Electric Potential-Augmented ECT Benchmark Dataset
收藏资源简介:
该数据集是由清华大学研究团队构建的电势增强型电容层析成像基准数据库,旨在为物理引导的机器学习方法提供标准化测试平台。数据集包含20,000个仿真样本,覆盖单棒状、双棒状、环状和分层四种典型流型,每个样本包含介电常数分布图像、28维电容向量以及8个激励对应的全场电势分布图,数据通过COMSOL-MATLAB有限元仿真流程生成。数据集通过显式保留电势场这一关键物理中间量,为理解电容层析成像的软场效应提供了物理可解释的数据基础,主要应用于电容层析成像的正反问题研究、物理引导的深度学习算法开发,以及跨域泛化性能评估等领域。
This dataset is a potential-enhanced electrical capacitance tomography (ECT) benchmark database constructed by the research team from Tsinghua University, aiming to provide a standardized testbed for physics-guided machine learning methods. The dataset contains 20,000 simulation samples covering four typical flow regimes: single rod, dual rod, annular, and stratified flow. Each sample includes a dielectric constant distribution image, a 28-dimensional capacitance vector, and full-field potential distribution maps corresponding to 8 excitations. The data is generated via a finite element simulation workflow combining COMSOL and MATLAB. By explicitly retaining the potential field as a critical physical intermediate variable, the dataset provides a physically interpretable data foundation for understanding the soft-field effect of electrical capacitance tomography. It is mainly applied to research on forward and inverse problems of ECT, development of physics-guided deep learning algorithms, and cross-domain generalization performance evaluation, among other research fields.




