Source-linked AI summary
Bridging Nano and Micro-scale X-ray Tomography for Battery Research by Leveraging Artificial Intelligence
Jonathan Scharf, Mehdi Chouchane, Donal P. Finegan, Bingyu Lu, Christopher Redquest, Min-cheol Kim, Weiliang Yao, Alejandro A. Franco, Dan Gostovic, Zhao Liu, Mark Riccio, František Zelenka, Jean-Marie Doux, Ying Shirley Meng
TL;DR
Battery research needs imaging and analysis methods that can resolve increasingly complex structures across relevant length scales. This review examines X-ray CT, AI/ML analysis, complementary characterization, and computational modeling, concluding that their combination supports predictive multiscale battery models while technical constraints remain for broad application.
Problem
Increasingly complex battery datasets require advanced analysis and complementary methods to extract detailed morphological information and connect experiments with multiscale models.
Method
The review surveys X-ray CT technologies, battery applications, AI/ML analysis, correlative workflows, stochastic generation, and computational modeling for multiscale characterization.
Results
Combining multiscale 3D characterization with AI/ML and modeling may support performance-predictive battery models incorporating phenomena across multiple length scales.
Takeaways & Limitations
CT-derived morphology and AI/ML-generated representative volumes can support predictive electrochemical modeling while reducing the number of required 3D-resolved CT characterizations.
Takeaways & Limitations
Broad correlative application remains constrained by protective-atmosphere requirements, sample and transfer-device design, and chemistry- or instrument-dependent imaging limitations.
Abstract
from arXiv · showhide
X-ray Computed Tomography (X-ray CT) is a well-known non-destructive imaging technique where contrast originates from the materials' absorption coefficients. Novel battery characterization studies on increasingly challenging samples have been enabled by the rapid development of both synchrotron and laboratory-scale imaging systems as well as innovative analysis techniques. Furthermore, the recent development of laboratory nano-scale CT (NanoCT) systems has pushed the limits of battery material imaging towards voxel sizes previously achievable only using synchrotron facilities. Such systems are now able to reach spatial resolutions down to 50 nm. Given the non-destructive nature of CT, in-situ and operando studies have emerged as powerful methods to quantify morphological parameters, such as tortuosity factor, porosity, surface area, and volume expansion during battery operation or cycling. Combined with powerful Artificial Intelligence (AI)/Machine Learning (ML) analysis techniques, extracted 3D tomograms and battery-specific morphological parameters enable the development of predictive physics-based models that can provide valuable insights for battery engineering. These models can predict the impact of the electrode microstructure on cell performances or analyze the influence of material heterogeneities on electrochemical responses. In this work, we review the increasing role of X-ray CT experimentation in the battery field, discuss the incorporation of AI/ML in analysis, and provide a perspective on how the combination of multi-scale CT imaging techniques can expand the development of predictive multiscale battery behavioral models.
1. A Brief History of X-ray Computed Tomography
X-ray CT evolved from medical imaging into a quantitative, multi-disciplinary tool whose decreasing voxel sizes expanded battery and materials applications. Advances in laboratory, synchrotron, and AI-assisted workflows now support more detailed characterization and modeling of complex samples.
- Origins and principles: X-ray CT reconstructs non-invasive cross-sectional and 3D images from X-ray attenuation, providing information about internal morphology and structure.The technique’s contrast originates from materials’ absorption coefficients.
- Analysis and modeling: AI and ML increasingly assist segmentation, complex-dataset analysis, and links between experimental CT data and multi-physics or multi-scale modeling.These methods can reduce processing time while precisely labeling features of interest.
- Origins and principles: Since its first patient brain scan in 1971, CT has expanded from medicine into scientific and engineering applications.The historical record tracks publication year and reported voxel size across medical and electrochemical-storage studies.
- Technological development: CT development has progressively reduced voxel sizes, with MicroCT, synchrotron, and NanoCT systems extending imaging across increasingly fine scales.Figure 1 distinguishes application fields by color and system types by marker shape.
- Present capabilities: Laboratory CT systems now support multi-energy imaging and improved spatial and temporal performance, while synchrotron upgrades target faster imaging.Laboratory sources still lack sufficient temporal resolution for many operando and in-situ structural-dynamics studies lasting minutes to hours.
2. X-ray CT in the Battery Field
X-ray CT provides non-destructive three-dimensional battery characterization, enabling measurement of electrode morphology and its evolution during operation. Instrument choice involves resolution, field of view, scan time, chemistry-specific contrast, and sample-preparation trade-offs.
- Capabilities: X-ray CT enables non-destructive three-dimensional characterization of battery architecture, morphology, and internal structure.Compared with destructive vacuum-based techniques, CT can support in-situ and operando studies.
- Morphological parameters: Surface area, volume, particle size, porosity, pore networks, and tortuosity can be extracted from reconstructed battery volumes.These parameters support analysis of wettability, volume expansion, particle evolution, transport pathways, and electrode structure.
- Experimental trends: Smaller voxel sizes generally coincide with reduced field of view, while nano-regime NanoCT studies are ex-situ and synchrotron studies support in-situ or operando measurements.The combination of limited field of view and long scan times makes dynamic nano-scale experiments difficult.
- Experimental trends: In the micro-regime, laboratory MicroCT supports common in-situ or operando studies, particle-scale analysis, and models using reconstructed porosity and tortuosity.MicroCT can resolve features such as 10–100 µm particles and micro-sized lithium-metal dendrites.
- Chemistry and constraints: Battery chemistry strongly influences instrument selection because lithium weakly interacts with the higher-energy X-rays used in traditional MicroCT.Studies targeting lithium commonly use lower-energy ZPC with NanoCT or SRXTM, but face field-of-view and sample-preparation constraints.
- Segmentation and analysis: Proper segmentation of active material, binder, and pores is mandatory for reliable particle-size, porosity, and tortuosity measurements.Thresholding and related analysis methods distinguish phases using their grey-value distributions.
- Chemistry and constraints: Sample preparation and in-situ cell design remain important challenges because attenuation, contrast, signal-to-noise, sample width, and representativeness must be balanced.Maximizing resolution and contrast can require limiting sample dimensions, constraining statistical representativeness.
3. CT Analysis, Simulation, and Modelling
CT analysis uses artifact correction and phase segmentation to obtain reliable electrode morphology, which can then support stochastic or tomography-derived 3D battery models. MicroCT and NanoCT provide complementary routes, with tradeoffs between phase resolution, representativeness, and model complexity.
- Species segmentation: Battery CT typically segments active material, binder, and pores to extract particle size, porosity, and tortuosity.Global thresholding separates phases using their grey-value distributions, while local methods account for neighborhood statistics.
- CT artifacts and filtering: Segmentation quality depends on data quality and proceeds through preprocessing, phase segmentation, and postprocessing stages.The workflow includes artifact removal, filtering, sharpening, thresholding, and denoising.
- Computational modeling: Battery models use CT-derived morphology or stochastically generated electrode structures to represent geometry and calculate electrochemical performance.Stochastic generation controls composition, particle-radius distribution, porosity, thickness, overlap, surface area, and inactive-phase morphology.
- MicroCT-based models: MicroCT-reconstructed structures can require stochastic addition of carbon and binder because spatial resolution above ~500 nm makes inactive-phase extraction difficult.This approach increases the reliability of 3D computational models while retaining experimentally reconstructed electrode texture.
- NanoCT-based models: NanoCT can distinguish inactive phase, active material, and porosity for direct use in Generation III models, but its narrow field of view raises representativeness concerns.The resulting multiphase structures can also be difficult to import into finite element or finite volume models because of numerous interfaces.
- Advanced model representations: Generation III models increasingly resolve inactive-phase structure and reduce reliance on average geometrical parameters, enabling explicit treatment of cell heterogeneities.Manufacturing-process simulations provide another route toward representative models and predictive digital twins, despite approximations such as spherical particles.
4. Future of Battery X-ray CT
Future battery CT research centers on correlative, multiscale characterization that combines complementary imaging with AI/ML, stochastic generation, and predictive modeling. These approaches can improve segmentation, recover unresolved structure, and connect experimental data with models, although sensitive samples and high-resolution preparation remain practical constraints.
- Correlative workflow characterization: Correlative tomography combines low- and high-resolution imaging with complementary tools to overcome X-ray CT’s chemical and nanoscale morphology limitations.The approach integrates techniques such as NanoCT, MicroCT, electron tomography, STEM/EDS, and FIB-based characterization.
- Correlative workflow characterization: Higher-resolution electron-tomography segmentations significantly improved ML-assisted pore segmentation in wider-field NanoCT data.Electron tomography supplied a high-resolution training dataset while NanoCT provided a wider field of view.
- Correlative workflow characterization: Correlative battery studies resolved electrode constituents, SEI layers, porosity networks, and carbon-binder domains, enabling state-of-charge and tortuosity analyses.These studies combined high-contrast or ptychographic X-ray CT, FIB cross-sections, NanoCT, and MicroCT.
- Correlative workflow characterization: Sub-MicroCT separator fibrils were stochastically generated from SEM and CT data and significantly influenced predicted effective diffusion coefficients.The workflow incorporates features too small to observe directly in MicroCT into transport modeling.
- Practical constraints: Sensitive samples require protective-atmosphere handling, while nanoscale workflows may require difficult sample reduction and higher-throughput milling tools than FIB.The stated alternatives include Plasma FIB, laser PFIB, and broad ion beam milling.
- AI/ML and multiscale modeling: AI/ML methods such as GANs can generate detailed or extended electrode microstructures and accelerate prediction of manufacturing effects on uncharacterized compositions.GANs can be trained on CT volumes or slices and combined with physical manufacturing models.
- AI/ML and multiscale modeling: Open multiscale datasets with characterization metadata are identified as important for reducing dependence on specialized facilities and supporting ML and multiphysics models.The repositories should include both measurement data and conditions describing how characterization was performed.
- AI/ML and multiscale modeling: AI/ML-orchestrated workflows may link CT characterization, multiscale physical models, experimental comparison, and prediction of synthesis or manufacturing conditions.The proposed workflows can couple models sequentially or iteratively at different length scales and fidelities.
5. Conclusion
The conclusion presents CT as a non-destructive source of morphological data for predictive battery models and highlights AI/ML generation and multiscale characterization as future directions. Together, these approaches are expected to support models spanning multiple length scales and broader computational use of realistic electrode structures.
- CT and predictive modeling: CT-derived tomograms and morphological parameters can be incorporated into predictive models of battery performance.Examples include particle distribution, porosity, and tortuosity.
- AI/ML-enabled modeling: AI/ML techniques such as GANs can generate realistic multiphase porous electrode microstructures and reduce the number of required CT characterizations for 3D-resolved models.The generated volumes are intended to remain representative for simulations.
- Multiscale characterization: Combining FIB-SEM, TEM, MicroCT, and NanoCT may enable performance-predictive models that incorporate phenomena across multiple length scales.The conclusion also describes consolidating characterization data and models in open-source repositories and VR tools.