A Multimodal and Fine-Tuned Large-Language Model for XPLE Cable Health Assessment

View at publisher Summarise with VIRAT
Authors: Rudai Yan, Changyou Suo, Yan Xu
Cited by: 0
References: 6

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Health assessment of underground cross-linked polyethylene (XLPE) power cables is crucial for maintaining reliable operation of a power grid. Existing methods are mostly based on the condition data, without sufficient utilization of the water tree formed in the cables, a very important health indicator. To makefull use of heterogenous data of cable health information, this letter proposes a multimodal large-language model (LLM) method for cable health assessment, which simultaneously uses water tree images and condition monitoring data, including insulation resistance (IR), IR Ratio, very low frequency tan δ, and age, as the input and generates a numerical health index. For training data collection, lab experiments are conducted to capture water tree images of retired cables. Then, the images and cable condition data are combined to fine-tune a LLM through low-rank adaptation (LoRA), which is highly computational efficient and requires much less memory. The proposed method is implemented with Gemma 3-4B and tested on real-world underground cable data.

Fields of Study & Subjects

High voltage insulation and dielectric phenomena Power Transformer Diagnostics and Insulation Electrical Fault Detection and Protection

Publication details

Journal / source
IEEE Transactions on Power Delivery
Volume
41
Issue
3
Pages
1508–1511
Publication date
30 March 2026
Language
EN