The Research On Aggregate Size And Shape Based On Convolution Neural Network

Concrete is an important building material for national infrastructure construction,and fine aggregate is one of the main raw materials of concrete.Its particle size and shape have a significant impact on various properties of concrete.High-performance concrete requires fine aggregates of good quality,and fine aggregates of good quality must have a reasonable particle size.In the industry,the particle size distribution(grading)detection is time-consuming and laborious,and the geometrical size parameters of the aggregate cannot be directly obtained.The aggregate particle shape detection instrument is complicated,and only some indicators are measured indirectly,and the particle shape is not directly measured.The quality inspection of fine aggregate only stays in the macro factor evaluation.Therefore,based on the geometric parameter data extracted from the three fine aggregate images of 0.8-2.0mm,the paper studies the aggregate gradation,grain shape and quality.Firstly,the aggregate grading is tested by analogy with the screening method in the industry,and the particle grading is divided by the size parameters of the image.The cumulative mass in the cumulative area and cumulative volume equivalent screening method is used to obtain the image grading.Through the correction factors,the image grading and the screen grading are in good consistency.Secondly,feature selection is introduced to analyze the particle shape features of 12 2-D images of fine aggregate,and a new feature subset evaluation function MMFS is proposed based on mutual information and maximum information coefficient.The experimental results show that the accuracy of this method is improved by 3.8% compared to MIFS.Then,through the random forest feature selection model,the candidate feature subsets obtained by the MMFS algorithm are selected to obtain the 6 most important granular shape features.Taking the importance of the 6 most important grain shape features as weights,the formula of the grain shape synthesis coefficient is defined: grain shape synthesis coefficient = ?(grain shape feature × weight),the smaller the grain shape synthesis coefficient,the better the aggregate grain shape.Finally,particle size is an important attribute of aggregate quality supervision.Traditional methods evaluate aggregate quality with macro factors,resulting in the industry has not yet figured out the classification criteria for the mapping of aggregate particle size to aggregate quality.To solve this problem,based on the convolutional neural network,a classification model of fine aggregate particle size and shape quality is proposed to realize the intelligent detection of fine aggregate quality.The experimental results show that the accuracy of the model on the test set and the verification set is more than 98%,and the model performance is better than LeNet5 and VGG16,which can provide a reference for the classification of fine aggregate quality in industry practice.