| There are many security risks in the construction site of building structure,such as multiple types of work,simultaneous operation of various large machines,complex construction environment and so on,which lead to the frequent occurrence of various safety accidents in the construction stage.In the process of dynamic construction,the loads of building structure are variable,and all kinds of supporting structures interact with each other,resulting in poor stability of the overall structure.At the same time,under the influence of natural conditions,building quality and other factors,safety accidents such as collapses and overturns are very easy to occur,which often cause serious casualties.Supported by the national 13th five-year plan project "Research and demonstration of key technologies for construction safety in super high-rise buildings"(2016YFC0802002-03),the key technologies of intelligent safety monitoring based on deep learning,low-cost intelligent mobile terminal and machine vision are studied for the construction elevator,tower crane,bolt connection,high formwork and deep foundation pit.At the same time,research on the integrated technology of construction safety monitoring system is carried out.The main research contents of this paper are as follows:(1)A portable detection method of construction elevator ride comfort based on intelligent mobile terminal is proposed.An acceleration acquisition software is developed,which can collect the operation acceleration of elevators by calling the acceleration sensor built in the smart phone,and evaluate the ride comfort of elevators by using the evaluation standard.First of all,the accuracy of the proposed method is verified by the shaking table test,and compared with the elevator comprehensive performance tester.The results showed that the correlation coefficient of the two types of detection data is greater than 0.92.Then,the influence of different operation mode and load on the comfort of elevator was analyzed.The ride comfort test and evaluation of elevators in three buildings are carried out to verify the practicability and convenience of the test method.Finally,a method of construction elevator ride comfort evaluation based on one-dimensional convolution neural network is proposed.This method has high detection accuracy and can achieve end-to-end rapid monitoring.Combining with smart phones,it can further improve the portability and rapidity of construction elevator ride comfort detection and evaluation.(2)Aiming at the collision risk monitoring of tower crane hoisting objects and workers,an intelligent monitoring method for safe distance of tower crane based on machine vision and Faster R-CNN(Region-based Convolutional Neural Network)is proposed.safety distance monitoring method in the process of tower crane hoisting is studied.First of all,a Faster R-CNN model based on the hook and worker is established to realize the accurately identify and locate the hook and worker in the picture.Then,the correlation between the object ratio(the ratio of pixel length to actual length)and the distance(the mark and camera)is studied.The dynamic monitoring of the horizontal distance between the hook and the worker is realized,and the average monitoring error is only 3.79%.Finally,in order to realize the safety distance monitoring of hoisting based on zoom camera,the relationship among camera focal length,ratio value and distance is verified,which further improves the practicability of the method.(3)Aiming at the relaxation damage of bolt connection,a relaxation damage monitoring method of bolt connection based on object detection is proposed,which includes small relaxation damage monitoring by bolt relaxation angle and large relaxation damage monitoring by whether the screw extends out of the structure surface.For small relaxation damage monitoring.SSD(single shot multibox detector)model for bolt looseness detection is established,and the recognition accuracy of the model is evaluated.According to the location information,the relaxation angle of bolt is obtained.The test results showed that the accuracy of relaxation angle monitoring is 95.53%.At the same time,the influence of different lighting conditions and different shooting angles on the method is analyzed.The detection model is transplanted into the smartphone,and the real-time bolt classification system based on the smart phone is completed.The potential of deep learning model based on structural damage in the smart mobile terminal is preliminarily verified.For the large relaxation damage monitoring,a Faster R-CNN model for bolt looseness detection is t established,and the recognition accuracy of the model is 0.9503.The minimum recognizable screw height of this method is evaluated.At the same time,the influences of different lighting conditions,shooting angles,structural vibration and different types of bolts on the method are analyzed.The network camera is connected with the model to realize the real-time monitoring of bolt large relaxation damage.(4)In view of the stability and safety monitoring of high formwork and deep foundation pit support in the construction stage,a rapid emergency monitoring method of structural displacement based on semantic segmentation and intelligent mobile terminal is proposed.According to the characteristics of high formwork safety monitoring,a single measurement point displacement monitoring method based on intelligent mobile terminal and Mask R-CNN model is proposed to realize pixel-level recognition and segmentation of target.The monitoring performance of the proposed method is verified by static test,dynamic test and remote displacement monitoring.The displacement monitoring of measuring point(10 m)is realized by optical telephoto camera,and the average monitoring error is only 4.18%.According to the characteristics of deep foundation pit safety monitoring,a longitudinal multi-point displacement monitoring method based on intelligent mobile terminal and FCN(Fully Convolutional Neural Network)model is proposed to realize pixel-level recognition and segmentation of multiple targets,so as to realize longitudinal multi-point displacement monitoring by using one intelligent terminal camera.The feasibility test,sensitivity test and multi-point displacement monitoring are used to verify the monitoring effect of the proposed method.The results show that the average monitoring error of each measuring point is less than 2%.(5)Finally,based on the above research,an intelligent monitoring system platform of building structure construction safety based on camera,intelligent mobile terminal and deep learning is proposed,and the function of the platform is designed.Based on the Internet of things technology,the intelligent construction site platform can monitor the safety of structures with great hidden dangers,such as construction elevators,tower cranes,bolt connections,high formwork and deep foundation pit.The platform includes two modules:real-time monitoring by webcam and fast monitoring/detection by smartphone.According to the characteristics of the two monitoring methods,the platform system is designed,including the server software and the mobile software.The server software can realize the functions of monitoring management,data display,status evaluation and security warning.The mobile software collects the relevant data by calling the acceleration sensor and camera built in the smart phone,transmits data to the server through the LAN or 4G network,and displays it in the server software,which is convenient for the unified management of the construction site.In addition,the smartphone has powerful computing power,which can transplant the deep learning model into the smartphone to complete the construction elevator ride comfort evaluation and bolt relaxation damage identification.Therefore,the monitoring system improves the safety monitoring ability of the construction site,and can detect and prevent safety accidents in time. |
