Bearing Capacity And State Prediction Of Concrete Structures Based On Machine Learning

Concrete is one of the most widely used building materials at present,so the performance and state prediction of concrete structure are very important.Traditionally,objective laws are extracted from experimental phenomena and mechanical models are established,and furthermore,semi-empirical formulas are fitted based on experimental data.However,some key factors are simplified in a certain extent by mechanical model,while the empirical formula has large error and weak generalization ability.A certain degree of disease will gradually accumulate on the concrete structure,affecting the safety of the structure.At present,the health status of the structure is mainly evaluated by analyzing the text of disease manually,but this method with low efficiency is easily influenced by subjective experience.Machine learning is a powerful data fitting tool,which can find the nonlinear relationship between input and output from the data level,and provides a new means for data processing and analysis.Based on the above background,this paper proposes a method to predict the bearing capacity and state of concrete structures using machine learning.This method is a data-driven calculation and evaluation method for structural performance.With the advantage of high efficiency and accuracy of machine learning algorithm,an efficient,accurate,multi-parameter and multi-data type data analysis model of concrete structure is established.In this paper,different machine learning models are established for different types of data,and the problems of shear calculation and overall health evaluation of structures,which are difficult to be dealt with by traditional methods,are selected as cases.The feasibility of establishing a general machine learning model for analyzing and processing traditional data and text data is discussed.Finally,the machine learning-based interpretability method provides reliability for the model prediction results.The main work and achievements are as follows:(1)An effective digital data analysis model based on machine learning algorithm was established.Firstly,the database of shear load test samples of reinforced concrete deep beams was established,and the shear factors such as material properties and geometric parameters were considered comprehensively.Then 270 samples were selected for feature processing,with input features including material properties,geometric dimensions,loading positions,etc.,and the measured values of shear capacity were taken as the output.Then,the ensemble learning model was established by programming programs based on Python language.Finally,the model was used to predict the shear capacity of deep beams,and the prediction accuracy was 90%.The comparison between the prediction results of the model and the calculation results of the empirical formula showed that the machine learning model has higher accuracy and smaller error,and the feasibility of the digital data analysis of concrete structures was verified.(2)Combined with natural language processing technology,an effective text data analysis model was established based on machine learning algorithm.Firstly,based on 263 bridge disease text samples,the text-health status level database was established.Then NLP technology and engineering experience were used to complete the word segmentation,feature processing and vector transformation of the disease text,and the key information of the disease was transformed.Finally,with text vector as input and bridge health evaluation level as output,the machine learning model was trained.The results showed that the classification effect of the model can reach 90% accuracy,and the feasibility of text type data processing and analysis was verified.(3)The interpretability of machine learning model prediction was analyzed based on existing research results.Specifically,Shap method was used to quantitatively calculate the contribution of shear influence parameters to shear capacity of shear wall at the level of data set.Shaply values of each parameter were analyzed from the perspectives of single sample and whole sample,and the results were in line with general research conclusions.The cross influence among parameters was also discussed.Meanwhile,SHAP method was used to calculate the shaply value of keywords in the text classification model,and the results showed that SHAP method can explain the text classification model of disease to a certain extent.