Detection Method Of Thermal Engineering Defects In Exterior Thermal Insulation Buildings Based On Machine Learning

As the great influence on building energy consumption,thermal engineering defect detection is an important content of building energy conservation inspection.Especially in cold region,because of construction failure and periodic frost heave caused by climate,thermal engineering defects in the residential building are very common,and seriously affect the building energy consumption,indoor thermal comfort and security of retaining structure.However,the detection and evaluation methods of infrared thermography recommended in the current standard are of high workload,large error and high omission rate,which brings resistance to the universal application of regular thermal engineering defect detection of existing residential building.Therefore,the need for automatic detection of thermal defects in severe cold regions is very urgent.This paper took Harbin as a typical city in cold region and in as an example,the external thermal insulation structure was selected as the research object to study rapid identification method for external thermal defects of external thermal insulation structure based on a large number of residential building structure research.This paper preliminary get common type of thermal engineering defects through the research of thermal engineering defects of residential building envelope in a typical city Harbin in cold region.After analysis the thermal infrared image,this paper presented a thermal defects automatic detection system based on image segmentation and image recognition.The system can achieve automatic identification and area extraction of thermal defects after 6 steps of pretreatment,size correction,the suspicious area segmentation,defect classification and wall temperature calculation of the main wall and defect area extraction.In order to obtain the critical temperature gradient parameter in the detection system of suspicious area and related inspection matters needing attention,transient CFD simulation was taken to study the surface temperature distribution characteristics under the change of defect parameters.Eventually four traditional machine learning methods,PCA-BP,PCA-SVM,gray histogram-BP,gray histogram-SVM and the two deep learning methods,convolution neural network based on Le Net-5 and transfer learing based on Inception-V3 was trained to classify the temperature anomaly area in the infrared image.By analyzing the results,the classification method suitable for building thermal defects was obtained.The results show that the highest recognition rate of PCA-SVM is 96% in the traditional machine learning method,and the accuracy rate of the two methods of deep learning is 95%,of which the transfer learning has the advantages of fast learning speed and high accuracy.The results of this study have a certain theoretical guiding significance for the attention and determination parameters of thermal defects detection,and provide a new idea and method for thermal defect detection in building construction acceptance,energy saving detection and existing construction inspection.