基于卷积神经网络图像处理的新拌混凝土强度识别

Identification of compressive strength of fresh concrete based on convolutional neural network image processing

混凝土抗压强度是评价混凝土结构质量和安全的代表性参数,但传统的评价混凝土抗压强度的方法多为后验方法,对已硬化后的混凝土进行抗压强度试验,此类方法成本高、浪费严重、耗时巨大且具有滞后性。因此,提出一种基于卷积神经网络的新拌混凝土抗压强度识别预测方法。根据不同特性(水灰比、骨料粒径、砂率)的混凝土具有不同的图像表面特征,对不同配合比情况下的新拌混凝土进行图像采集以及对卷积神经网络的改进来实现混凝土强度较为精准的识别预测,其中改进的卷积神经网络模型识别准确率高达97.94%。将该方法与现有的强度检测方法结合使用以提高混凝土抗压强度的检测效率和准确性。

Compressive strength of concrete is a representative indicator to evaluate the quality and safety of concrete structures.However,the traditional methods for evaluating the compressive strength of concrete are mostly posterior methods,which carried out the compressive strength test of hardened concrete,these kind of methods have the characteristics of costly,wasteful,time consuming and hysteretic.This paper presents a new method for identify and predict the compressive strength of freshly mixed concrete based on the convolutional neural network.According to different characteristics(water-cement ratio,nominal particle of aggregate,sand-aggregate ratio)of concrete has different image surface characteristics,through image acquisition of fresh concrete in different mix proportion and improvement of convolutional neural network to realize accurate identification and prediction of concrete strength that the identification accuracy of the improved convolutional neural network model is as high as 96.94%.This method is combined with the existing concrete strength testing methods to improve the testing efficiency and accuracy of concrete compressive strength.