基于声纹数据标准化的变压器质量缺陷检测研究

摘要：

【目的】针对传统声纹检测方法受数据质量不统一及模型泛化能力弱的问题，研究声纹数据标准化方法并构建基于深度学习的质量检测模型，支撑电力变压器无损检测和智能运维。【方法】通过分析变压器声纹特性与缺陷检测瓶颈，构建了涵盖信号采集、降噪处理、特征提取的标准化预处理流程，提升数据质量与一致性。引入基于CNN-Transformer混合架构深度学习模型，实现对多种典型缺陷的识别。【结果】形成涵盖声压级、信噪比、奇偶次谐波比、高频能量占比和谱熵等多维度特征的声纹标准化表征体系，经标准化预处理后的数据能有效提升模型性能，实现对直流偏磁、局部放电等质量缺陷的识别。【结论】本研究为变压器声纹数据提供了标准化处理框架与高精度识别模型，对提升电力设备运维质量具有重要推动意义。

Abstract:

[Objective] To address the problem of the inconsistent data quality and weak model generalization, this study aims to investigate acoustic data standardization methods and construct a deep learning-based quality detection model to support non-destructive testing and intelligent maintenance of power transformers. [Methods] By analyzing the characteristics of transformer acoustic signals and the bottlenecks in defect detection, a standardized process covering signal acquisition, noise reduction, and feature extraction is established to improve data quality and consistency. A deep learning model based on a CNN-Transformer hybrid architecture is introduced to identify multiple typical defects. [Results] A standardized acoustic characterization system is established, encompassing multi-dimensional features such as sound pressure level, signal-to-noise ratio, odd-even harmonic ratio, high-frequency energy ratio, and spectral entropy, which can effectively enhance model performance, enabling accurate identification of quality defects such as DC bias and partial discharge. [Conclusion] This research provides a standardized processing framework and a high-precision recognition model for transformer acoustic data, contributing significantly to improving the quality of power equipment maintenance.

引用格式：王童,王正,安丰柱.基于声纹数据标准化的变压器质量缺陷检测研究[J].标准科学,2026(1):71-79.

基金项目：本文受国家电网公司总部科技项目“电力设备智慧巡检与精准作业机器人关键技术研究”（项目编号：5108-202218280A-2-249-XG）资助。

作者简介：王童，通信作者，博士，高级工程师，研究方向为信息与自动化技术标准化。