Research On Carbonation Of Concrete Based On Statistical Analysis And Machine Learning

As the problem of structural durability becomes more and more prominent,concrete carbonation,as the most common deterioration factor of reinforced concrete structures,has attracted more and more attention.Although a large number of studies have explored the effects of various parameters such as concrete strength,water-cement ratio,etc.on the carbonization process,most of the parameter research content only stays at the qualitative stage,and few carbonization models have a lot of consideration for parameter selection before they are established.To this end,based on statistical methods and machine learning principles,and the basis of existing carbonization research,this paper relies on a large number of carbonization data to quantitatively analyze and evaluate the carbonization parameters.The quantitative analysis considers the correlation between parameters and carbonization depth and The role of parameters in reducing the uncertainty of carbonation depth.The study found that concrete strength and aggregate-cement ratio have obvious advantages over other parameters.Based on single parameter analysis,considering the interaction between parameters,this paper calculates the corresponding parameters Combination,model validation results show that parameter combination can achieve effective control and prediction of concrete carbonation with few parameters.Aiming at the differences between existing structures and new structures in the prediction of concrete carbonation and the shortcomings of current prediction models,this paper provides corresponding solutions based on machine learning methods.For the existing structure,this paper uses a small number of parameters to establish a big data prediction platform for concrete carbonation of the existing structure.The data platform can supplement data at any time during use and update the model in real-time.In the long run,this is to solve the problem of data dependence of machine learning models.A feasible solution,the reliability of continuously updating the model will increase with time.On this basis,this paper proposes a corresponding simplified practical model.The practical model is based on a large amount of data,involves few parameters,and has high accuracy.For the carbonization problem of new structures,this paper combines the powerful ability of machine learning algorithms to learn from data and the advantages of the clear physical meaning of carbonization theoretical equations.At the same time,it makes up for the data dependence of machine learning.Differential equations are used to constrain the model learned by the neural network in the data,and the physical constraints of the new structural concrete are established-a data-driven neural network model.This study expounds on the machine learning model and the training process of the model from a new perspective.The verification results show that the model has a high performance.At the same time,it also proves that machine learning methods can not completely rely on experimental data,and fit well with theoretical research,and the application is not limited to simple mathematical equations,but also can be used for complex engineering problems,considering that machine learning methods are used in theoretical learning and There is great potential in data learning,and related methods can be transferred to other areas of durability research,such as chloride ion erosion,acid rain erosion,etc.