Evaluation Of Concrete Strength Grade And Corrosion Residual Strength Based On Faster RCNN

In engineering practice,the problems of substandard concrete strength grade and concrete corrosion and aging are common.How to accurately and quickly detect concrete strength grade and residual strength is of great engineering importance.The traditional methods for concrete strength grade detection have many drawbacks such as structural damage,complicated operations,expensive equipment and low accuracy,while the methods for residual strength of corroded concrete detection need high condition demands.In order to overcome these drawbacks,a method based on the Faster RCNN was proposed to evaluate the strength grade and residual strength of concrete.The main contents of the research are as follows:(1)The implementation principle of the Faster RCNN was introduced first,and further improvements are then made on the VGG16 network for feature extraction.The influences of three key parameters,including the anchor size,batch size and learning rate on the model identification accuracy were investigated.(2)For evaluation of the concrete strength grade,six groups of concrete specimens with strength grades of C20,C30,C35,C40,C50 and C55 were made.The pictures of microscopic surface features of concrete specimens were taken with digital microscope cameras from the third to the sixth month after the fabrication of concrete specimens.The pictures taken at the third and the forth month were used to make the datasets used to train the model.An optimal model for concrete strength grade detection was obtained through parameter analysis,which achieved a mean average precision(m AP)of 99.07%.The optimal model was then used to detect new pictures taken at the fifth and the sixth month,achieving a m AP of 91.13% and 89.49 % respectively.(3)For evaluation of residual strength of corroded concrete,the textile-reinforced concrete(TRC)was adopted in the study.Twenty TRC specimens were fabricated and divided into five groups randomly.Five different corrosion degrees of TRC specimens were obtained by immersing them into the water tank for different days.The microscopic surface features of the specimens with different corrosion degrees were taken with digital microscope cameras and used to make datasets.An optimal model was trained with prepared datasets through parameter analysis,which achieved a m AP of 98.98% for residual strength detection of corroded TRC.The robustness of the optimal model was verified with new pictures.The results from the study indicate the effectiveness and feasibility of using the deep learning method to evaluate the concrete strength grade and corrosion residual strength,which provides a good reference for solving actual relevant engineering problems.