CDChangyu Deng

Changyu Deng

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Changyu Deng has a total of 21 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Frequency dependence of effective conductivity induced by spatial variance in isotropic heterogeneous media

    by Changyu Deng, Bogdan Epureanu, Bogdan-Ioan Popa, Wei Lu

    It is an essential and challenging task to determine effective conductivity of heterogeneous media. Different from homogeneous media, heterogeneity causes the effective conductivity to behave as a function of frequency under an alternating current (AC). There have been few reports to quantify this function despite the wide applications of AC on heterogeneous media. This Letter derives the analytical form of the effective conductivity as a function of frequency caused by heterogeneity, where the local properties of the media are isotropic and independent of frequency. We use simulation and experiment data to validate our formulation and find that our form shows better agreement with the data than other models. This research provides a practical guidance on the interpretation and application of impedance and conductivity data.

  • A Generic Battery Cycling Optimization Framework with Learned Sampling and Early Stopping Strategies

    by Changyu Deng, Wei Lu

    Battery optimization is a challenging task due to the large amount of cost and time required to evaluate different configurations in experiment or simulation. It is especially costly to optimize cycling performance since cycling the batteries is costly and time-consuming. In this project, we report an optimization framework with learned sampling and early stopping strategies, designed to optimize battery parameters in an efficient way to reduce total cycling time. It consists of a pruner and a sampler. The pruner stops unpromising cycling batteries to save the budget for further exploration. The sampler can predict the next promising configurations based on query history. The framework can deal with categorical, discrete and continuous parameters, and can run in an asynchronously parallel way to allow multiple simultaneous cycling cells. We demonstrated the performance by a parameter fitting problem for a calendar aging model. Comparisons with current methods are made to demonstrate the effectiveness of our method to reduce cycling time. Our framework can foster battery optimization in simulations and experiments.

  • A Battery Cycling Optimization Framework Coupling Prediction Models and Bayesian Optimization

    by Changyu Deng, Wei Lu

    Energy storage is widely used in many fields, for instance, electrical grid, electric vehicles, portable devices and so forth. Rechargeable batteries such as lithium-ion, lithium-oxygen, sodium-ion, lead acid, and more broadly, supercapacitors, are highly needed to achieve high capacity and long duration while maintaining low cost. An essential task of battery research and development, in both industry and academia, and for both simulation and experiment, is to optimize the parameters for best performance. A special property of battery cycling optimization is that we evaluate the performance as a function of cycles when calculating the score of the objective function. A huge challenge is the difficulty to query the objective function due to the cost to obtain the performance, i.e., long cycling time (in both simulation and experiment) causes battery optimization to be a time-consuming task. What is worse, the space of parameter is often large and thus a great number of queries are necessary. To shorten computation time in simulation, or reduce cost in experiment, we require powerful optimization algorithms. We aim for an asynchronously parallel algorithm which considers early stopping and is flexible to incorporate prior knowledge of the batteries. Our optimization algorithm consists of two major components, a predictor and a sampler. For battery cycling, some information is revealed before finishing all cycles, thus the predictor can predict the final outcome with uncertainty based on existing information. The uncertainty of the prediction is described by a probability distribution. The sampler determines new query points based on existing observations; the observations include finished cycles and predictions from the predictor. The next searching points are those with highest expected improvement over existing query points. We compare our algorithm with current optimization methods to show significant reduction of optimization time.

  • Top Cited Papers
    Open Access

    Self-directed online machine learning for topology optimization

    by Changyu Deng, Yizhou Wang, Can Qin, Yun Fu, Wei Lu

    Topology optimization by optimally distributing materials in a given domain requires non-gradient optimizers to solve highly complicated problems. However, with hundreds of design variables or more involved, solving such problems would require millions of Finite Element Method (FEM) calculations whose computational cost is huge and impractical. Here we report Self-directed Online Learning Optimization (SOLO) which integrates Deep Neural Network (DNN) with FEM calculations. A DNN learns and substitutes the objective as a function of design variables. A small number of training data is generated dynamically based on the DNN’s prediction of the optimum. The DNN adapts to the new training data and gives better prediction in the region of interest until convergence. The optimum predicted by the DNN is proved to converge to the true global optimum through iterations. Our algorithm was tested by four types of problems including compliance minimization, fluid-structure optimization, heat transfer enhancement and truss optimization. It reduced the computational time by 2 ~ 5 orders of magnitude compared with directly using heuristic methods, and outperformed all state-of-the-art algorithms tested in our experiments. This approach enables solving large multi-dimensional optimization problems.


  • A Continuous Model on the Electrochemical Impedance Spectroscopy of Solid-State Electrolyte

    by Changyu Deng, Wei Lu

    Solid state batteries have emerged as a potential next-generation energy storage device due to safety and energy density advantages. Development of electrolyte is one of the most important topics in solid state batteries. The ionic conductivity is a crucial factor to characterize the electrolyte for research and applications. Electrochemical Impedance Spectroscopy (EIS) is a popular measurement technique to obtain the conductivity; a typical experiment setup is to connect the electrolyte with two conductive electrodes, and apply an alternating current. As for post-processing, the intersection of the EIS curve with the real axis is treated as the bulk resistance. This simple interpretation, however, discards other information revealed by EIS such as the slope of the curve at low frequency. What is worse, some features on the curve are not fully interpreted. For instance, the semicircle in the Nyquist plot, often interpreted as charge transfer, also occurs when the electrolyte is connected to two non-reacting electrodes such as stainless steel and gold. To better understand the transport mechanism and interpret EIS curves, we introduce a continuous model to quantify the ion transport and current flow in the electrolyte. Experiments are conducted to verify our model. The produced EIS curves from the model are compared with experiment data to show good agreement.

  • Open Access

    A Facile Process to Fabricate Phosphorus/Carbon Xerogel Composite as Anode for Sodium Ion Batteries

    by Changyu Deng, Wei Lu

    Sodium ion batteries have become popular due to their low cost. Among the possible anode materials for sodium ion batteries, phosphorus has great potential owing to its high theoretical capacity. Previous research that yielded high capacity and long duration for a phosphorus anode used expensive materials such as black phosphorus (BP) and phosphorene. To take advantage of the low cost of sodium ion batteries, we report a simple and low-cost method to fabricate the anode: condensing red phosphorus on carbon xerogel. Even with a large particle size (∼ 50 m) and a high mass loading (2 mg cm−2), the composite cycled at 100 mA g−1 yielded a capacity of 357 mA g−1 or 2498 mAh g P − 1 based on phosphorus after subtracting the contribution of carbon. The average coulombic efficiency is as high as 99.4%. When cycled at 200 mA g−1, it yielded a capacity of 242 mAh g−1 or 1723 mAh g P − 1 , with average degradation rate only 0.06% in 80 cycles. Our research provides an innovative approach to synthesize anodes for sodium ion batteries at extremely low cost, with performance exceeding or comparable to state-of-the-art materials, which will promote their commercialization.


  • Achieving Consistent Diffusivity Measurement between GITT and EIS

    by Changyu Deng, Wei Lu

    Accurate measurement of lithium diffusivity in electrode particles is essential for analyzing and predicting the performance of Li-ion batteries. There are two most widely used techniques: Galvanostatic Intermittent Titration Technique (GITT) and Electrochemical Impedance Spectroscopy (EIS). For GITT, a direct current (DC) pulse is applied to the cell. For EIS, the cell is connected to an alternating current (AC) source. The responses of the batteries are used to calculate diffusivity. GITT and EIS, despite different techniques, measure the same material property and are supposed to yield identical results. However, what has puzzled the community for a long time is that these two techniques often give an order of magnitude difference in the results. Our research aims to resolve this large discrepancy between these two techniques. By analyzing the diffusion profile and approximation error for various particle geometries, we show that these two techniques are consistent only when the current excitation does not impact deep inside the particles, which corresponds to the condition of short pulse time for GITT or high frequency for EIS. GITT does not depend on particle size by its assumption while EIS does. Thus, we propose an innovative approach to use EIS to determine diffusivity accurately independent of particle size or geometry. We further demonstrate experimentally that the two techniques yield identical results under the right measurement conditions. This work may provide insight on the scattering of measured diffusivity data in literature. Figure 1

  • A Framework to Optimize Electrode Morphology

    by Changyu Deng, Wei Lu

    Since the invention of lithium-ion batteries, porous electrodes have been manufactured in essentially the same way: granulating active materials into powder, mixing them with some additives such as carbon black and binder, and pasting them on conductive substrate. Optimizations have been focusing on tuning parameters such as particle sizes, porosity and electrode thickness, with the assumption of a uniform electrode. Some recent work demonstrated better cell performance by engineering the shape of the electrode with features such as nano cylinders, tunnels and fractal branches. 3D printing has been used to fabricate the structure of batter electrode. On the simulation side, however, there have been few reports on freeform electrode shape optimization. Most shape optimizations assume a particular electrode pattern with a few adjustable dimensional parameters. Here we report a method to optimize porosity distribution of electrode without a prior assumption of the pattern. An NMC 333 electrode with lithium metal as the counter electrode is considered as an example. To find the optimal solid material distribution for maximum specific energy, we represented the NMC material distribution by discrete nodal design variables. To solve this high dimensional optimization problem, we used a gradient-free optimization algorithm, self-directed online learning optimization. We show that the method can produce an electrode pattern with the specific energy more than 15% higher than that of the uniform electrode. This generic distribution optimization approach provides a powerful tool for the design and optimization of porous electrodes. Figure 1

  • Open Access

    Self-Directed Online Machine Learning for Topology Optimization

    by Changyu Deng, Yizhou Wang, Can Qin, Wei Lu

    Topology optimization by optimally distributing materials in a given domain requires gradient-free optimizers to solve highly complicated problems. However, with hundreds of design variables or more involved, solving such problems would require millions of Finite Element Method (FEM) calculations whose computational cost is huge and impractical. Here we report a Self-directed Online Learning Optimization (SOLO) which integrates Deep Neural Network (DNN) with FEM calculations. A DNN learns and substitutes the objective as a function of design variables. A small number of training data is generated dynamically based on the DNN's prediction of the global optimum. The DNN adapts to the new training data and gives better prediction in the region of interest until convergence. Our algorithm was tested by compliance minimization problems and fluid-structure optimization problems. It reduced the computational time by 2 ~ 5 orders of magnitude compared with directly using heuristic methods, and outperformed all state-of-the-art algorithms tested in our experiments. This approach enables solving large multi-dimensional optimization problems.