神经网络在高耐久性混凝土骨料优选中的应用
发布时间：2018-09-05 18:06
【摘要】：粗骨料占混凝土体积的50%以上,来源广泛,质量难以控制。其质量直接影响混凝土的耐久性,有必要对其进行优选。以往,人们研究粗骨料对混凝土耐久性的影响大多是通过试验进行的,试验研究费时费力。因此,有必要引入一种先进的方法来预测粗骨料对混凝土耐久性的影响,从而提高效率。粗骨料与混凝土耐久性之间的关系是非线性的,很难建立明确的数学表达式,本文将人工神经网络理论引入到混凝土的研究之中。本文通过选用不同技术指标(平均粒径、吸水率、表观密度、强度)的典型骨料,研究其对混凝土抗渗、抗冻、抗碳化性的影响,结果表明:当水灰比为0.32时,粒径(5~30 mm)较大的骨料配制的混凝土的抗渗性最好;当水灰比为0.40和0.49时,粒径(5~20 mm)适中的骨料所配制的混凝土的抗渗性最好。粒径对混凝土抗碳化性的影响规律与其对抗渗性的影响规律类似,对抗冻性的影响不是很明显。对于不同品种骨料配制的混凝土的抗渗性而言,均是玄武岩配制的混凝土最好。处于低水灰比(0.32)时,石灰岩配制的混凝土的抗渗性最差;在其它水灰比(0.40和0.49)时,花岗岩配制的混凝土的抗渗性最差。骨料品种对混凝土抗冻、抗碳化的影响规律与其对抗渗性的影响规律相同。本文基于以上混凝土耐久性试验所得数据,建立了混凝土耐久性的神经网络预测模型。在建立模型之前,本文对较为常用的四种优化的神经网络算法(弹性梯度下降法、附加动量法、自适应学习率算法、L-M算法)进行比较,发现L-M算法无论是在训练时间还是在预测误差和收敛性方面都具有其它三种算法无法比拟的优势,本文选择L-M算法进行训练。综合分析粗骨料对混凝土耐久性的影响因素之后,选定水灰比、龄期、平均粒径、吸水率、表观密度和强度6个因素作为混凝土抗渗性和抗碳化性预测模型的输入变量,分别以氯离子扩散系数、碳化深度作为输出变量,基于108组抗渗数据和135组抗碳化数据,分别建立结构为6-17-1和6-15-1的混凝土抗渗性与抗碳化性预测模型。选定水灰比、冻融循环次数、平均粒径、吸水率、表观密度和强度6个因素作为混凝土抗冻性预测模型的输入变量,以相对动弹性模量作为输出变量,基于抗冻试验所得103组数据建立结构为6-21-1的混凝土抗冻性预测模型。运用以上建立的混凝土抗渗、抗冻、抗碳化性预测模型对测试样本进行预测,平均预测误差分别为4.44%、4.15%、5.16%,均在6%以内,预测值与试验值非常接近,可以满足实际工程的需要。为了验证建立的模型的适用性,本文以抗渗、抗冻性预测模型为例,收集了一些工程中相关数据,利用已建立的神经网络预测模型分别对其进行预测。预测误差分别为11.00%、9.85%,比本文对试验数据的预测误差稍大一些,但是可以满足工程要求。
[Abstract]:Coarse aggregate accounts for more than 50% of the volume of concrete. The quality of concrete directly affects the durability of concrete, it is necessary to select it. In the past, the influence of coarse aggregate on concrete durability was mostly studied through experiments, which were time-consuming and laborious. Therefore, it is necessary to introduce an advanced method to predict the effect of coarse aggregate on the durability of concrete so as to improve the efficiency. The relationship between coarse aggregate and concrete durability is nonlinear, it is difficult to establish a clear mathematical expression. In this paper, the artificial neural network theory is introduced into the study of concrete. In this paper, the effects of different technical indexes (average particle size, water absorption, apparent density and strength) on the impermeability, freezing resistance and carbonation resistance of concrete are studied. The results show that when the water-cement ratio is 0.32, When the water-cement ratio is 0.40 and 0.49, the impermeability of concrete with moderate particle size (5 ~ 20 mm) is the best. The effect of particle size on the carbonation resistance of concrete is similar to that on the impermeability of concrete, but the effect on freezing resistance is not obvious. For the impermeability of concrete prepared with different kinds of aggregate, the concrete made of basalt is the best. When the water-cement ratio is low (0.32), the impermeability of concrete prepared by limestone is the worst, and the impermeability of concrete prepared by granite is the worst when the other water-cement ratio (0.40 and 0.49). The effect of aggregate variety on the frost resistance and carbonation resistance of concrete is the same as that of its impermeability. Based on the above data of concrete durability test, a neural network prediction model of concrete durability is established in this paper. Before establishing the model, four kinds of optimization neural network algorithms (elastic gradient descent method, additional momentum method, adaptive learning rate algorithm) are compared in this paper. It is found that L-M algorithm has incomparable advantages in terms of training time, prediction error and convergence. This paper chooses L-M algorithm to train. After synthetically analyzing the influence factors of coarse aggregate on concrete durability, six factors, such as water-cement ratio, age, average particle size, water absorption, apparent density and strength, are selected as input variables of the prediction model of concrete impermeability and carbonation resistance. Taking chloride diffusion coefficient and carbonation depth as output variables, based on 108 groups of anti-seepage data and 135 groups of anti-carbonation data, the prediction models of anti-permeability and anti-carbonation of concrete with structures of 6-17-1 and 6-15-1 were established, respectively. Six factors, such as water-cement ratio, freeze-thaw cycle times, average particle size, water absorption, apparent density and strength, are selected as input variables and relative dynamic modulus of elasticity as output variables. Based on 103 sets of data obtained from frost resistance test, a concrete frost resistance prediction model with structure of 6-21-1 was established. The prediction model of concrete impermeability, frost resistance and carbonization resistance established above is used to predict the test samples. The average prediction error is 4.44 / 4.15 / 5.16, which is less than 6% respectively. The predicted value is very close to the test value, which can meet the needs of practical engineering. In order to verify the applicability of the established model, the prediction model of impermeability and frost resistance is taken as an example, and some relevant data in engineering are collected, and the established neural network prediction model is used to predict the model respectively. The prediction error is 11.00 and 9.85, which is a little larger than that of the test data in this paper, but it can meet the engineering requirements.
【学位授予单位】：石家庄铁道大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TU528.041
本文编号：2225045
[Abstract]:Coarse aggregate accounts for more than 50% of the volume of concrete. The quality of concrete directly affects the durability of concrete, it is necessary to select it. In the past, the influence of coarse aggregate on concrete durability was mostly studied through experiments, which were time-consuming and laborious. Therefore, it is necessary to introduce an advanced method to predict the effect of coarse aggregate on the durability of concrete so as to improve the efficiency. The relationship between coarse aggregate and concrete durability is nonlinear, it is difficult to establish a clear mathematical expression. In this paper, the artificial neural network theory is introduced into the study of concrete. In this paper, the effects of different technical indexes (average particle size, water absorption, apparent density and strength) on the impermeability, freezing resistance and carbonation resistance of concrete are studied. The results show that when the water-cement ratio is 0.32, When the water-cement ratio is 0.40 and 0.49, the impermeability of concrete with moderate particle size (5 ~ 20 mm) is the best. The effect of particle size on the carbonation resistance of concrete is similar to that on the impermeability of concrete, but the effect on freezing resistance is not obvious. For the impermeability of concrete prepared with different kinds of aggregate, the concrete made of basalt is the best. When the water-cement ratio is low (0.32), the impermeability of concrete prepared by limestone is the worst, and the impermeability of concrete prepared by granite is the worst when the other water-cement ratio (0.40 and 0.49). The effect of aggregate variety on the frost resistance and carbonation resistance of concrete is the same as that of its impermeability. Based on the above data of concrete durability test, a neural network prediction model of concrete durability is established in this paper. Before establishing the model, four kinds of optimization neural network algorithms (elastic gradient descent method, additional momentum method, adaptive learning rate algorithm) are compared in this paper. It is found that L-M algorithm has incomparable advantages in terms of training time, prediction error and convergence. This paper chooses L-M algorithm to train. After synthetically analyzing the influence factors of coarse aggregate on concrete durability, six factors, such as water-cement ratio, age, average particle size, water absorption, apparent density and strength, are selected as input variables of the prediction model of concrete impermeability and carbonation resistance. Taking chloride diffusion coefficient and carbonation depth as output variables, based on 108 groups of anti-seepage data and 135 groups of anti-carbonation data, the prediction models of anti-permeability and anti-carbonation of concrete with structures of 6-17-1 and 6-15-1 were established, respectively. Six factors, such as water-cement ratio, freeze-thaw cycle times, average particle size, water absorption, apparent density and strength, are selected as input variables and relative dynamic modulus of elasticity as output variables. Based on 103 sets of data obtained from frost resistance test, a concrete frost resistance prediction model with structure of 6-21-1 was established. The prediction model of concrete impermeability, frost resistance and carbonization resistance established above is used to predict the test samples. The average prediction error is 4.44 / 4.15 / 5.16, which is less than 6% respectively. The predicted value is very close to the test value, which can meet the needs of practical engineering. In order to verify the applicability of the established model, the prediction model of impermeability and frost resistance is taken as an example, and some relevant data in engineering are collected, and the established neural network prediction model is used to predict the model respectively. The prediction error is 11.00 and 9.85, which is a little larger than that of the test data in this paper, but it can meet the engineering requirements.
【学位授予单位】：石家庄铁道大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TU528.041
【参考文献】
相关期刊论文 前2条
1 万广培;李化建;黄佳木;;混凝土内养护技术研究进展[J];混凝土;2012年07期
2 王爱勤,张承志;水工混凝土的碱骨料反应问题[J];水利学报;2003年02期
，本文编号：2225045
本文链接：https://www.wllwen.com/kejilunwen/huagong/2225045.html
