Multi-scale Intelligent Recognition For Damage Of Reinforced Concrete Bridge

In the process of service,different types of structural damage and aging will occur,and as the bridge continues to operate,the damage and aging will be further developed,seriously threatening the safe operation of the bridge.To reduce bridge accidents,accurate and effective bridge damage identification is essential."To promote the deep integration of information technology and manufacturing technology,with a focus on the industrialization of a new generation of artificial intelligence technology and integrated applications,to promote the deep integration of artificial intelligence and the real economy" was proposed in the three-year action plan to promote the development of a new generation of artificial intelligence industry,the development and application of artificial intelligence have risen to the level of national strategy.There is a trend to carry out bridge damage identification based on artificial intelligence technology for bridge damage identification.Traditional identification methods mainly rely on manual work and face the limitations of strong subjectivity and low efficiency,while bridge damage identification based on deep learning has the advantages of high accuracy,high efficiency,and intelligence.Therefore,it is vital to carry out intelligent bridge damage recognition based on deep learning technology.At present,research on bridge damage identification based on deep learning has been carried out at home and abroad,but there is a lack of multi-scale research on bridge damage identification.Therefore,concrete bridges were used as the research object and carries out the research on multi-scale intelligent recognition of bridge damage based on deep learning in this paper.Firstly,machine learning algorithms and deep learning neural networks were used to analyze the dynamic response text signals of bridges to achieve macro-scale damage identification;Secondly,local damage identification of cracks and surface spalling is performed by object detection algorithms to achieve fine-scale accurate localization of local damage of bridges;Finally,concrete cracks were identified at the pixel level by semantic segmentation neural network to achieve micro-scale damage identification of cracks.The main research of this paper is as follows.(1)Macroscale identification of bridge damage based on dynamic response was carried out.The bridge damage data set was established through the impact test and finite element simulation of indoor test beams.Bridge damage global damage recognition was divided into three parts: a.bridge damage and damage state recognition;b.bridge damage location recognition,for bridge damage location intelligent recognition,was divided into single damage location recognition without considering damage quantification,single damage location recognition considering damage quantification,bridge multi-damage location multi-label classification recognition,c.bridge damage clustering analysis under the condition of incomplete a priori information.By studying the above three levels of bridge damage identification,the research framework of bridge damage was realized,including state of bridge damage identification,bridge damage location identification,unsupervised clustering analysis of bridge damage information,and global damage intelligence identification of bridge damage.(2)Fine-scale identification of cracks based on object detection was carried out.Firstly,the concrete cracks were used as the research object,and the labeling of the crack images was performed by the moderately dense labeling method to establish the crack dataset for object detection;Then,the YOLO-V5 algorithm was established for crack object detection.Model training and parameter optimization were performed using the training and validation datasets;Finally,the optimized model was evaluated,and the optimized YOLO-V5 model was used for crack object detection identification.The recognition results show that YOLO-v5 can achieve accurate recognition for crack images under different conditions,reflecting the strong robustness of the YOLO-v5 model.(3)The object detection for concrete surface spalling identification was carried out at fine-scale.Firstly,to simulate the surface spalling damage of concrete,concrete wind corrosion tests were conducted indoors,and a concrete wind corrosion data set was established considering the effects of water damage,scratches,and background noise,etc.Then,YOLO-v4 was used as the backbone of the concrete erosion damage detection algorithm,and the algorithm was improved by combining the transformer principle,and an MHSA-YOLOV4 model was proposed.Finally,model validation was performed through model training,parameter tuning,and model comparison.The validation results show that the MHSA-YOLOV4 model has excellent object detection performance and can accurately identify concrete erosion damage for different damage levels and different interference factors,reflecting the excellent generalization ability of the MHSA-YOLOV4 model.(4)Microscale identification of concrete cracks based on semantic segmentation algorithm was carried out.In order to improve the weight of the crack region,the UNet neural network base model was improved by introducing an attention mechanism and a residual network,and then a deep residual convolutional neural network(DRACNN)based on the attention mechanism was proposed.