Research On Automatic Temperature Interpretation Method Of Thermal Paint Images

The working temperature of an aero engine has a significant impact on its performance.Using thermal paint based temperature measurement technology to measure the temperature of its components is an essential key step in the engine test process.The manual interpretation method of thermal paint has poor stability and low accuracy,thus it is necessary to realize the automatic temperature interpretation of thermal paint.At present,the automatic temperature interpretation method based on thermal paint image mainly has the following two problems: First,the color distortion of thermal paint image caused by the lighting and manual operation seriously affects the accuracy of temperature interpretation;Second,the current automatic temperature interpretation method of thermal paint image only establishes a single universal interpretation model,and its interpretation accuracy can’t meet the needs of practical applications.How to use image processing,machine learning and deep learning methods to achieve automatic temperature interpretation model under various interference factors with higher accuracy is a hot topic of current research.This thesis investigates the status of art of automatic temperature interpretation technology of thermal paint,analyzes the reasons that affect the temperature interpretation accuracy,and focuses on the pre-processing method of thermal paint image and the temperature interpretation model based on the image characteristics of thermal paint to improve the interpretation precision.First,based on image processing and deep learning algorithms,this thesis proposes thermal paint image preprocessing methods including color cast correction,brightness correction,image smoothing and background segmentation.These methods mostly solve the problem of color distortion of thermal paint image,and are validated by comparative experiments.Then,based on the discoloration law and characteristics of thermal paint,the clustering method for thermal paint temperature interval division is studied,including two clustering segmentation algorithms,the one is a fast clustering method based on finding the density peak point,and the other is a circular elimination algorithm based on the proportion of membership.Finally,in different color spaces,the effects of temperature interpretation methods based on different features and different classifiers are analyzed,and on the basis of these comparative experiments,the step-by-step interpretation model based on the discoloration law of thermal paint is proposed.In this thesis,some comparative experiments of multiple indicators such as overall temperature interpretation accuracy,overall temperature average error and temperature category average error are performed on the image dataset of multiple thermal paints.In the temperature interpretation experiment,a variety of algorithms such as SVC and KNN are compared and analyzed.The step-by-step interpretation algorithm is compared with the best-available KNN algorithm,and the overall temperature average error on the thermal paints test dataset of KN3,KN6 and KN8 is reduced from 12.73 ℃ to 1.66 ℃,from 21.10 ℃ to 6.89 ℃,from 6.53 ℃ to 5.48 ℃.The effectiveness of the step-by-step interpretation model is verified,which provides technical support for the wide application of automatic temperature interpretation of thermal paint image.