基于贝叶斯神经网络的钢筋与混凝土黏结强度预测模型研究

A Prediction Model for Bond Strength Between Concrete and Steel Bars Based on Bayesian Neural Networks

提出一种新的钢筋与不同种类骨料混凝土间黏结强度的计算方法。结合748组试验数据,引入神经网络动态更新理论,建立了基于神经网络方法的黏结强度多参数预测模型,通过对黏结强度影响因素显著性的分析,利用神经网络参数剔除法简化了预测模型,并提出了临界锚固长度计算式。结果表明:钢筋与混凝土黏结强度主要受混凝土抗拉强度、锚固长度、钢筋直径和混凝土保护层厚度等因素的影响;建议模型具有较高预测精度,模型预测值与试验值比值的平均值和变异系数分别为1.052和0.337,为预测钢筋与混凝土黏结强度提供了新思路。

A new calculation method of bond strength between steel bars and concrete with different aggregate types was proposed.Combined with 748 groups of test data and the introduction to dynamic renewal theories of the neural network,a prediction model with multi parameters for bond strength based on neural network method was established,the significance analysis of influencing factors of bond strength was conducted,the prediction model was simplified by the parameter elimination method of the neural network,and the formula for critical anchorage length was put forward.Important factors that influenced bond performances were tensile strength,anchorage length,diameters of steel bars,and thickness of concrete covers.The proposed model was of higher prediction accuracy and the mean value and coefficient of variation for the ratio of the prediction values to test values were 1.056 and 0.377 respectively.It provided a new method to predict bond strength between steel bars and concrete.