Device for identifying apparent quality of lining concrete of hydraulic tunnel

Abstract

The invention discloses a hydraulic tunnel lining concrete apparent quality recognition device which comprises a recognition trolley. The identification trolley comprises a power system, a vehicle control system, a positioning system, an image acquisition system, a storage system and an image identification processing system; and identifying the apparent quality of the tunnel lining concrete by using the trained hydraulic tunnel lining concrete apparent quality classification model and segmentation model. The method has the advantages that the unmanned recognition trolley and the intelligent photographing and analysis system are adopted, manual participation and manual subjective analysis are reduced, meanwhile, two cameras in each direction shoot the same area of the tunnel, the accuracy of concrete apparent quality judgment is improved, the shot images are classified and then segmented, and the accuracy of concrete apparent quality judgment is improved. The identification efficiency and the identification precision can be improved; the model adopts a newest Deeplab segmentation framework and applies cavity convolution, so that the training loss rate is reduced, and the segmentation effect is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of construction quality inspection, and is particularly applicable to an apparatus for identifying the apparent quality of lining concrete of a hydraulic tunnel. Background Art

[0002] As an important part of water conservancy project construction, the safety of a hydraulic tunnel is directly related to the construction safety and operation safety of the water conservancy project. Hydraulic tunnels are usually arranged in the strata, and are easily affected by ground stress. At the same time, it is difficult to achieve 100% excellent construction quality. Therefore, under the combined influence of ground stress and construction quality, cracks will occur on the lining surface of the hydraulic tunnel, and even the surface lining concrete will fall off. There are also phenomena such as honeycombing, pitting, and exposed steel bars in some areas.

[0003] Currently, the detection of the apparent quality of lining concrete of a hydraulic tunnel is mainly carried out by manual on-site measurement, recording, marking, and drawing. This method is often inefficient, dangerous, and greatly affected by subjectivity, making it difficult to ensure the objectivity and accuracy of the detection results. There are also some projects that embed conductive coatings in the lining concrete of the tunnel and judge whether there are cracks in the lining concrete by detecting the resistance value. However, this method is troublesome in construction, and cannot fully reflect the apparent morphology of the cracks. At the same time, the dark and humid environment of the tunnel has a great impact on the detection effect of the conductive coating, so it is difficult to promote. Using ground penetrating radar can accurately detect the existence of cracks, but both it and the method of embedding conductive coatings are difficult to systematically describe the apparent quality of concrete, and only have a good detection effect on cracks, and are difficult to describe honeycombing, pitting, and exposed steel bars, etc.

[0004] With the development of artificial intelligence, many domestic and foreign scholars have also tried to apply image recognition technology to tunnel crack detection. However, most of them require manual holding of a camera to take pictures in the tunnel, and then bring the shooting results back to the laboratory for analysis and processing. This method is difficult to ensure the shooting quality and is inefficient. At present, most of the research on the rapid recognition of the apparent quality of tunnels focuses on only one defect, namely cracks, and there is little research on the quality recognition of defects such as honeycombing, pitting, and exposed steel bars, and even less targeted research on the recognition of the apparent quality of lining concrete of hydraulic tunnels. Summary of the Invention

[0005] The purpose of the present invention is to provide an apparatus for identifying the apparent quality of lining concrete of a hydraulic tunnel, which is used to solve the problems of high cost, low efficiency, difficult maintenance, and low detection accuracy in the existing detection technologies for the apparent quality of hydraulic tunnels.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions: The apparatus for identifying the apparent quality of lining concrete of a hydraulic tunnel according to the present invention includes an identification trolley; the identification trolley includes a power system, a vehicle control system, a positioning system, an image acquisition system, a storage system, and an image recognition and processing system; The power system provides driving power for the vehicle body and supplies power to the electrical modules carried on the vehicle body. The vehicle control system controls the vehicle body to travel along the center line of the bottom of the tunnel through symmetrically installed laser rangefinders. The image acquisition system includes six CCD cameras for taking apparent images of the tunnel lining concrete at the current position of the vehicle body. The positioning system is used to determine the current position of the vehicle body, and further determine the acquisition area of the apparent image of the tunnel lining concrete. The storage system is used to store the apparent images of the tunnel lining concrete and the images processed by the image recognition processing system. The image recognition processing system is used to analyze and process the apparent images of the tunnel lining concrete, and identify the apparent quality of the tunnel lining concrete by using the trained classification model and segmentation model for the apparent quality of hydraulic tunnel lining concrete.

[0007] Further, the recognition trolley includes two crawler wheels arranged symmetrically, a shock-absorbing layer mounted above the two crawler wheels, and a bottom plate on the shock-absorbing layer; the shock-absorbing layer is composed of springs and shock-absorbing beams.

[0008] Further, the power system includes a driving power source and a motor; the driving power source is arranged on the bottom plate at the tail of the recognition trolley; the motor is arranged below the shock-absorbing layer and is respectively connected to the driving power source and the crawler wheels to provide driving power for the vehicle body.

[0009] Further, the laser rangefinders are symmetrically installed on the left and right side mounting plates perpendicular to the bottom plate.

[0010] Further, the six CCD cameras are divided into three groups and symmetrically installed on the left and right side mounting plates and the top plate of the recognition trolley; each group is equipped with a gyroscope, and each CCD camera is equipped with a shadowless lamp.

[0011] Further, the identification of the apparent quality of the tunnel lining concrete by using the trained classification model and segmentation model for the apparent quality of hydraulic tunnel lining concrete specifically includes the following steps: S1, collecting pictures of the hydraulic tunnel lining concrete with quality problems, good quality and interference objects on the surface; S2, performing grayscale processing on the collected pictures, sequentially marking the interference objects, and classifying and marking the quality problems; S3, dividing the pictures collected in step S1 and the marked pictures into four data sets A, B, C, and D, where data sets A and B include all the pictures collected in step S1 and the marked pictures; data sets C and D include the pictures of the tunnel lining concrete with quality problems collected in step S1 and the corresponding marked pictures; S4. Create a classification model and a segmentation model for the surface quality of the lining concrete of hydraulic tunnels based on convolutional neural network computer vision and deep learning; S5. Use datasets A and B to train and validate the classification model for the surface quality of the lining concrete of hydraulic tunnels for the classification and discrimination of surface problems of the lining concrete of hydraulic tunnels; S6. Use datasets C and D to train and validate the segmentation model for the surface quality of the lining concrete of hydraulic tunnels for extracting the problem areas on the surface of the lining concrete of hydraulic tunnels; S7. Obtain a binary image of the problem area through threshold segmentation, extract and vectorize the feature points of the binary image, and characterize the problem area according to the disease category.

[0012] Furthermore, the disease categories include cracks, exposed steel bars, and honeycombing and pockmarks; the crack problems and exposed steel bar problems are characterized by length, width, diameter, and inclination angle; the honeycombing and pockmark problems are characterized by area.

[0013] Furthermore, the length calculation method is: The inclination angle calculation method is: , where l is the length of the crack or exposed steel bar, and n are the n coordinate points of the skeleton feature points of the crack or exposed steel bar, x i and y i are the x and y coordinates of the i-th coordinate point of the skeleton feature point respectively, x i+1 and y i+1 are the x and y coordinates of the (i + 1)-th coordinate point of the skeleton feature point respectively; θ is the inclination angle; the crack width or exposed steel bar diameter calculation method is: for any feature point a on the skeleton within the problem area, take 10 adjacent feature points to perform linear fitting, determine the normal line of the fitting line, and take the straight-line distance between the two intersection points of the normal line and the contour boundary of the problem area as the crack width or exposed steel bar diameter at feature point a; the average value of the crack widths or exposed steel bar diameters at all feature points is the crack width or exposed steel bar diameter of the problem area.

[0014] Furthermore, the honeycombing and pockmark area calculation method is: according to the vectorized honeycombing and pockmark contour feature points extracted from the binary image, number the representative points of each honeycombing and pockmark hole, use the Alpha Shapes algorithm to extract the outer boundary, and after sorting the representative points on the outer boundary, use the n-sided polygon area calculation formula to calculate the honeycombing and pockmark area.

[0015] The advantages of the present invention are: 1. The present invention uses an unmanned recognition vehicle and an intelligent photography and analysis system, reducing manual participation and subjective human analysis. At the same time, two cameras in each direction photograph the same area of the tunnel, improving the accuracy of judging the apparent quality of concrete.

[0016] 2. The recognition vehicle of the present invention is driven by double tracks, can adapt to poor working environments, and uses a gyroscope to ensure the stability of the shooting process, without being affected by excessive terrain undulations. During the vehicle's movement, the invention device photographs the surface topography of the lining concrete while transmitting the photos to the processing system to analyze the apparent quality of the concrete, improving the overall work efficiency.

[0017] 3. The present invention uses shadowless lights for illumination, which can effectively reduce the shadow area, reduce the interference of shadows on the quality of the photographed images, and reduce the influence of shadows on the image recognition process, facilitating the improvement of the detection accuracy of the apparent quality of the lining concrete in hydraulic tunnels.

[0018] 4. The present invention adds anti-interference object training to the training of the classification model and segmentation model for the apparent quality of the lining concrete in hydraulic tunnels, improving the accuracy of model classification. By classifying the images first and then segmenting them, it is beneficial to improve the recognition efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of the device for recognizing the apparent quality of the lining concrete in the hydraulic tunnel described in the present invention.

[0020] Figure 2 It is a cross-sectional schematic diagram of the device for recognizing the apparent quality of the lining concrete in the hydraulic tunnel described in the present invention.

[0021] Figure 3 It is a flowchart of the method for classifying, segmenting, and characterizing the problem areas of the apparent quality of the lining concrete in the hydraulic tunnel described in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0023] The device for recognizing the apparent quality of the lining concrete in the hydraulic tunnel described in the present invention includes a recognition vehicle; the recognition vehicle includes a power system, a vehicle control system, a positioning system, an image acquisition system, a storage system, and an image recognition and processing system.

[0024] As Figure 1 、 Figure 2As shown in the figure, the identification trolley is driven by double tracks and includes two symmetrically arranged track wheels 3, a shock-absorbing layer 5 mounted above the two track wheels, and a bottom plate 6 on the shock-absorbing layer. The shock-absorbing layer 5 is composed of springs, shock-absorbing beams, shock-absorbing materials, etc., to prevent excessive vibration of the upper equipment and affect the use effect. The bottom plate 6 is combined into a cuboid by mounting plates and various processing devices are loaded inside. Using a tracked vehicle improves the vehicle's ability to adapt to the tunnel environment and avoids operating failures such as vehicle skidding due to reasons such as accumulated water on the ground.

[0025] The power system carried by the identification trolley includes a driving power source 2 and a motor 1. The driving power source 2 is arranged on the bottom plate and is located at the rear of the identification trolley. The motor 1 is arranged below the shock-absorbing layer 5 and is respectively connected to the driving power source 2 and the track wheels 3 to provide driving power for the vehicle body. At the same time, the driving power source 2 also provides power for the electrical modules carried on the vehicle body.

[0026] In one embodiment, a common power source 4 can be additionally provided to provide power for the electrical modules carried on the vehicle body. This design enables the driving power source 2 to independently provide power for the identification trolley, ensuring the stable operation of the power system.

[0027] The vehicle control system 7 is installed in the front of the bottom plate and is used to control the running parameters of the trolley in real time. Laser rangefinders 8 are symmetrically installed below the left and right mounting plates perpendicular to the bottom plate and are fed back to the vehicle control system in real time to ensure that the vehicle travels on the center line at the bottom of the tunnel.

[0028] The image acquisition system includes six CCD cameras 10 for taking the apparent image of the tunnel lining concrete at the current position of the vehicle body. Each camera is equipped with a shadowless lamp 11. Two cameras, two shadowless lamps and a gyroscope 14 are installed on each of the left and right mounting plates 12 and the top plate 13. That is, the six CCD cameras are divided into three groups and symmetrically installed on the left and right mounting plates and the top plate of the identification trolley. Each group is configured with a gyroscope, and each CCD camera is configured with a shadowless lamp. The gyroscope on each side controls the running angles of the cameras and shadowless lamps on that side. The shadowless lamp reduces the shadows generated on the surface of the hydraulic tunnel lining due to unevenness under light. The gyroscope adjusts the working angles of the cameras and shadowless lamps in real time according to the traveling situation of the vehicle, avoiding the influence on the photo quality caused by vehicle bumps. At the same time, the image acquisition system is interconnected with the positioning system and the storage system 15 to mark and store the positions of the pictures.

[0029] The positioning system is installed at the rear of the bottom plate. It determines the current position of the trolley in the hydraulic tunnel according to the traveling distance of the trolley. It is associated with the image acquisition system to record the position of each taken picture, and then determines the acquisition area of the apparent image of the tunnel lining concrete.

[0030] The storage system is used to store the photos taken by the camera and the photos processed by the recognition processing system. The photos taken by the camera and their corresponding tunnel location information are first stored in the storage system and then transmitted to the image recognition and analysis device. The picture information processed by the image recognition processing system 16 is transmitted back to the storage device. Therefore, the storage system is connected to both the acquisition system and the picture recognition processing system. The user can read the stored information through the external data socket 17 or add an information transmission device to achieve remote real-time reading of the stored information. The main device of the image recognition processing system is the GPU. After inputting the picture, it runs the program code to identify, segment, and output the results of the picture.

[0031] The image recognition processing system is used to analyze and process the apparent images of the tunnel lining concrete, and identify the apparent quality of the tunnel lining concrete by using the trained classification model and segmentation model for the apparent quality of the hydraulic tunnel lining concrete.

[0032] The identification of the apparent quality of the tunnel lining concrete by using the trained classification model and segmentation model for the apparent quality of the hydraulic tunnel lining concrete specifically includes the following steps, and the process is as Figure 3 shown.

[0033] S1. Collect pictures of the hydraulic tunnel lining concrete with quality problems, good quality, and with interference objects. That is, collect pictures of the hydraulic tunnel lining concrete with good surface quality, pictures with surface quality problems, and pictures with interference objects such as splicing seams and wires on them. These pictures with interference objects can be of the hydraulic tunnel lining concrete with good surface quality or with surface quality problems.

[0034] S2. Grayscale the collected pictures, mark the interference objects in sequence, and classify and mark the quality problems. The main quality problems include cracks, honeycombing and pockmarks, and exposed steel bars. For the pictures with surface quality problems, the specific problem categories need to be marked. For the pictures with interference objects, the interference objects need to be marked.

[0035] S3. Divide the pictures collected in step S1 and the marked pictures into four data sets, namely A, B, C, and D. Among them, data sets A and B include all the pictures collected in step S1 and all the marked pictures; data sets C and D include the collected pictures of the tunnel lining concrete with surface quality problems in step S1 and the corresponding marked pictures. The collected pictures with surface quality problems and the corresponding marked pictures both include pictures with and without interference objects.

[0036] S4. Create a classification model and a segmentation model for the surface quality of the lining concrete of hydraulic tunnels based on computer vision and deep learning with convolutional neural networks. The classification model is constructed based on Resnet101, and the segmentation model is constructed based on DeepLab combined with Resnet101. Both are implemented using TensorFlow, and OpenCV is used for image preprocessing.

[0037] S5. Use datasets A and B to train and validate the classification model for the surface quality of the lining concrete of hydraulic tunnels, which is used for the classification and discrimination of surface problems of the lining concrete of hydraulic tunnels.

[0038] S6. Use datasets C and D to train and validate the segmentation model for the surface quality of the lining concrete of hydraulic tunnels, which is used to extract the problem areas on the surface of the lining concrete of hydraulic tunnels.

[0039] It should be noted that before the model is trained, in order to reduce the waste of picture pixels and improve the recognition accuracy of microcracks, it is necessary to segment the pictures in the training set, input the segmented pictures into the classification model and the segmentation model, and then splice the pictures after the model outputs.

[0040] Optimize the parameters such as the number of pictures read at one time, the number of training rounds, and the learning rate for the classification model. When the test accuracy meets the set threshold, it means the model can be used; otherwise, continue to optimize the parameters for training. The accuracy formula is: In the formula, PPV is the accuracy, TP is the number of cases predicted as diseases and actually being diseases, and FP is the number of cases where the background is predicted as diseases.

[0041] Optimize the parameters such as the number of pictures read at one time, the number of training rounds, and the learning rate for the segmentation model. When the test accuracy meets the set threshold, it means the model can be used; otherwise, continue to optimize the parameters for training. The accuracy formula is: In the formula, represents the test accuracy, is the pixel area of the prediction result, is the pixel area of the actual result.

[0042] S7. Through the classification model and the segmentation model, the problem categories on the surface of the lining concrete of hydraulic tunnels can be obtained, such as cracks, honeycombing and pitted surfaces, and exposed steel bars, etc., and the problem areas can be determined. Then, a binary image of the problem area is obtained through the threshold segmentation method, the feature points of the binary image are extracted and vectorized, and the problem area is characterized according to the disease category.

[0043] Characterize the crack problems and exposed steel bar problems by length, width, diameter and inclination angle; characterize the honeycombing and pitted surface problems by area.

[0044] The calculation method for the length of the crack or exposed steel bar problem is as follows: .

[0045] The calculation method for the inclination angle is as follows: .

[0046] Wherein, l is the length of the crack or exposed steel bar, n is the n coordinate points of the characteristic points of the crack or exposed steel bar skeleton, x i and y i are the x and y coordinates of the i-th coordinate point of the skeleton characteristic point respectively, x i+1 and y i+1 are the x and y coordinates of the (i + 1)-th coordinate point of the skeleton characteristic point respectively; θ is the inclination angle.

[0047] The calculation method for the crack width or exposed steel bar diameter is as follows: For any characteristic point a on any skeleton in the problem area, take 10 characteristic points adjacent to a for linear fitting, determine the normal line of the fitting line, and take the straight-line distance between the two intersection points of the normal line and the contour boundary of the problem area as the crack width or exposed steel bar diameter at the characteristic point a; The average value of the crack width or exposed steel bar diameter at all characteristic points is the crack width or exposed steel bar diameter of the problem area.

[0048] The calculation method for the honeycomb pitted surface area is as follows: According to the vectorized honeycomb pitted surface contour characteristic points extracted from the binary image, number the representative points of each honeycomb pitted surface hole, use the Alpha Shapes algorithm to extract the outer boundary, and after sorting the representative points on the outer boundary, use the n-sided polygon area calculation formula to calculate the honeycomb pitted surface area.

[0049] The calculation method for the n-sided polygon area is as follows: In the formula, S is the area of the outer boundary polygon, n is the number of sides of the outer boundary polygon or the number of representative points of the outer boundary, and are respectively the -th representative point's x and y coordinates, and are respectively the -th representative point's x and y coordinates.

Claims

1. A device for identifying the apparent quality of hydraulic tunnel lining concrete, characterized in that: It includes an identification vehicle; the identification vehicle includes a power system, a vehicle control system, a positioning system, an image acquisition system, a storage system and an image recognition processing system; The power system provides driving power for the vehicle body and provides power for the power modules carried on the vehicle body; The vehicle control system controls the vehicle body to travel along the center line of the tunnel bottom through symmetrically installed laser rangefinders; The image acquisition system includes 6 CCD cameras for capturing the surface image of the tunnel lining concrete at the current vehicle body position; The positioning system is used to determine the current position of the vehicle body, and then determine the acquisition area of ​​the tunnel lining concrete surface image; The storage system is used to store the tunnel lining concrete surface image and the image processed by the image recognition and processing system; The image recognition and processing system is used to analyze and process tunnel lining concrete surface images, and to identify the tunnel lining concrete surface quality by using the trained hydraulic tunnel lining concrete surface quality classification model and segmentation model.

2. The device for identifying apparent quality of hydraulic tunnel lining concrete according to claim 1 is characterized in that: The identification vehicle comprises two symmetrically arranged crawler wheels, a shock-absorbing layer mounted above the two crawler wheels, and a bottom plate on the shock-absorbing layer; the shock-absorbing layer is composed of springs and shock-absorbing beams.

3. The device for identifying apparent quality of hydraulic tunnel lining concrete according to claim 2, characterized in that: The power system includes a driving power supply and a motor; the driving power supply is arranged on the bottom plate and located at the rear of the identification vehicle; the motor is arranged below the shock-absorbing layer and is respectively connected to the driving power supply and the track wheels to provide driving power for the vehicle body.

4. The device for identifying apparent quality of hydraulic tunnel lining concrete according to claim 2, characterized in that: The laser rangefinder is symmetrically mounted on left and right mounting plates which are perpendicular to the bottom plate.

5. The device for identifying apparent quality of hydraulic tunnel lining concrete according to claim 4, characterized in that: The six CCD cameras are divided into three groups and symmetrically mounted on the left and right mounting plates and the top plate of the identification vehicle; each group is equipped with a gyroscope, and each CCD camera is equipped with a shadowless lamp.

6. The device for identifying apparent quality of hydraulic tunnel lining concrete according to claim 1, characterized in that: The method of identifying the apparent quality of the hydraulic tunnel lining concrete by using the trained apparent quality classification model and segmentation model specifically comprises the following steps: S1, collect pictures of the surface of hydraulic tunnel lining concrete with quality problems, good quality and interference objects; S2, grayscale the collected images, mark the interference objects in turn, and classify and mark the quality problems; S3, divide the collected pictures and marked pictures in step S1 into four data sets: A, B, C, and D, where the A and B data sets include the collected pictures and all marked pictures in step S1; the C and D data sets include the collected pictures with quality problems on the lining concrete surface in step S1 and the corresponding marked pictures; S4, computer vision and deep learning based on convolutional neural networks to create a classification model and segmentation model for the surface quality of hydraulic tunnel lining concrete; S5, using datasets A and B to train and validate the classification model of hydraulic tunnel lining concrete surface quality, which is used to identify the classification and discrimination of hydraulic tunnel lining concrete surface problems; S6, using datasets C and D to train and validate the segmentation model of the surface quality of the hydraulic tunnel lining concrete, is used to extract the problem areas on the surface of the hydraulic tunnel lining concrete; S7, obtain a binary image of the problem area through a threshold segmentation method, extract feature points of the binary image and vectorize them, and characterize the problem area according to the disease category.

7. The device for identifying apparent quality of hydraulic tunnel lining concrete according to claim 6, characterized in that: The disease categories include cracks, exposed steel bars and honeycombed surfaces; crack problems and exposed steel bar problems are characterized by length, width, diameter and inclination; honeycombed surface problems are characterized by area.

8. The device for identifying apparent quality of hydraulic tunnel lining concrete according to claim 7, characterized in that: The length calculation method is: , the inclination angle calculation method is: , l is the length of the crack or exposed steel bar, n is the coordinates of the crack or exposed steel bar skeleton feature points, x i and y i are the x and y coordinates of the i-th coordinate point of the skeleton feature point, x i+1 and y i+1 are the x and y coordinates of the i+1th coordinate point of the skeleton feature point respectively; θ is the inclination angle, and the calculation method of the crack width or the exposed steel bar diameter is: for a feature point a on any skeleton in the problem area, take 10 feature points adjacent to a for straight line fitting, determine the normal of the fitting line, and take the straight-line distance between the two intersection points of the normal and the contour boundary of the problem area as the crack width or the exposed steel bar diameter at the feature point a; the mean of the crack width or the exposed steel bar diameter at all feature points is the crack width or the exposed steel bar diameter of the problem area.

9. The device for identifying apparent quality of hydraulic tunnel lining concrete according to claim 7, characterized in that: The honeycomb pockmark area calculation method is as follows: according to the vectorized honeycomb pockmark contour feature points extracted from the binary image, the representative points of each honeycomb pockmark hole are numbered, the outer boundary is extracted by using the Alpha Shapes algorithm, and after the representative points on the outer boundary are sorted, the honeycomb pockmark area is calculated by using the n-gon area calculation formula.

Images