Study On The Principles And Methods Of Temperature Interpretation Of Thermal Paint By Spectrum

In engineering applications,the working temperature of the equipment is an important parameter.Thermal paint is often used for temperature testing of complex equipment such as aero engines,which is widely used in production and scientific research.As temperature changes,the color of thermal paint changes irreversibly.By comparing the color of the components with the templates,the highest temperature of the components can be measured.Currently,the temperature of thermal paint is mainly interpreted by technicians or by images.However,there are many disadvantages of image based interpretation,such as large image acquisition workload,high time cost,application limitation,susceptibility to external light,and low temperature interpretation accuracy.To overcome these shortcomings and improve the interpretation accuracy,this thesis explores the correspondence between the maximum temperature thermal paint experienced and its diffuse reflectance spectrum,and proposes a method to interpret the temperature by diffuse reflectance spectrum on the basis of analyzing the characteristics of thermal paint.The temperature interpretation method based on diffuse reflectance spectrum of thermal paint contains researches such as the acquisition of diffuse reflectance,preprocessing of spectral data,spectral feature extraction and the interpretation model.Firstly,a probe coupling fitting with Y-shaped reflection fiber is designed to collect diffuse reflectance.The spectral data of the standard templates is collected many times,from which the abnormal samples are removed by setting a threshold.The best preprocessing method is determined after comparing the preprocessing effect of spectral data by Savitzky-Golay filtering,Standard Normal Variable Transformation,Multivariate Scattering Correction and derivative.Based on the characteristics of the spectral curve,the feature extraction method for extracting banded statistical information is proposed,which is verified that the thermal paint’s spectral features of different temperatures have great discrimination with this method by comparative experiments.Finally,after comparing the interpretation error of the standard samples of some machine learning algorithms and model optimization methods such as K-Nearest Neighbors,Artificial Neural Networks,Partial Least Squares Regression,Support Vector Machines and Random Forests,a temperature interpretation model based on Random Forest and Adaboost method is proposed,with good generalization performance and high accuracy.It is showed that the root mean square errors of temperature interpretation on standard samples of KN3,KN6 and KN8 based on diffuse reflectance spectrum of thermal paint are all less than 8 ℃,and the accuracy of 10 °C is more than 92%;The temperature interpretation errors on butterfly-shaped templates and the turbine blades are almost within ± 10 ℃,which is lower than that of image based interpretation method.