基于机器学习预测再生骨料混凝土抗压强度与弹性模量的研究

Prediction of Compressive Strength and Elastic Modulus of Recycled Aggregate Concrete Based on Machine Learning

由于成分的复杂性及来源的不确定性,再生骨料混凝土的力学性能与普通混凝土相比有较大差别。为了更准确地把握再生骨料混凝土的基本力学性能,基于BP神经网络及ANFIS自适应神经模糊推断系统分别建立了两种软计算模型,对再生骨料混凝土的抗压强度与弹性模量进行预测。以试件形状、尺寸、有效水灰比、骨料水泥比、再生粗骨料取代率、再生粗骨料与天然粗骨料各自的最大粒径、体密度与吸水率等11个参数作为输入变量,以再生骨料混凝土的抗压强度或弹性模量作为输出变量,利用统计的146组试验数据对模型进行训练及测试,并将预测值与实测值进行对比。研究结果表明:两种模型均能较准确地预测再生骨料混凝土的抗压强度及弹性模量;相比BP模型,ANFIS模型的预测精度稍有提高,但计算效率明显偏低,因此在后续研究中需对其进行优化。

Due to its constituent complexity and source uncertainty,recycled aggregate concrete(RAC)differs from natural aggregate concrete(NAC)in terms of many mechanical properties.To better estimate the fundamental mechanical properties of RAC,two soft comput‐ing models based respectively on the back-propagation neural network(BP)and the adaptive neural fuzzy identification system(ANFIS)are developed to evaluate the compressive strength and elastic modulus of RAC.A total of eleven variables,including the specimen shape,speci‐men size,effective water-to-cement ratio,aggregate-to-cement ratio,replacement ratio of recycled aggregates(RAs)in lieu of natural aggre‐gates(NAs),and nominal maximum size,bulk density,and water absorption ratio of RA and NA,are considered as the input parameters,while the compressive strength or elastic modulus of RAC is taken as the final output.A test database made up of 146 data sets collected from the published literatures is used to train and test the two models separately.The predictions by the two models are compared with the test results.It is indicated that both prediction models are capable of predicting with good accuracy the compressive strength and elastic mod‐ulus of RAC.Comparatively,the ANFIS method performs slightly better than the BP model,yet the former's computational efficiency is markedly lower as opposite to the latter.Thus,the ANFIS method is necessary to be properly optimized in future studies.