基于遗传算法改进的BP神经网络模型的磨损机制智能识别-论文.pdf

《基于遗传算法改进的BP神经网络模型的磨损机制智能识别-论文.pdf》由会员上传分享，免费在线阅读，更多相关内容在应用文档-天天文库。

1、2014年1月润滑与密封Jan．2014第39卷第1期LUBRICATIONENGINEERINGVo1．39No．1DOI：10．3969／j．issn．0254—0150．2014．01．005基于遗传算法改进的BP神经网络模型的磨损机制智能识别盛晨兴程俊李文明段志和马奔奔(1．高性能船舶技术教育部重点实验室(武汉理工大学)湖北武汉430063；2．武汉理工大学能源与动力工程学院湖北武汉430063；3．西安交通大学润滑理论及轴承研究所陕西西安710049；4．交通运输部南海救助局广东广州510310)摘要：通过提取磨粒形状特征参数、颜色特征参数和表面纹理等特征参数对磨粒形态进行量化表征，

2、并以此为输入矢量，引入遗传算法(GA)改进BP神经网络对磨粒进行自动分类识别，建立遗传算法改进的BP神经网络模型，并给出具体的算法实现过程。分别应用遗传算法改进的BP神经网络模型和未引入遗传算法改进的BP神经网络模型对磨粒图像进行智能识别。实验结果表明，遗传算法改进的BP神经网络综合了遗传算法的全局优化和BP算法局部搜索速度快的特点，网络识别率较高，具有较好的全局性。关键词：特征提取；磨粒识别；遗传算法；BP算法；神经网络中图分类号：THIl7．1；TP301．6文献标识码：A文章编号：0254—0150(2014)1—024—5IntelligentRecognitionofWearMech

3、anismBasedonImprovedBPNeuralNetworkModelbyGeneticAlgorithmShengChenxing，ChengJun，LiWenmingDuanZhihe，MaBenben·(1．KeyLaboratoryofHighPerformanceShipTechnologyofMinistryofEducation(WuhanUniversityofTechnology)，WuhanHubei430063，China；2．SchoolofEnergyandPowerEngineering，WuhanUniversityofTechnology，WuhanH

4、ubei430063，China；3．TheoryofLubricationandBearingInstitute，Xi’anJiaotongUniversity，Xi’anShaanxi710049，China；4．NanhaiRescueBureauoftheMinistryofTransport，GuangzhouGuangdong510310，China)Abstract：Aimprovedbackpropagation(BP)neuralnetworkbyGeneticalgorithmwasintroducedtorealizetheauto—maticclassification

5、andrecognitionofweardebris，basedonthequalitativecharacterizationofthemorphologicalfeaturesoftheweardebrismakinguseofthecharacteristicparametersofweardebrisshape，color，andsurfacetexture．Aneuralnet—workmodelbasedontheimprovedbackpropagation(BP)neuralnetworkbyGeneticalgorithmwasestablishedtoclassifyand

6、recognizetheweardebrisusingthoseparametersastheinputvectors．Thealgorithmoftheestablishedmodelwasde—tailed．BycomparingtheresultsofautomaticrecognizingtheweardebrisbytheimprovedBPneuralnetworkandthepres—entedBPneuralnetwork，itshowsthattheimprovedbackpropagation(BP)neuralnetworkcombinestheglobaloptimiz

7、a—＼p＼qp8tionfeatureofgeneticalgorithmandthefastspeedfeatureinlocalsearchofBPalgorithm，whichhasahighrecognitionrateandbetterglobalsearchfeature．Keywords：characterizationextracting；wearrecognition；Genet