基于深度学习的SCC工作性能预测技术

SCC Performance Prediction Technology Based on Deep Learning

基于深度学习技术研究了一种在自密实混凝土(SCC)出机前根据拌合物图像信息预测其工作性能的方法。录制了25组不同工作性能的SCC搅拌过程视频,按坍落扩展度、T_(500)实测值并结合目测将其划分为合格、流动性不足及离析三种工作性能,通过将视频分帧处理为图像集,采用图像分类和目标检测两种方法建立深度学习模型,通过模型对拌和物图像特征的学习及训练,完成对SCC工作性能的预测。结果表明,图像分类和目标检测两种方法在验证集上均可达到98%以上的准确率,可为实时调整配合比进而实现SCC智能化提供参考。

The deep learning technology was used to study a method of predicting the performance of SCC based on the mixture image information during the mixing process.Twenty-five sets of videos of the SCC mixing process with different performances were recorded.According to the slump flow and T_(500)measured values and combined with the visual inspection,the SCC mixes were classified into three performances:qualified,insufficient fluidity and segregation.By processing the videos into image sets,the deep learning models were built using image classification and target detection respectively.The models learn and train the image features of the mixes to realize the prediction of SCC performance.The results show that both image classification and target detection methods can achieve more than 98%accuracy on the validation set,which provides a reference for adjusting the mix proportion in real-time and realizing the smart production of SCC.