| The domestic and international construction industry is in the stage of large-scale rehabilitation and reconstruction.The fiber-reinforced composite materials as new strengthening materials in recent decades have been widely concerned in the field of strengthening.Under the background of carbon peaking and carbon neutrality goals,its application prospect as a green material is bound to be broader.In the case of beam strengthening in strengthening projects,for example,the reinforced concrete beams being strengthened have safety problems-susceptible to debonding failure,including intermediate crack debonding and plate end debonding（noted as Type I and Type II safety problems in this study）,so their application is limited to some extent.The recognition and prediction of both types of safety problems is a widespread concern in the safety prediction of the strengthened beams.Scholars at home and abroad have conducted a series of studies on this safety issue and proposed different analytical models combining experiments and experiences,but the safety of the strengthened beams is affected by many indicators.Factors such as experimental environment,instrumentation,and people can affect the result of experiments and modeling so most of the prediction models constructed have large errors.With the growth of computer computing power and data scale in recent years,machine learning techniques have been applied to various fields.Introducing machine learning techniques to the safety prediction of the strengthened structures could provide new ideas to solve problems in this field.In this paper,a modeling study of a machine learning-based safety prediction model for the strengthened beams is conducted for the above problem,including three sub-models: safety pattern recognition model,Class I safety prediction model,and Class II safety prediction model.The research process is as follows.Firstly,domestic and foreign indicators and prediction models considered for Type I and Type II safety problems are reviewed and analyzed;secondly,the literature research method is used to establish a primary indication system for Type I,Type II safety prediction,and safety pattern recognition;thirdly,the correlation analysis and the gray relation analysis are used to study the indicators and then the final indication system is established;then classification algorithms in machine learning: K-nearest neighbor,decision tree,random forest,BP neural network,logistic regression,and support vector machine are used to build safety pattern recognition models;in addition,regression algorithms in machine learning: linear regression,ridge regression,decision tree,random forest,and BP neural network algorithms are used to build Class I and Class II safety prediction models;last but not least,the sparrow search algorithm is used to optimize the weights and thresholds of the BP neural network model;finally,the reliability evaluation of the model,L-M algorithm modeling,indicator importance analysis,and decision support system techniques are used to establish a safety prediction model for the strengthened beams.The following conclusions are obtained.The support vector machine is most suitable for the recognition of safety patterns,and the BP neural network model is suitable for the prediction of Class I and Class II safety problems;the accuracy of the BP neural network model optimized by the sparrow search algorithm is significantly improved;the established BP neural network and L-M prediction models are better than the prediction models suggested by the codes.the established safety prediction model can theoretically make effective safety prediction for the strengthened beams.Finally,the results and shortcomings of this study are summarized,the problems to be solved,and the directions to be improved are pointed out. |
