Research On Intelligent Crack Evaluation Technology Of Concrete Beam Bridge Based On Reinforcement Learning

For concrete beam bridges,the most common disease is cracks.Therefore,the health evaluation of concrete beam bridges should start with the detection and evaluation of structural cracks.At present,many research results have been achieved in the intelligent detection of bridge cracks,which can quickly and accurately extract the crack shape feature information and provide data support for crack evaluation,However,the evaluation of bridge cracks still depends on manpower.The traditional manual evaluation has some shortcomings,such as long time-consuming,heavy workload and strong subjectivity of evaluation results.Taking the cracks of concrete beam bridges as the research object,this paper proposes a concrete beam bridge crack evaluation algorithm based on DQN algorithm to automatically evaluate the cracks of concrete beam bridges,so as to provide scientific data support for the maintenance decision-making of concrete beam bridges in the operation period.(1)Firstly,this paper summarizes and analyzes the formation mechanism and harm degree of concrete structure cracks,takes the crack width,crack length,crack trend,crack longitudinal position and crack structural plane as the parameter indexes to evaluate the crack grade,evaluates the crack according to the experience of experts,and arranges the samples,A three-level fracture state database is formed to prepare the data set for the training of reinforcement learning network model.(2)Starting from the normal service limit state of the structure.The parameters of crack evaluation are dimensionless,so that each parameter has the same influence level on the result of crack evaluation.A fracture evaluation model based on DQN algorithm is designed.The results of expert evaluation are compared with those of intelligent evaluation to verify the accuracy of fracture intelligent evaluation algorithm.(3)Through statistical analysis of the crack data of each grade after evaluation,the relationship between each grade of crack and crack parameters is obtained,and a new standard for dividing the crack state grade is proposed.The crack is divided into five scales according to the severity.Compared with the current code,this method divides the crack grade more finely.Compared with the expert evaluation results,the crack evaluation model based on dqn algorithm proposed in this paper has a recognition accuracy of 97.0%on the test set,has good generalization ability,can overcome the subjectivity of the evaluation results and improve the work efficiency,and can be used for the automatic evaluation of concrete bridge cracks.Compared with the original classification results,the accuracy of the proposed fracture classification standard on the randomly selected sample set is 96.6%,which can accurately classify the fracture grade...