Speaker
Description
DC Current Transformers (DCCTs) are essential instruments for non-interceptive beam current measurement in particle accelerators. The zero-flux modulation principle demands exceptional symmetry between paired magnetic cores to achieve sub-$\mu$A offset stability. Conventional core matching based on static magnetic parameters provides only an engineering approximation, as it neglects the dynamic magnetization behavior under AC modulation. This paper presents a novel approach employing unsupervised machine learning techniques applied to 6 dynamic magnetic parameters ($\mu_\mathbf{a}$, $\delta$, $B_\mathbf{r}$, $B_\mathbf{m}$, $H_\mathbf{c}$, $H_\mathbf{m}$) measured at 50 kHz sinusoidal excitation for 19 Fe-based nanocrystalline cores. Principal Component Analysis (PCA) reduces the feature space while preserving $89.64\%$ of total variance. An adaptive multi-objective K-Means strategy successfully isolates anomalous specimens ($K=2$), while a density-based evaluation framework partitions the remaining operational cores into 5 highly homogeneous sub-groups. This two-tier matching scheme enables a physically rigorous core pairing that accounts for real-world dynamic magnetization and domain wall losses under actual DCCT operating conditions.
Funding Agency
- This work is supported in part by National Natural Science Fund (No. 12275294).
Footnotes
† huangwei@ihep.ac.cn
| Paper status | Resubmitted proceeding files received and assigned to an editor. Accepted. |
|---|