【作者】 李鹏飞；

【导师】 张胜寒；

【作者基本信息】 华北电力大学 ， 环境工程（专业学位）， 2023， 硕士

【摘要】 作为用电和发电大国,变电站和输电线铁塔等钢结构因环境腐蚀浪费严重。为了延长以上设施的使用寿命,减少经济损失,采用涂装防腐技术已成为操作方便、经济实惠的首选方案。在借鉴了锌牺牲阳极保护阴极的作用,涂料中加锌已然成为涂层防腐的主要手段之一。现如今,富锌底漆的多样性和质量良莠不齐的局面依然存在,以次充好的现象也时有发生。为防止经济浪费,需要有效鉴定不同含锌量的富锌底漆,并通过定性分析对样品分类识别。近年来,红外光谱分析技术因其分析速度快,不破坏完整性以及无污染等优点在各种检测领域被广泛应用。本研究使用40%和70%的富锌底漆、云铁中间漆和聚氨酯面漆制成的干膜样品,采用标准化学方法检测含锌的含量,并结合傅里叶红外光谱仪测定样本的特征信息,结合化学计量学方法进行漆膜样品的红外光谱识别和分析。本文重点探讨了红外光谱技术结合化学计量学模型建立的几个关键问题,包括光谱预处理、波长筛选和校正模型的建立等。为研究漆膜样品中不同含锌量的富锌底漆的无损检测提供了新的参考,采用校正模型的预测准确率评估了红外光谱在辨别不同富锌底漆含锌量方面的应用。本文研究了漆膜样品光谱数据处理方法。针对波峰差异,将数据分为六个波数段,并研究了不同预处理方法及组合对数据处理的效果。结果表明,经过二阶导数处理的预处理方法最佳。在最佳预处理条件下比较了各波段采用蒙特卡洛、以及其他选择方法在建立模型时的效果。经比较,PCA模型的鉴别率超过了70%,同时也进行了PCR和PLSR的比较分析,PLSR比PCR需要更少的成分就能获得同样的预测准确度。运用偏最小二乘(PLS)算法来降维度,并实现光谱样本分类和辨别,仿真实验结果显示,准确率高达90.30%。采用BP神经网络相结合的红外光谱分析法,并利用K折和双交叉验证法对模型进行验证,其预测均方根误差和百分比误差明显低于化学计量学方法建立的模型。本文使用红外光谱结合化学计量学进行模型计算,运用PLS和BP神经网络以及PCA主成分分析方法。通过建立富锌底漆样品含锌量的红外光谱预测模型,得出的预测均方根误差均在0.8以上,相关系数最高达0.99,证明红外技术识别富锌底漆方面具有可行性。更多还原

【Abstract】 As a big power consumer and power generation country,steel structures such as substations and transmission line towers are seriously wasted due to environmental corrosion.In order to prolong the service life of the above facilities and reduce economic losses,the use of coating anti-corrosion technology has become the first choice for convenient operation and economic benefits.Drawing on the function of zinc sacrificial anode to protect the cathode,adding zinc to the coating has become one of the main means of coating anticorrosion.Nowadays,the diversity and quality of zincrich primers still exist,and the phenomenon of shoddy ones also occurs from time to time.In order to prevent economic waste,it is necessary to effectively identify zincrich primers with different zinc contents,and classify and identify samples through qualitative analysis.In recent years,infrared spectroscopic analysis technology has been widely used in various detection fields due to its advantages of fast analysis speed,no damage to integrity and no pollution.In this study,dry film samples made of 40% and 70% zincrich primers,mica iron intermediate paints,and polyurethane topcoats were used to detect the zinc content by standard chemical methods,and combined with Fourier transform infrared spectroscopy to determine the characteristics of the samples Information,combined with chemometric methods for infrared spectrum identification and analysis of paint film samples.This paper focuses on several key issues in the establishment of infrared spectroscopy combined with chemometric models,including spectral preprocessing,wavelength screening,and establishment of calibration models.To provide a new reference for the study of non-destructive testing of zinc-rich primers with different zinc content in paint film samples,the application of infrared spectroscopy in distinguishing zinc content of different zinc-rich primers was evaluated using the prediction accuracy of the calibration model.In this paper,the method of processing spectral data of paint film samples is studied.According to the peak difference,the data is divided into six wavenumber segments,and the effects of different preprocessing methods and combinations on data processing are studied.The results show that the preprocessing method processed by the second derivative is the best.Under the optimal preprocessing conditions,the effects of Monte Carlo and other selection methods in building models were compared for each band.After comparison,the identification rate of the PCA model exceeded70%.At the same time,a comparative analysis of PCR and PLSR was also carried out.PLSR requires fewer components than PCR to obtain the same prediction accuracy.The Partial Least Squares(PLS)algorithm is used to reduce the dimension,and realize the classification and identification of spectral samples.The simulation experiment results show that the accuracy rate is as high as 90.30%.The infrared spectrum analysis method combined with BP neural network was used,and the model was verified by K-fold and double cross validation.In this paper,infrared spectroscopy combined with chemometrics is used for model calculation,and PLS and BP neural network and PCA principal component analysis method are used.By establishing an infrared spectrum prediction model for the zinc content of zinc-rich primer samples,the root mean square errors of the predictions obtained are all above 0.8,and the correlation coefficient is as high as 0.99,which proves that infrared technology is feasible in identifying zinc-rich primers.更多还原