A method for evaluating the roughness of concrete roughening based on a convolutional neural network

Abstract

The present invention discloses a method for evaluating the roughness of concrete chiseling based on a convolutional neural network. The method uses a depth camera to obtain pictures of the chiseled concrete surface, establishes the evaluation index of the chiseled concrete, namely the roughening roughness, based on the chiseling point cloud information obtained after preprocessing. Through the splitting tensile test of layered casting specimens with different degrees of chiseling on the contact surface, the relationship between the roughening roughness and the interfacial bonding strength is established, and the chiseling degree is classified into under-chiseling, ideal chiseling, and over-chiseling. Finally, by establishing a convolutional neural network, based on the camera photos and the chiseling classification, a data set is established and data augmentation is carried out, and the classification and evaluation can be automatically obtained after the camera takes pictures and inputs them into the network. The present invention obtains the surface information of the bonding surface by means of lidar and camera scanning of the solidified concrete surface, constructs a method for the evaluation index system of the bonding surface quality, and automatically detects the chiseling situation of the bonding surface based on deep learning, realizing the intelligent evaluation of the chiseling quality of dam concrete in alpine regions.

Description

Technical Field

[0001] The invention belongs to the technical field of intelligent management of engineering construction, and relates to a method for evaluating the roughness of concrete chiseling based on a convolutional neural network. Background Art

[0002] At present, there is no unified standard for the treatment standards and detection methods of the roughness of the bonding surface of layered concrete pouring at home and abroad. According to existing research and practice, there are three common detection methods: sand filling method, fractal dimension method and visual recognition method. The sand filling method is one of the earliest detection methods. It reflects the degree of concavity and convexity of the bonding surface by pouring a certain amount of sand on the bonding surface and then measuring the height of the sand. This method is simple to operate, but its accuracy is not high, which is not conducive to construction quality control. The fractal dimension method is a detection method based on mathematical theory. It reflects the complexity and irregularity of the bonding surface by taking a complete fractal analysis of the surface morphology of the bonding surface to obtain its fractal dimension value. This method can theoretically better describe the roughness of the bonding surface, but its equipment cost is high, the operation is complex, the data processing volume is large, and it is not suitable for engineering construction in high-altitude cold areas. The human eye recognition method is one of the commonly used methods on construction sites. Construction technicians usually guide the roughening process based on highly subjective terms such as "no milky skin on the construction joint surface, slightly exposed fine aggregate", and judge the roughening quality based on personal experience, making it impossible to finely control the construction quality of multi-layer poured concrete.

[0003] With the improvement of the resolution accuracy of optical camera equipment and the multi-dimensional development of artificial intelligence neural networks, visual recognition technology, as a technology that uses computers and artificial intelligence to analyze and understand the content and information in images, has provided new ideas for concrete construction quality inspection. In the harsh construction environment of high cold and high altitude, workers' attention and endurance are accelerated to decline, which not only significantly reduces efficiency, but also easily causes construction safety accidents. Therefore, it is necessary to give full play to digital and intelligent means to solve many adverse problems in the construction of water conservancy projects, such as large temperature difference between day and night, significant reduction in manpower efficiency, and continuous construction in ultra-long low temperature seasons, so as to improve the efficiency of safe construction of water conservancy projects in high cold and high altitude areas.

[0004] At present, the horizontal seam surface treatment adopts high-pressure water artificial roughening. Due to the restriction of the process, the roughening effect is difficult to meet the uniformity requirements, and the process qualification inspection lacks scientific and quantitative objective evaluation standards and refined detection methods. Summary of the invention

[0005] The object of the present invention is to provide a method for evaluating the roughness of concrete roughening based on a convolutional neural network. According to the requirements of the intelligent control system for concrete construction, the present invention obtains the surface information of the bonding surface by scanning the surface of the solidified concrete with devices such as lidar and cameras, constructs a method for the evaluation index system of the bonding surface quality, and automatically detects the roughening condition of the bonding surface based on deep learning, so as to realize the intelligent evaluation of the roughening quality of dam concrete in alpine regions.

[0006] The present invention is realized through the following technical solutions:

[0007] A method for evaluating the roughness of concrete roughening based on a convolutional neural network, characterized by comprising the following steps:

[0008] S1: Obtain roughening photos by using a depth camera, and preprocess the collected pictures;

[0009] S2: Based on the information after preprocessing, combined with multiple correlation analysis methods such as Pearson detection, derive a roughening roughness R calculation formula applicable to the evaluation of the roughening treatment of the concrete blank layer surface, and use the roughening roughness R calculation formula to calculate the roughening roughness;

[0010] S3: Construct a roughening evaluation standard based on the relationship between the roughening roughness and the interfacial bonding strength;

[0011] S4: Construct a roughening evaluation classification model based on a convolutional neural network;

[0012] S5: Evaluate the roughness of concrete roughening in the project based on the roughening evaluation classification model.

[0013] Furthermore, in the step S1, the preprocessing steps specifically include:

[0014] 1.1. Denoise the point cloud of the original photo. The point cloud denoising adopts a multi-gradient method. First, limit the area through direct filtering, then use statistical filtering to remove discrete and interfering points, then use least squares filtering to remove noise points according to the distance threshold from the point to the fitting surface, and finally apply bilateral filtering to smooth the denoising result.

[0015] 1.2. Correct the roughening point cloud to keep it horizontal. Use matlab to implement the RANSAC algorithm to fit the best plane of the point cloud to represent the plane of the roughening point cloud, and adjust this plane to achieve the purpose of keeping the roughening point cloud horizontal. After inputting the point cloud data, after processing and calculation, output the normal vector of the fitted plane. Adjust this normal vector to be perpendicular to the horizontal plane, and multiply the point set coordinates by the rotation leveling matrix calculated by matlab to achieve the purpose of correction.

[0016] Furthermore, in the step S2, it specifically includes:

[0017] 2.1. For the scanned point cloud data, separate and store the three-phase coordinates; traverse z the coordinate sequence to obtain the highest point of the plane to be detected , the volume of the envelope body formed by the intersection of the plane where it is located and the irregular surface formed by the point cloud , traverse the z coordinate sequence again to obtain the lowest point of the plane to be detected . The maximum uneven height difference of the plane to be detected ; the volume of the envelope body formed by the intersection of the plane where it is located and the irregular surface formed by the point cloud , and its calculation formula is:

[0018] ;

[0019] ;

[0020] To improve the calculation speed and reduce data redundancy, use the Delaunay triangulation method to calculate:

[0021] ;

[0022] ;

[0023] ;

[0024] ;

[0025] In the formula: Z C is the z coordinate of all points in the detection area, S C is the scanned area of the area to be detected, V CS is the volume of the projection of the point cloud surface of the concrete surface in the area to be detected onto the zero plane, R C is the roughness of the decisive surface to be detected for sand piling, R S is the roughness of the decisive surface to be detected for concrete, , and are the coordinates of the three vertices of the triangle after Delaunay triangulation, , and are the z coordinates of the three vertices of the triangle after Delaunay triangulation;

[0026] 2.2. According to RS and R C The Pearson correlation coefficient is used to determine the contribution rate of a single parameter to the textured roughness, and then the calculation formula of the textured roughness R is obtained; the specific operation is as follows: According to the R S , R C and of the three-dimensional point cloud data of multiple groups of randomly roughened surfaces, perform Pearson detection through spss software to obtain R S and R C；

[0027] ; ; ;

[0028] In the formula: is the contribution rate of the sand pile decisive roughness, is the contribution rate of the concrete decisive roughness, is the correlation of the sand pile decisive roughness, is the correlation of the concrete decisive roughness, R is the textured roughness.

[0029] Furthermore, in the step S3, the specific steps of the evaluation method include:

[0030] 3.1. Design the concrete mix ratio according to the concrete mix ratio design test on the construction site, pour standard specimens, and the volume of the specimens is 150mm×150mm×150mm. In order to achieve the purpose of layered pouring, the concrete specimen is composed of the bottom test block A and the top test block B, and the volumes of the test block A and the test block B are both close to 150mm×150mm×75mm.

[0031] First, pour the test block A, make a mold and mark the position of 78mm of the mold height as a reference. Move the mixed concrete into the mold and stop adding slurry when the center convex surface reaches the horizontal mark. Place the mold on the vibrating table and vibrate for 1 - 2 minutes, and demold after curing for 24 hours. Due to pouring fluctuations, the height of the test block will be between 75 and 80mm. Mark a reserved distance of 3mm to facilitate manual roughening to 75mm to construct test blocks with different degrees of roughening.

[0032] 3.2. Perform roughening treatments with different degrees on the top surface of the bottom test block A to make the volume of the roughened test block A approach 150mm×150mm×75mm and the top is rough; make several specimens with shallow roughening, medium roughening and deep roughening respectively for the roughening treatment; uniformly take pictures to obtain the surface point cloud, and repeat steps S1 and S2 to obtain the textured roughness.

[0033] In the above step 3.2, a method combining a handheld chiseling bit and manual chiseling is used to perform chiseling treatment on the top surface of the bottom test block A to different degrees: the chiseling treatment of each bottom test block A is carried out according to the thickness of the test block A. Since macroscopically, the bonding strength of both overly smooth and overly rough bonding surfaces is relatively low, and at the same time, bonding surfaces with different degrees appear in the training set. For the convenience of production, the specific operations are as follows: For test block A with a thickness less than 76 mm, little or no chiseling is carried out, and the thickness after treatment approaches 75 mm, simulating under-chiseling, with the characteristic that the surface is basically flat or has a small amount of chiseling marks; for test block A with a thickness of 76 - 78 mm, moderate manual chiseling is simulated, and the thickness after treatment approaches 75 mm, simulating moderate chiseling, with the characteristic that the surface chiseling is uniform; for test block A with a thickness greater than 79 mm, excessive manual chiseling is simulated, and the thickness after treatment approaches 75 mm, simulating over-chiseling, with the characteristic that the surface has uneven depths; after the chiseling treatment, the volume of test block A all approaches 150 mm × 150 mm × 75 mm; a concrete group from a smooth surface to over-chiseling is formed.

[0034] 3.3. Pour test block B. After rinsing the chiseled surface of the bottom test block A clean, place it at the bottom of the mold to ensure that the test block is in close contact with the mold bottom plate, then assemble the mold. Then pour the concrete into the mold above the bottom test block A, and then place it on a flat vibrating table and vibrate for 1 - 2 minutes to form a complete standard test block of 150 mm × 150 mm × 150 mm. After that, continue to cure the specimen to the target age.

[0035] 3.4. Conduct the splitting tensile test on the specimen. The chiseling degree is characterized by its interlayer bonding force, and the interlayer bonding force of chiseling is characterized by the splitting tensile strength of the test result; specifically:

[0036] Taking the required splitting tensile strength as the standard value, obtain the variation law of the average splitting tensile strength of each chiseling group at 7 days and 28 days. Corresponding the roughening roughness obtained by photographing the concrete surface to three regions: the low roughening roughness value with a splitting tensile strength lower than the standard value corresponds to the under-chiseling region, the high roughening roughness value with a splitting tensile strength lower than the standard value corresponds to the over-chiseling region, and the rest corresponds to the ideal chiseling region. Different pictures are corresponding to different chiseling degrees according to their roughening roughness, and at the same time, a network training set with this strength standard is obtained. After training, the purpose of obtaining the chiseling degree through its appearance picture without performing the splitting tensile experiment is achieved.

[0037] The above required splitting tensile strength standard value is the design value according to different requirements.

[0038] Furthermore, in the above step S4, the specific method for constructing the convolutional neural network includes:

[0039] 4.1. Build a convolutional neural network model. Based on the Mobilenet V2 network model, introduce the CA attention mechanism to build the Mobilenet-CA model with lightweight, high efficiency, and high image recognition performance. The experimental results on the public dataset dogs-vs-cats show that Mobilenet-CA achieves higher accuracy than Mobilenet V2 in image classification tasks and also inherits the characteristics that are convenient for deployment on mobile devices. At the same time, introduce the VGG16-BN network model similar to the Mobilenet CA network for comparison.

[0040] 4.2. Build a dataset of concrete roughened surface images.

[0041] First, use the original images scanned by the camera as the training set and label the three corresponding partitions of the concrete roughened surface.

[0042] Then, use perspective transformation to correct the tilted target content, and project the image with a tilted perspective onto the horizontal perspective image through the perspective transformation matrix to support effective model training and evaluation, and reduce the interference of building structure differences and environmental changes on the application.

[0043] Secondly, through data augmentation techniques such as random cropping, random rotation, brightness and saturation variation, and mirroring, a variety of training samples are generated. After data augmentation, the constructed dataset has 12,000 image data with classification labels, which can better train the model to adapt to diverse concrete roughened surface image inputs. Obtain a trained classification model for segmenting the ring knife point cloud.

[0044] 4.3. Select appropriate hyperparameters for model training through hyperparameter search. Use the SGD and Adam optimizers for model training and compare their loss functions and convergence speeds. At the same time, experiment with different initial learning rates during model training; the results show that for the Vgg16-BN network model, the best performance is achieved when the initial learning rate is 0.01, with a fast convergence speed and good model learning effect; the same is true for the Mobilenet-CA network model.

[0045] 4.4. Evaluate the two network models through multiple different metrics. Conduct model training on the two network models based on the self-built dataset and optimize the learning rate in combination with the theory of deep learning model training. Finally, evaluate the model performance by evaluating indicators such as training time, prediction time, total number of parameters, and computing unit occupancy. Although the Mobilenet CA network is slightly inferior to the Vgg16-BN network model in some accuracy metrics, its training efficiency and prediction speed are higher than those of the Vgg16-BN network model, and the number of parameters and computational complexity are also lower, making it easy to deploy on terminal devices.

[0046] Further, in step S5, the photos obtained by the depth camera on site are input into the chiseling evaluation and classification model, and chiseling classification and evaluation are automatically performed through deep learning. The chiseling classification is: under-chiseled area, ideal chiseling area, over-chiseled area.

[0047] Advantages of the present invention:

[0048] (1) In view of the subjective randomness of manual evaluation and the complex operation and high cost of mechanical detection in the determination of the roughness of the concrete multi-layer pouring construction blank surface, the method of the present invention provides a method for quickly and accurately detecting surface roughness parameters, realizing the quantitative description of the roughness of the chiseled surface.

[0049] (2) The method of the present invention uses a non-contact method of camera shooting, which will not cause damage to the surface and interior of the concrete and will not affect the strength and durability of dam concrete. Using digital detection, not only the detection speed is very fast but also the measurement result is accurate. It can complete the detection of a large area of concrete surface in a short time and has good versatility. It is not limited to professional and technical personnel who have received complex training, nor is it restricted by the shape and size of the concrete. Its automatic detection results can be fed back to the entire intelligent chiseling process quality control system, facilitating the implementation of intelligent and refined construction.

[0050] (3) The method of the present invention proposes the roughening roughness to characterize the irregularity after the surface roughening treatment of the concrete. It is detected and quantified based on the physical sand piling method through the three-dimensional space reconstruction method. Compared with the sand piling method, the operation is more convenient and the accuracy interference is low. Through point cloud noise reduction processing, more than 90% of the outlier points and heterogeneous points are filtered. The point cloud plane deviation correction algorithm is used to solve the problem that the calculation result is distorted due to the large influence of the scanning device skew or ground uneven settlement on the roughening roughness calculation model. The calculation formula of the roughening roughness applicable to the evaluation of the surface roughening treatment of the concrete blank surface is derived. R of the calculation formula.

[0051] (4) In the method of the present invention for studying the interfacial bonding performance of concrete, the tensile strength measured by the splitting tensile test is used to evaluate the interfacial bonding performance. As a form of indirect tensile test, it is more convenient to operate in studying the interfacial bonding performance of concrete and it is easier to obtain accurate results. At the same time, the model established based on the test is more in line with the engineering practice. During the test process, the casting method is changed by improving the detachable mold, so as to simulate the stress condition of the concrete cast in layers at the construction site and improve the reducibility of the test. Through the test and measurement of 36 specimens in the embodiment of the present invention, the failure forms and the variation law of the splitting tensile strength under different roughnesses are observed: as the roughness of the roughened bonding surface increases, the splitting tensile strength of the specimen gradually increases within a certain range, while when the roughness of the roughened surface is too high, the splitting tensile strength of the specimen begins to decrease. That is, when the concrete is cast in layers, it is necessary to control the roughness of the roughened bonding surface within an ideal range to obtain a higher interfacial bonding strength.

[0052] (5) According to the specifications to be followed in the data acquisition process, including resolution, environmental conditions, shooting distance and angle, etc., the method of the present invention constructs an image dataset of the roughness of the roughened concrete applicable to this project. The image data of the roughened concrete during the data acquisition process comprehensively considers the data quality, diversity and acquisition specifications, and can support effective model training and evaluation. Through data augmentation techniques such as random cropping, random rotation, alienation of brightness and saturation, and mirroring, diverse training samples are generated. After data augmentation, the dataset constructed in the embodiment of the present invention has a total of 12,000 image data with classification labels, which can better train the model to adapt to the diverse input of the roughened concrete surface images and improve the performance and generalization ability of the model.

[0053] (6) The method of the present invention learns two network models, namely the Mobilenet V2 network and the VGG16 network, and an emerging attention mechanism, CA attention. By introducing the CA attention mechanism into the Mobilenet V2 network model, a Mobilenet-CA model is constructed, which can improve the image recognition performance while maintaining lightweight and high efficiency. In the test on the public dataset dogs-vs-cats, the accuracy of the Mobilenet-CA model is 98.25%, which is 0.53% higher than the accuracy of the Mobilenet V2 model at 97.72%. During the model training based on the self-built dataset, appropriate hyperparameters are selected through hyperparameter search for model training. In model training, the SGDm and Adam optimizers are used, and their loss functions and convergence speeds are compared, both showing good optimization effects. At the same time, different initial learning rates are also experimented with during model training. The results show that for the Vgg16-BN network model, the initial learning rate of 0.01 performs best, with a fast convergence speed and good model learning effect; while for the Mobilenet-CA network model, an initial learning rate of 0.01 can also achieve good training results.

[0054] (7) The method of the present invention adjusts the neural network optimization method and hyperparameters according to the accuracy index and the losses of the training set and the test set, enabling the neural network model to have better prediction accuracy. The accuracy of the vgg16-bn model is 99.63%, the precision is 99.64%, and the recall is 99.56%. It performs slightly better than Mobilenet-CA on different accuracy indicators, especially in accurately finding all positive examples, but it requires a longer prediction response time. While Mobilenet-CA has a better performance in terms of prediction response time, only 0.172s, which is 0.074s faster than 0.246s of vgg16-bn, and is suitable for scenarios with high requirements for response speed. The lightweight Mobilenet-CA model has fewer parameters and FLOPs, occupying less memory resources, so it performs better in cases where memory and computing resources are limited.

[0055] (8) In view of the different performance characteristics of the Mobilenet-CA network model and the Vgg16BN network model, the method of the present invention creates an evaluation method for the concrete roughening linear combination model, develops a concrete roughening roughness detection platform with high accuracy and easy operation, and conducts tests in actual engineering applications. Description of the Drawings

[0056] Figure 1 Comparison before and after noise reduction of the concrete surface point cloud;

[0057] Figure 2For the comparison before and after the deviation correction of the concrete surface point cloud;

[0058] Figure 3 For the pouring schematic diagram of test block A and test block B;

[0059] Figure 4 For the relationship diagram between the roughening roughness and the interfacial bonding strength;

[0060] Figure 5 For the under-roughened photo of the classification result;

[0061] Figure 6 For the ideal-roughened photo of the classification result;

[0062] Figure 7 For the over-roughened photo of the classification result. Detailed implementation manners

[0063] The present invention will be further described below in conjunction with the detailed implementation manners. The detailed implementation manners are further explanations of the principle of the present invention, and do not limit the present invention in any way. The same or similar technologies to the present invention do not exceed the protection scope of the present invention.

[0064] The present invention will be further described in detail below with reference to the accompanying drawings.

[0065] A method for evaluating the roughness of concrete roughening based on a convolutional neural network according to the present invention is used for visual recognition and intelligent evaluation of the roughening quality of dam concrete in alpine regions.

[0066] The present invention uses non-contact depth camera technology to accurately capture the surface of the concrete after roughening. Through preprocessing such as point cloud noise reduction and leveling on the two-dimensional visual features and three-dimensional information of the surface depth obtained by the depth camera, a roughening roughness evaluation index for concrete roughening is established, providing a scientific basis for quantitatively evaluating the roughening quality.

[0067] The present invention conducts splitting tensile tests on the test blocks poured in layers, and adopts different roughening degrees at the layered positions, establishes the relationship between the roughening roughness and the interfacial bonding strength, and classifies the roughening degree into under-roughened, ideal-roughened and over-roughened based on this. This classification provides a clear classification standard for subsequent intelligent evaluation.

[0068] The present invention establishes a Mobilenet V2 convolutional neural network model introducing the CA attention mechanism. Based on the camera photos and roughening classification, a data set is established and data enhancement is performed. After the camera takes a photo, the classification and evaluation can be automatically obtained by inputting it into the network.

[0069] The specific steps are as follows:

[0070] Step 1: Preprocess the collected roughened pictures;

[0071] The depth camera is installed above the concrete surface to be roughened for data acquisition, obtaining a point cloud file containing the three-dimensional information of the roughened concrete surface. Affected by environmental factors such as light, temperature, and humidity, as well as the machine equipment itself, the point cloud file contains a large number of noise points distributed in the concrete surface point cloud. At the same time, the depth camera cannot always be exactly perpendicular to the concrete surface, and there is usually a slight inclination. Therefore, noise reduction and leveling processing are required.

[0072] (1.1)Noise reduction is performed on the original photo point cloud. The point cloud noise reduction adopts a multi-gradient method. First, the region is limited by direct filtering, then discrete and interfering points are removed by statistical filtering, and then noise points are removed according to the distance threshold from the point to the fitting surface using least squares filtering. Finally, bilateral filtering is applied to smooth the noise reduction result.

[0073] (1.2)The roughened point cloud is corrected to make it horizontal. The RANSAC algorithm is implemented using Matlab to fit the best plane of the point cloud to represent the plane of the roughened point cloud, and the purpose of keeping the roughened point cloud horizontal is achieved by adjusting this plane. After inputting the point cloud data and processing the calculation, the normal vector of the fitted plane is output. The normal vector is adjusted to be perpendicular to the horizontal plane, and the point set coordinates are multiplied by the rotation leveling matrix calculated by Matlab to achieve the purpose of correction.

[0074] The comparison of the point cloud before and after multi-gradient noise reduction and smoothing processing in this embodiment is as Figure 1 shown; the comparison before and after correction is as Figure 2 shown.

[0075] Based on the preprocessed information, the roughening roughness applicable to the evaluation of the roughening treatment of the concrete blank surface is deduced R calculation formula.

[0076] In order to determine the specific composition of the roughening roughness, in this example, the decisive roughness of sand pile R s and the decisive roughness of concrete R c are analyzed for their correlation with the extreme value of the height difference of the roughened surface h c , where R s and R c represent the average depth of the roughened surface, and h c represents the maximum fluctuation of the rough surface. The Pearson test is used to study the correlation R s and R c .

[0077] Pearson detection can evaluate the strength of the linear relationship between the decisive roughness of the sand pile and the decisive roughness of the concrete with a relatively fast calculation speed. The Pearson detection method of SPSS is used to realize data analysis, and the results are shown in Table 1. h c The number of variable samples participating in the partial correlation analysis of

[0078] Table 1 Analysis results of Pearson detection method:

[0079]

[0080] Descriptive statistics R s and R c are both 100. The mean of R s is 4.6098, which is very close to the mean of R c . R s The standard deviation of R c is 2.0919, and the fluctuation range is similar to that of

[0081] Table 2 is the correlation analysis result.

[0082]

[0083] R s and R c The significance levels of R s and R c and h c all have correlations. R s and h c The Pearson correlation coefficient of R c and h c is 0.769, which is greater than 0.5, indicating a positive correlation and a relatively strong correlation. R c The Pearson correlation coefficient of

[0084] is 0.812, which is also greater than 0.5, indicating a higher positive correlation strength. Therefore, Rc In most cases, it matches the roughness results observed by the naked eye. However, when affected by the roughening tool or process and the roughening is uneven, R c it differs greatly from the actual roughness of the roughened surface. In such cases, R s and R c the deviation values are also large, and the individual R c cannot reflect the actual roughness of the roughened surface. As can be seen from Table 1 and Table 2, R s and R c the significance row level sig (two-tailed) is 0.079, far higher than 0.05, and the Pearson correlation coefficient is 0.251, indicating that there is no obvious linear correlation between the two. That is to say, R s and R c the data are independent and contain different roughness information respectively. Therefore, R s and R c these two data are needed to support the composition of the textured roughness.

[0085] Step 2: By synthesizing various correlation analysis methods such as Pearson detection, derive the textured roughness R calculation formula applicable to the evaluation of the roughening treatment of the concrete blank surface.

[0086] (2.1) For the scanned point cloud data, separate and store the three-phase coordinates; traverse z the coordinate sequence to obtain the highest point of the plane to be detected, and the volume of the envelope formed by the intersection of the plane where is located and the irregular surface formed by the point cloud. Traverse the z coordinate sequence again to obtain the lowest point of the plane to be detected; the maximum uneven height difference of the plane to be detected. The volume of the envelope formed by the intersection of the plane where

[0087]

[0088] is located and the irregular surface formed by the point cloud, and its calculation formula is:

[0089] To improve the calculation speed and reduce data redundancy, the Delaunay triangulation method is used for calculation.

[0090] ;

[0091] ;

[0092] ;

[0093] ;

[0094] In the formula: Z C is the z coordinates of all points in the detection area, S C is the scanning area of the area to be detected, V CS is the volume of the surface point cloud of the concrete surface in the area to be detected projected onto the zero plane, R C is the roughness of the decisive surface to be detected for sand piling, R S is the roughness of the decisive surface to be detected for concrete, , and are the coordinates of the three vertices of the triangle after Delaunay triangulation, , and are the z coordinates of the three vertices of the triangle after Delaunay triangulation.

[0095] (2.2) Determine the contribution rate of a single parameter to the roughening roughness according to the Pearson correlation coefficient of R S and R C , and then obtain the calculation formula of the roughening roughness R ; The specific operation is as follows: According to the R S , R C and of the three-dimensional point cloud data of multiple groups of randomly roughened surfaces, perform Pearson detection through spss software to obtain R S and R C ;

[0096] ; ; ;

[0097] In the formula: is the contribution rate of the decisive roughness for sand piling, is the contribution rate of the decisive roughness for concrete, For the decisive roughness correlation of sand piling, For the decisive roughness correlation of concrete, R Is the textured roughness.

[0098] Step 3: Construct a roughening evaluation method based on the relationship between textured roughness and interlayer bonding strength. The process is schematically shown as Figure 3 shown.

[0099] (3.1) Design a test mix ratio of concrete according to the concrete mix ratio on the construction site and pour standard specimens with a volume of 150mm×150mm×150mm. To achieve the purpose of layered pouring, the concrete specimen is composed of the bottom test block A and the top test block B, and the volumes of both test blocks A and B are close to 150mm×150mm×75mm.

[0100] Pour test block A first, mark the position of 78mm on the mold, move the mixed concrete into the mold, and stop adding slurry when the center convex surface reaches the horizontal mark. Place the mold on the vibrating table and vibrate for 1 - 2 minutes, then demold after curing for 24 hours.

[0101] (3.2) Use a method combining a handheld roughening drill bit and manual roughening to perform roughening treatment on the top surface of the bottom test block A to different degrees. Classify according to the thickness of the bottom test block A. When the thickness does not exceed 76mm, it is shallow roughening; when it is 76 - 78mm, it is medium roughening; when it is above 79mm, it is deep roughening. At the same time, make the volume of test block A all approach 150mm×150mm×75mm. After uniformly taking pictures to obtain the surface point cloud, repeat the previous steps to calculate its textured roughness.

[0102] (3.3) Then pour test block B. After rinsing the roughened surface of the bottom test block A and placing it at the bottom of the mold to ensure tight fit between the test block and the mold bottom plate, assemble the mold, and then pour the concrete above the bottom test block A into the mold. Then place it on the flat vibrating table and vibrate for 1 - 2 minutes to form a complete 150mm×150mm×150mm standard test block, and then continue to cure the specimen to the target age.

[0103] (3.4) Conduct a splitting tensile test on the test block, and use its result to characterize the interlayer bonding strength of roughening. After analysis, according to the change law of the average splitting tensile strength of each group at 7 days and 28 days, as Figure 4 shown, the textured roughness obtained from taking pictures of the concrete surface is corresponding to three zones: under - roughening zone, ideal roughening zone, and over - roughening zone.

[0104] Step 4: Construct a roughening evaluation and classification model based on a convolutional neural network.

[0105] (4.1)Construct a convolutional neural network model. Based on the Mobilenet V2 network model, introduce the CA attention mechanism to construct the Mobilenet-CA model with lightweight, high efficiency, and high image recognition performance. The experimental results on the public dataset dogs-vs-cats show that Mobilenet-CA achieves higher accuracy in image classification tasks compared to Mobilenet V2, and also inherits the characteristics that are convenient for deployment on mobile devices. At the same time, introduce the VGG16-BN network model similar to the Mobilenet CA network for comparison.

[0106] (4.3)Construct a dataset of concrete roughened roughness images. First, use the original images scanned by the camera as the training set and label the corresponding three partitions of the concrete roughened roughness. Then, use perspective transformation to correct the tilted target content, and project the images with a tilted perspective onto the horizontal perspective images through the perspective transformation matrix to support effective model training and evaluation, and reduce the interference of building structure differences and environmental changes on the application. Secondly, through data augmentation techniques such as random cropping, random rotation, brightness and saturation variation, and mirroring, diverse training samples are generated. After data augmentation, the constructed dataset has a total of 12,000 image data with classification labels, which can better train the model to adapt to diverse concrete roughened surface image inputs. Obtain a trained classification model for segmenting the ring knife point cloud.

[0107] (4.3)Select appropriate hyperparameters for model training through hyperparameter search. Use the SGD and Adam optimizers for model training and compare their loss functions and convergence speeds. At the same time, different initial learning rates are also experimented with during model training. The results show that for the Vgg16-BN network model, the best performance is achieved when the initial learning rate is 0.01, with a fast convergence speed and good model learning effect; the same is true for the Mobilenet-CA network model.

[0108] (4.4)Evaluate the two network models through multiple different metrics. Conduct model training on the two network models based on the self-built dataset and optimize the learning rate in combination with the theory of deep learning model training. Finally, evaluate the model performance by evaluating metrics such as training time, prediction time, total number of parameters, and computing unit occupancy. Although the Mobilenet CA network is slightly inferior to the Vgg16-BN network model in some accuracy metrics, its training efficiency and prediction speed are higher than those of the Vgg16-BN network model, and the number of parameters and computational complexity are also lower, making it easy to be deployed on terminal devices.

[0109] Step 5: After the camera scanning is completed, the photos are input into the deep learning model, and the classification and evaluation of the roughening situation are automatically performed through deep learning.Figure 5 For the under-chiseled photos of classification results, Figure 6 For the ideal chiseled photos of classification results, Figure 7 For the over-chiseled photos of classification results.

Claims

1. A method for evaluating the roughness of concrete chiseling based on a convolutional neural network, characterized by including the following steps: S1: Use a depth camera to obtain chiseling photos and preprocess the collected pictures; specifically: 1.

1. Denoise the original photo point cloud; the point cloud denoising adopts a multi-gradient method. First, limit the area through direct filtering, then use statistical filtering to remove discrete and interfering points, and then use least squares filtering to remove noise points according to the distance threshold from the point to the fitting surface. Finally, apply bilateral filtering to smooth the denoising result; 1.

2. Rectify the roughened point cloud to make it horizontal; use matlab to implement the RANSAC algorithm to fit the best plane of the point cloud to represent the plane of the roughened point cloud, and adjust this plane to make the roughened point cloud horizontal; after inputting the point cloud data, process and calculate, and output the normal vector of the fitted plane; adjust this normal vector to be perpendicular to the horizontal plane, and multiply the point set coordinates by the rotation leveling matrix calculated by matlab to achieve the purpose of rectification; S2: Based on the preprocessed information, combined with Pearson detection and various correlation analysis methods, deduce the formula for calculating the roughening roughness R applicable to the evaluation of the roughening treatment of the concrete blank layer surface, and calculate the roughening roughness using the formula for calculating the roughening roughness R; specifically: 2.

1. For the point cloud data obtained by scanning, store the three-phase coordinates separately; traverse the z-coordinate sequence to obtain the highest point Z of the plane to be detected max , Z max The volume V of the envelope formed by the intersection of the plane where it is located and the irregular surface formed by the point cloud ss , traverse the z-coordinate sequence again to obtain the lowest point Z of the plane to be detected min ; the maximum uneven height difference Δh of the plane to be detected; Z max The volume V of the envelope formed by the intersection of the plane where it is located and the irregular surface formed by the point cloud ss , and its calculation formula is: Δh = Z max -Z min ; V ss = ∫∫(Z max - Z c ) dxdy; To improve the calculation speed and reduce data redundancy, use the Delaunay triangulation method for calculation; Where: Z C is the z - coordinate of all points in the detection area, S C is the scanning area of the area to be detected, V CS is the volume of the surface point cloud of the concrete in the area to be detected projected onto the zero plane, R C is the roughness of the decisive surface to be detected for sand piling, R S is the roughness of the decisive surface to be detected for concrete, Y i , Y j and Y k are the y - coordinates of the three vertices of the triangle after Delaunay triangulation, Z i , Z y and Z k are the z - coordinates of the three vertices of the triangle after Delaunay triangulation; 2.

2. Determine the contribution rate of a single parameter to the honing roughness according to the Pearson correlation coefficient of R S and R C , and then obtain the calculation formula of the honing roughness R; the specific operation is as follows: According to the R of the three-dimensional point cloud data of multiple groups of randomly chiseled surfaces S , R C and Δh are subjected to Pearson detection through spss software to obtain R S and R C ; Where: b s is the contribution rate of decisive roughness of sand pile, b c is the contribution rate of decisive roughness of concrete, p s is the correlation of decisive roughness of sand pile, p c is the correlation of decisive roughness of concrete, and R is the roughened roughness; S3: Construct a roughening evaluation standard based on the relationship between roughening roughness and interfacial bonding strength; S4: Construct a roughening evaluation classification model based on a convolutional neural network; S5: Evaluate the roughening roughness of concrete in the project based on the roughening evaluation classification model.

2. The method for evaluating the roughening roughness of concrete based on a convolutional neural network according to claim 1, characterized in that, In the said S3, the construction of the evaluation standard includes: 3.

1. Pour several standard specimens according to the concrete mix ratio on the construction site; the total volume of each standard specimen is 150mm×150mm×150mm, and it is composed of a bottom test block A and a top test block B that are stacked half by half up and down; Pour test block A first; make a mold and mark the position of 78mm of the mold height as a reference, move the mixed concrete into the mold, and stop adding slurry when the center convex surface reaches the marked position; place the mold on the vibrating table and vibrate for 1 - 2 minutes, and demold after curing for 24 hours; 3.

2. Conduct different degrees of roughening treatment on the top surface of the bottom test block A, so that the volume of the roughened test block A approaches 150mm×150mm×75mm and the top is rough; make several specimens with shallow roughening, medium roughening, and deep roughening respectively for the roughening treatment; take unified photos to obtain the surface point cloud, and repeat step S1 and S2 to obtain the roughening roughness; 3.

3. Pour test block B; after washing the roughened surface of the bottom test block A clean, place it at the bottom of the mold to ensure that test block A is closely attached to the mold bottom plate, then assemble the mold, and then pour the concrete above the bottom test block A in the mold; then place it on the flat vibrating table and vibrate for 1 - 2 minutes to form a complete 150mm×150mm×150mm standard specimen, and then continue to cure the specimen to the target age; 3.

4. Conduct a splitting tensile test on the specimen; use its interfacial bonding strength as the standard to characterize the roughening degree, and use the splitting tensile strength of the test result to characterize the interfacial bonding strength of the roughening; specifically: Taking the required splitting tensile strength as the standard value, the variation law of the average splitting tensile strength of each roughening group at 7 days and 28 days is obtained. The roughening roughness obtained by photographing the concrete surface is corresponded to three zones: the low roughening roughness value with splitting tensile strength lower than the standard value corresponds to the under-roughening zone, the high roughening roughness value with splitting tensile strength higher than the standard value corresponds to the over-roughening zone, and the rest corresponds to the ideal roughening zone; each picture is corresponded to the corresponding roughening degree according to its roughening roughness, and at the same time, a network training set with this strength standard is obtained; after training, it is used to obtain the roughening degree of the detection object through picture comparison.

3. The method for evaluating the roughness of concrete roughening based on a convolutional neural network according to claim 2, wherein: In step 3.2, the roughening treatment of each bottom specimen A is selected according to the thickness of specimen A, and the specific operation is as follows: For specimen A with a thickness less than 76 mm, little or no roughening is carried out, and the thickness approaches 75 mm after treatment, simulating under-roughening, and the feature is that the surface is flat or has a small amount of roughening marks; For specimen A with a thickness of 76 - 78 mm, moderate manual roughening is simulated, and the thickness approaches 75 mm after treatment, simulating moderate roughening, and the feature is that the surface roughening is uniform; For specimen A with a thickness greater than 79 mm, excessive manual roughening is simulated, and the thickness approaches 75 mm after treatment, simulating over-roughening, and the feature is that the surface has different depths; After the roughening treatment, the volume of specimen A all approaches 150 mm × 150 mm × 75 mm; a concrete group from smooth surface to over-roughening is formed.

4. The method for evaluating the roughness of concrete roughening based on a convolutional neural network according to claim 2, wherein In S4, the specific method for constructing the convolutional neural network includes: 4.

1. Construct a convolutional neural network model; Based on the Mobilenet V2 network model, introduce the CA attention mechanism to construct the Mobilenet-CA model; At the same time, introduce the VGG16-BN network model similar to the Mobilenet CA network for comparison; 4.

2. Construct an image dataset of concrete roughening roughness; Taking the original pictures of standard specimens scanned by a depth camera as the training set and marking the corresponding three partitions of the concrete roughening roughness; Use perspective transformation to correct the tilted target content, project the image with a tilted perspective to a horizontal perspective image through the perspective transformation matrix to support effective model training and evaluation, and reduce the interference of building structure differences and environmental changes on the application; Through random cropping, random rotation, alienation of brightness and saturation, and mirror data augmentation, generate diverse training samples, construct image data with classification labels, train the model to adapt to diverse concrete roughened surface image inputs, and obtain a trained classification model for segmenting the core cutterhead point cloud; 4.

3. Select appropriate hyperparameters for model training through hyperparameter search; Use the SGD and Adam optimizers for model training, and compare their loss functions and convergence speeds; At the same time, experiment with different initial learning rates during model training; 4.

4. The two network models were evaluated through multiple different metrics; model training for the two network models was carried out based on a self-built dataset, and the learning rate was optimized in combination with the deep learning model training theory; the model performance was evaluated by assessing metrics such as training time, prediction time, total number of parameters, and computing unit occupancy; the Mobilenet CA network was selected to construct a chiseling evaluation classification model based on a convolutional neural network.

5. The method for evaluating the roughness of concrete chiseling based on a convolutional neural network according to claim 1, wherein In S5, the photos obtained on-site through a depth camera are input into the chiseling evaluation classification model, and chiseling classification and evaluation are automatically performed through deep learning. The chiseling classifications are: under-chiseled area, ideal chiseling area, over-chiseled area.

Images