基于声纹数据标准化的变压器质量缺陷检测研究

打开文本图片集

Abstract：[Objective]Toaddresstheproblemof the inconsistentdata qualityand weak model generalization，this study aimsto investigate acousticdatastandardization methods and constructa dep learning-based quality detectionmodel to support non-destructive testingand intellgent maintenance of power transformers.[Methods]By analyzing the characteristicsof transformeracoustic signalsandthe botlenecks indefectdetection，astandardized processcovering signalacquisition，noisereduction，andfeatureextractionisestablished toimprovedataqualityandconsistency.A deeplearningmodelbasedonaCNN-Transformer hybridarchitecture isintroducedto identifymultiple typicaldefects. [Results]Astandardizedacousticcharacterizationsystemis established，encompasingmulti-dimensionalfeatures such as soundpresure level，signal-tonoiseratio，od-evenharmonicratio，high-frequencyenergyratio，andspectratropy which can efectively enhance model performance，enabling accurate identification ofquality defects such as DC bias andpartialdischarge.[Conclusion]Thisresearch providesastandardizedprocesing frameworkandahigh-precision recognitionmodelfor transformeracousticdata，contributing significantlytoimproving thequalityofpower equipment maintenance.

Keywords： power transformer;acoustic fingerprint;data standardization;qualitydefects;detection

0 引言

电力变压器作为电网的核心设备，其运行状态直接关系到整个电力系统的安全、稳定与可靠性。（剩余9839字）