will be obvious as it will be mirrored by negative change in the cell’s E-i response. However, a battery nearing failure can maintain acceptable E-i characteristics, creating a challenge in recognizing its condition. The physical and chemical phenomena comprising the failing electrode will give some signal that the cells are being compromised. For this grant, UEP and the CUNY-EI used both EIS and CIT data to develop models of battery behavior. More detailed but time-consuming to gather, EIS data allowed development of a physics-based model of the response of the system (See Figure E) that can provide detailed information about specific failure modes and localize them in the anode or the cathode. CIT data was gathered on a larger scale (more than 100 cells), and allowed the development of a neural network model (See Figure F) that was shown to effectively predict future cell behavior using only a small amount of early cell cycling data. In both cases, it was shown that transient response data provided useful information about the future state of cells and will be useful in further modelling and prediction efforts.
