Prediction Of Foamed Concrete Compression Strength And Thermal Conductivity Based On BP Neural Network

Thermal conductivity and compressive strength of foamed concrete are two important indexes in engineering application.Two of them is coupling relationship.At present, the performance prediction of the foamed concrete is mainly based on the mathematical model and the finite element model, which is mainly through the hole structure.Many valuable results have been achieved, but still do not have a recognized mature forecasting system.In this paper,we try to use the theory of artificial neural network to predict the performance of foamed concrete,using the good generalization ability of foamed concrete and the main parameters of mix ratioto obtain the prediction results. The research contents are mainly composed of the following aspects:Firstly,We introduce the foamed concrete, neural network, pore structure and thermal conductivity, pore structure and compressive strength etc,and introducing several common models of thermal conductivity of composite materials.Secondly,The experimental data were pretreated by the normalized function,then the standard gradient descent algorithm, adaptive learning rate algorithm and Levenberg Marquardt training algorithm, respectively, using the experimental data to network training for five times. Three different algorithms in training are compered by training time and fitting accuracy of performance advantages and disadvantages.Through the performance of three different algorithms of time average value and the fitting accuracy of average value, the Levenberg Marquardt training algorithm has shorter training time and higher precision. Therefore, this paper selects the Levenberg Marquardt algorithm as the training algorithm of BP neural network model.The results showed that when the amount of data and the number of hidden neurons are small, the error of Levenberg-Marquardt training algorithm is not greater than 30%, and the average convergence speed of training time is less than 0.01 seconds.Thirdly,the BP neural network model is established by using the Matlab,and selected the Levenberg-Marquardt algorithm to predict the thermal conductivity and compressive strength of foamed concrete.The experimental data are divided into training group and control group, the neural network is used to fit the data of the training group, and the model training is completed when the fitting results meet the error precision.We choose a set of model which fit better tocompere with control group to validate the accuracy of model prediction.The results show that the BP neural network model can accurately fit the experimental data, and the error of the control group is less than 8% by using the generalization ability of the model.Finally,this paper will present a new possible hypotheses based on the traditional Levy model,the hypotheses is that the concrete base(including pore structure) as a homogeneous matrix(not including pore structure),according to this hypothesis, the Levy model is better than traditional Levy model to predict the thermal conductivity.The prediction results are analyzed and compared with the prediction results of BP neural network model. The comparison results show that the Levy model is lower prediction accuracy with higher porosity,the prediction error of the same data of the BP neural network model is Levy model’s 58.4%.