面向机械振动信号的自主信号处理大语言模型智能体

打开文本图片集

关键词：大语言模型；神经符号；可信人工智能；多智能体；决策链；故障诊断 中图分类号： TH165+ 3 文献标志码：A DOI：10.16385/j.cnki.issn.1004-4523.202508034

An LLM-based agent for autonomous signal processing mechanical vibration signals

LI Qi¹²， ZHANG Xinran1，HU Wenyang'，ZHANG Feibin1，QIN Zhaoye1， CHUFulei' （1. University，Beijing1oo084，China； 2. Haven O6511，USA）

Abstract：Vibrationsignalaalysisisaoerstonemachineconditionmoitoringfaultdiagnosis，yetitfacesacentraldilma Tradtionalexpertsytemsverigidwokflows，whileendtonddepmodels，despiteteiradaptiveleagabilts，suferrobeing ‘blackboxes’withinsuicientreproducibilityTispaperintroducesaneuro-symbolicmulti-agentframeworkforautonomoussigal procesing.Theframeworkutilizesalargelanguagemodel（LLMasacentraldecision-maker，coordinatingatoboxinterpretable， symbolicsignalprocsingperatosteableatonoousvibratioignalaalysisnddagnsisTheframeorkadoptsaaEeute ReviewmultiagtcitetutieatielytizetesigalprocesigdisionchainTsuetegicalosistectelaing preventicoctoeatorall，allatseallyuatedsoimesildanticrafoatooe Specifically，teyareconstraiedbysmantiifoationfortheMtocompreendValidationonbaringfaultdiagnosisdatasetsshows thathisframeorkanautoomouslygeneatesigalprocessngdecisionchainsithlearpsicalmeanngassuefullypoduced expert-evel，intepretabledagostcalgoritssuchas‘veloespectru-kurtosis’Insingle-doaintestsntegaUesity bearingdataset，theGemini-2.5-pro versionreached anaccuracy 97.8% .Incross-domain testson theUniversity Ottawa variable-speed dataset，the framework，trained solely on‘acceleration’‘deceleration’conditions，achieved 99.3% accuracy on unseen conditions， provingitsgeneralizationabilityThisreseachprovidesapromsingneparadigmforbuildingtrustworthy，eproducible，salableext generation intelligent diagnostic systems.

Keywords：largelanguagemodel;neuro-symbolic;trustworthyAI;multi-agent;decisionchain;faultdiagnosis

振动信号分析是机械设备状态监测与故障诊断的核心技术之一。（剩余19967字）