基于智能颗粒的冷再生混合料压实状态感知

Cold-recycled Mixture Compaction State Sensing Based on SmartRock

为了更好地揭示压实过程中混合料颗粒间的相互作用,实现优化压实工艺与混合料设计的目的,基于智能颗粒传感装置开展了冷再生沥青混合料室内旋转压实试验。监测混合料在旋转压实过程中的受力、转角及加速度动态响应规律;研究智能颗粒动态响应规律与混合料相对压实度之间的关联性;揭示压实过程中颗粒间的互锁机制,并对混合料的压实状态做出评价。开展了不同粗骨料含量下的冷再生混合料旋转压实试验,研究不同粗骨料含量对压实过程的影响。试验结果表明:旋转压实过程中混合料相对压实度的变化规律与智能颗粒的动态响应有较好的相关性,初步实现旋转压实过程中混合料压实状态的智能感知;与45%和65%粗骨料含量的混合料相比,55%粗骨料含量的混合料更容易达到较高压实度,且在压实过程中的稳定性较高。此外,建立了基于改进的人工神经网络的压实度预测模型,选取智能颗粒实测的x、y方向的相对旋转角和加速度,z方向的接触力及集料在各筛孔的通过率作为模型的输入变量,采用Tensorflow2.0框架构建网络、循环神经网络结构与全连接深度神经网路结构,分别处理序列特征数据与连续特征数据,并通过拼接操作进行数据融合训练。经过600次迭代训练后,网络的损失函数和精度达到了收敛阈值,混合料相对压实度的实测值与预测值拟合直线方差为0.93,结果表明该模型对于混合料的相对压实度具有较好的预测能力。

To better reveal the interaction among mixture particles in the compaction process,and realize the purpose of optimizing compaction process and mixture design,the gyratory compaction test of cold-recycled asphalt mixture was carried out.The SmartRock sensing device was used to monitor the dynamic response rule of the force,rotation angle and acceleration of particles in the gyratory compaction process.The correlation between SmartRock dynamic response and relative compaction degree of mixture was studied.The interlocking mechanism between among particles during the compaction process was reveal,and the mixture compaction state was evaluated.The gyratory compaction test of cold-recycled mixture with different coarse aggregate contents was carried out to study the influence of different coarse aggregate contents on the compaction process.The test result indicates that there is a good correlation between the variation rule of relative compaction degree and the dynamic response of SmartRock during the compaction process.The intelligent perception of mixture compaction state during the gyratory compaction process was initially realized.Compared with the mixture with 45%and 65%coarse aggregate contents,the mixture with 55%coarse aggregate content is easier to achieve a better compaction degree,and has a higher stability during the compaction process.In addition,the prediction model of compactness based on improved artificial neural network was established.The relative rotation angle and acceleration in x and y directions,the contact force in z direction measured by SmartRock,and the passage rate of aggregate in each sieve hole were selected as the input variables of model.The Tensorflow2.0 framework was used to construct the network and the cyclic neural network structure.The sequential feature data and continuous feature data were processed respectively,and the data fusion training was carried out through the splicing operation.The loss function and accuracy of network reach the convergence threshold after 600 training iterations.The linear variance of measured value and predicted value of the relative mixture compaction was 0.93,indicating that the model has a good prediction ability for the relative mixture compactness.