Drainage pipeline health index construction method

Abstract

The invention provides a drainage pipeline health index construction method. The method comprises the following steps: establishing a drainage pipeline historical database; cleaning the daily pipeline inspection video to obtain detection data of the target pipeline; extracting a key frame from the detection video, screening out an image containing a target pipeline defect, adding the image into a defect classification database, and performing feature extraction on a drainage pipeline defect image to be evaluated by using a convolutional neural network model; defect information including category and size data is obtained; constructing a drainage pipeline health state evaluation index system; and calculating each index value based on the defect information data and the drainage pipeline health state index system to obtain the drainage pipeline health state. According to the invention, health state grades are divided according to the index value distribution condition, and the evaluation result greatly reduces the applicability error; the efficiency and accuracy of daily operation and maintenance and planned replacement decision making are improved, and a powerful guarantee is provided for efficient and accurate operation and maintenance of an urban drainage pipe network and urban green and safe construction.

Description

Technical Field

[0001] The invention relates to the technical field of pipeline safety risk assessment, and in particular to a method for constructing a drainage pipeline health index. Background Art

[0002] With the rapid development of cities, the length and complexity of drainage pipes are increasing; drainage pipes need to be regularly maintained after they are laid. The current urban pipe network maintenance strategy tends to be a passive firefighting operation and maintenance strategy, which has a heavy financial burden but poor performance. In the face of these problems, improving the operation and maintenance efficiency of municipal pipe networks has become a key task in promoting the development of cities in a green and safe direction. Accurately evaluating and predicting the operating status of the pipeline system can timely discover the structural and functional defects of the drainage pipes, predict the operation risks of the drainage pipes, and guide the optimization and operation and maintenance of the drainage system. First of all, it is necessary to establish a comprehensive and clear health assessment system.

[0003] At present, existing technologies mainly use methods such as AHP tomography analysis, expert consultation weight method (Delphi method), superior and inferior solution distance method, principal component analysis, fuzzy comprehensive evaluation, factor analysis, entropy weight method, etc. to evaluate the health status of pipelines; however, most of these methods rely on subjective weighting and have poor objectivity.

[0004] Secondly, there are some common drainage pipeline health status evaluation standards at home and abroad. For example, PACP issued by NASSCO in the United States uses the average defect grade to obtain the condition grade. The average grade cannot reflect the actual condition of the pipeline; SRM-4 published by WRc in the United Kingdom calculates the maximum defect score of the pipe section and compares it with the preset threshold to obtain the condition grade. This method uses the problem of the most serious local point of the pipe section to represent the defect of the entire pipe section, which is a generalization; and CJJ181-2012 issued by the Ministry of Housing and Urban-Rural Development of China uses the larger value of the average defect score and the maximum defect score to compare with the preset threshold to obtain the defect parameter, and then calculates the repair index RI. However, experience has shown that the RI of an intact pipe section may still be repaired at the second level. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a method for constructing a health index for a drainage pipe. The present invention has the characteristics of extremely high practicality, versatility, objectivity and rationality, is suitable for drainage pipes in all situations, and does not require subjective scoring.

[0006] The technical solution of the present invention is: a method for constructing a drainage pipeline health index, comprising the following steps: S1) Establish a historical database of drainage pipelines; S2) Detect the target pipeline according to the operation and maintenance requirements and obtain the detection video of the target pipeline; and perform data cleaning and alignment processing on the target pipeline detection data to obtain the detection video of the pipeline to be evaluated; S3), automatically extract key frames from the inspection video and screen out high-resolution images containing target pipeline defects; S4), adding the drainage pipe defect image to be evaluated into the defect classification database, and using the convolutional neural network model to extract features of the drainage pipe defect image to be evaluated; obtaining the defect feature image, and obtaining effective defect information, that is, category and size quantitative data through the defect feature image; S5) Constructing a drainage pipeline health status assessment index system based on the defect feature image of the drainage pipeline itself; S6), based on the defect quantification data and the defined drainage pipe health status assessment index system, calculate the value of each index to obtain the drainage pipe health status value, and classify and store the latest detection results in the drainage pipe history database.

[0007] Preferably, in step S1), the drainage pipe historical database includes drainage pipe body basic data, test result data and historical operation and maintenance decision data.

[0008] Preferably, in step S1), the drainage pipe ontology basic database includes a structure database, a function database and a defect classification database.

[0009] Preferably, in step S4), the convolutional neural network model is an improved VGGNet-4 network model, and the improved VGGNet-4 network model includes an input layer, a convolution block, a fully connected layer and an output layer; each of the convolution blocks has a corresponding convolution layer, and each of the convolution layers uses a 3x3 convolution kernel.

[0010] Preferably, in step S4), the convolutional neural network model has four convolutional layers and two fully connected layers, and corresponding pooling layers are connected between adjacent convolutional layers; a dropout layer is used after each fully connected layer to prevent overfitting; A flattening layer is also connected between the first fully connected layer and the fourth convolutional layer.

[0011] Preferably, in step S4), the improved VGGNet-4 network model is used to extract features of the drainage pipe defect image to be evaluated, specifically: Input the drainage pipe defect image to be evaluated through the input layer X ; The defect image X The low-level feature image is obtained through the first convolutional layer Z 1 ; Low-level feature image Z 1 After activation function Then pass through the first pooling layer to obtain the pooled feature map; After repeated processing through the convolutional layer, pooling layer, and fully connected layer, the output layer generates the prediction result.

[0012] Preferably, in step S4), the improved VGGNet-4 network model optimizes model parameters by back propagation algorithm; and based on the chain rule, the gradient of the loss function relative to each parameter is calculated, and the parameters of the model are updated according to the gradient, which specifically includes the following steps: S41), starting from the output layer, back propagating layer by layer, the output layer error gradient for: In the formula, is the true label; is the predicted probability; S42), the gradient of the second fully connected layer is: In the formula, W 2 is the weight matrix of the second fully connected layer, A 2 is the output of the first fully connected layer; express ReLU The derivative of L represents the loss function; represents the output of fully connected layer 2; represents the error gradient of the fully connected layer 2; S43), the gradient expression of the first fully connected layer is: in, is the weight matrix of the first fully connected layer, A 1 is the output of the flat layer; express ReLU The derivative of represents the output of the fully connected layer 1; represents the error gradient of the fully connected layer 1; S44), the gradient of the fourth convolutional layer is: In the formula, is the output of the third pooling layer. The gradient of the pooling layer is propagated through the "max pooling mask" in back propagation. Continue to calculate the weight gradient of each convolutional layer layer by layer, and use the chain rule to propagate the gradient back to the input layer. In the formula, is the error gradient of the current layer; is the weight matrix of the current layer; is the output of the previous layer; The weight gradient of each layer is obtained by combining the input of the previous layer and the error signal of this layer. δ Calculated; after calculating the gradients of all layers, the model parameters are updated using the gradient descent method: In the formula, represents the model weight; represents the learning rate; Repeat the above steps until the model converges on the training set.

[0013] Preferably, in step S4), the gap between the predicted probability and the true label is measured by using a cross entropy loss function, and the expression of the cross entropy loss function is: In the formula, C is the number of categories, It is i The true labels of defect categories, It is i The predicted probability of each defect class.

[0014] Preferably, in step S5), the drainage pipeline health status assessment index system is divided into two categories: one is applicable to the maintenance and replacement of different departments, including drainage pipeline maintenance indicators PMI and drainage pipe replacement indicators PRI ,in, PMI Help determine if the pipeline needs repair, PRI Helps decide whether the entire pipe section needs to be replaced, focusing on the severity of the defect; a type of structure suitable for maintenance and replacement is the same department, using a single indicator PRI Assess the overall health of your pipeline and make decisions regarding long-term monitoring and maintenance plans.

[0015] As a preferred embodiment, in step S5), the drainage pipe maintenance index PMI It is expressed as: In the formula, Indicates defects k The level value of Indicates the number of defect levels; Indicates pipeline importance; The drainage pipe replacement index PRI It is expressed as: In the formula, Indicates defects k The level value of Indicates defects k The area of A Indicates the inner surface area or cross-sectional area of ​​the pipe section to be calculated; PRI The application scenarios and goals are different in the two types of departmental architectures. Among them, in the architecture where maintenance and replacement are in the same department, PRI Responsible for maintenance and update scenarios, used to evaluate the overall health status of pipelines. This indicator is not only applicable to daily maintenance and upkeep needs, but also can provide a basis for decision-making for large-scale updates and renovations. Maintenance and replacement are in the architecture of different departments. PRI It is suitable for scenarios with division of labor and responsibilities, focusing on pipeline replacement needs and providing a basis for replacement decisions.

[0016] The beneficial effects of the present invention are: 1. The present invention is constructed based on defect feature information to reflect the actual service performance status of the pipeline; and the present invention can capture individual differences and comprehensively consider all the special considerations of the target drainage pipeline; 2. The present invention divides the health status level according to the calculated index value distribution and its corresponding pipeline physical and mechanical connotation. The index values ​​are continuous and objective, and the evaluation results greatly reduce the applicability error; 3. The present invention constructs health indicators based on defect levels and sizes, which are not restricted by regions and specifications and can be applied to different standards such as CJJ181, PACP or WRc. This adaptability enhances the versatility of the evaluation system and provides wider support for practical applications.

[0017] 4. The evaluation method indicators of the present invention have clear physical meanings and good representativeness. They are divided into two evaluation systems according to the departmental structure, both of which can guide maintenance decisions and update and replacement needs, and concisely and clearly improve the efficiency and accuracy of management personnel in operation and maintenance and replacement decisions, providing strong guarantees for the efficient and accurate operation and maintenance of urban drainage networks and the green and safe construction of cities. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flow chart of the method of the present invention; Figure 2 This is a framework diagram of the historical operation and maintenance decision database of the drainage pipeline of the present invention; Figure 3 Schematic diagram of the framework of the convolutional neural network model of the present invention; Figure 4 Schematic diagram of the drainage pipeline health status indicator evaluation system of the present invention; Figure 5 A schematic diagram for delineating the area of ​​a rupture defect of the present invention; Figure 6 A schematic diagram for delineating the area of ​​a deformation defect of the present invention; Figure 7 A schematic diagram for delineating the area of ​​corrosion defects of the present invention; Figure 8 It is a schematic diagram for delineating the area of ​​the misalignment defect of the present invention; Fig. 9 A schematic diagram for delineating the area of ​​undulation defects of the present invention; Fig.10 A schematic diagram for delineating the area of ​​a disjointed defect of the present invention; Fig.11 A schematic diagram of the area demarcation of the interface material shedding defect of the present invention; Fig.12 It is a schematic diagram for delineating the area of ​​the concealed connection defect of the branch pipe of the present invention; Fig.13 A schematic diagram for delineating the area of ​​a foreign body penetration defect of the present invention; Fig.14 A schematic diagram for delineating the area of ​​a leakage defect of the present invention; Fig.15 A schematic diagram for delineating the area of ​​deposition defects of the present invention; Fig.16 It is a schematic diagram for delineating the area of ​​scaling defects of the present invention; Fig.17 A schematic diagram for delineating the area of ​​an obstacle defect of the present invention; Fig.18 It is a schematic diagram for delineating the area of ​​residual wall and dam root defects of the present invention; Fig.19 This is a schematic diagram of the area delineation of the tree root defect of the present invention. DETAILED DESCRIPTION

[0019] The specific implementation of the present invention will be further described below in conjunction with the accompanying drawings: like Figure 1 As shown, this embodiment provides a method for constructing a drainage pipeline health indicator, comprising the following steps: S1), establishing a drainage pipeline historical database, and cleaning and aligning the data in the drainage pipeline historical database; In this embodiment, the drainage pipe historical database includes drainage pipe body basic data, detection result data and historical operation and maintenance decision data.

[0020] like Figure 2 As shown, the drainage pipe ontology basic database includes a structure database, a function database and a defect classification database.

[0021] In this embodiment, the structural status database includes pipeline physical data A 1 , Pipeline function data A2 、Pipeline environment A 3 ; Wherein, the pipeline physical data A 1 Including pipe length A 11 , Pipe diameter A 12 、Pipe material A 13 、Tube age A 14 , average burial depth A 15 , Pipeline slope A 16 ; The pipeline function data A 2 Including maintenance strategy A 21 Repair strategy A 22 , Update strategy A 23 ; The pipeline environment A 3 Including sewage type A 31 、Groundwater A 32 、Soil Type A 33 , cushion condition A 34 、Frost Factor A 35 、Other public facilities A 36 Traffic flow A 37 .

[0022] In this embodiment, the functional status database includes water factor data B 1 and non-hydraulic factor data B 2 ;in, The water factor data B 1 Including flow B 11 Flow rate B 12 , Load state B 13 , Pipeline slope length B 14 ; The non-hydraulic factor data B 2 Including random blocking B 21 ; Debris, grease and floating matter B 22 ; Pump station and screening equipment failure B 23 ; Historical operation and maintenance effect B 24 ; The defect classification database includes structural defect data C 1 , Functional defect data C 2 ; Wherein, the structural defect data C 1 Including rupture C 11 , deformation C 12 、Corrosion C 13 、Wrong mouth C 14 , Ups and downs C 15 , Disconnect C 16 , Interface material falls off C 17 、Branch pipe concealed connection C 18、Foreign body penetration C 19 , filtration C 20 ; The functional defect data C 2 Including deposition C 21 , scaling C 22 、Obstacle C 23 , Remaining walls and dam roots C 24 , tree root C 25 、Scum C 26 .

[0023] In this embodiment, the test result data includes customer data D 1 and pipeline data D 2 ; Wherein, the customer data D 1 Including customer relationship data 11 、Enterprise Resource PlanningD 12 , Supply Chain Management 13 .

[0024] The pipeline data D 2 Including pipeline internal condition data D 21 , Pipeline external condition data D 22 、Environmental data D 23 .

[0025] The historical operation and maintenance decision data includes maintenance plans and measures E 1 , Maintenance Records E 2 , Repair and Improvement Plan E 3 , Operation and maintenance decision-making process record E 4 , budget and resource allocation records 5 .

[0026] S2) Inspect the target pipeline according to the operation and maintenance requirements, and obtain the closed-circuit television inspection video of the target pipeline; and perform data cleaning and alignment processing on the target pipeline inspection data to obtain the inspection video of the pipeline to be evaluated; S3), extracting key frames from the inspection video and screening out high-resolution images containing target pipeline defects; S4), adding the drainage pipe defect image to be evaluated to the defect classification database, and using the convolutional neural network model to extract features of the drainage pipe defect image to be evaluated; obtaining a defect feature image, and obtaining defect classification and quantification data through the defect feature image; In this embodiment, Figure 3As shown in Table 1, the convolutional neural network model is an improved VGGNet-4 network model, which includes an input layer, a convolutional block, a fully connected layer, and an output layer; each of the convolutional blocks has a corresponding convolutional layer, and each of the convolutional layers uses a 3x3 convolutional kernel. The convolutional neural network model has four convolutional layers, and corresponding pooling layers are also connected between adjacent convolutional layers. The convolutional neural network model has two fully connected layers of 4096 neurons, and a dropout layer Dropout is used after each fully connected layer to prevent overfitting. A flat layer is also connected between the first fully connected layer and the fourth convolutional layer.

[0027] As a preferred embodiment of this invention, the improved VGGNet-4 network model is used to extract features of the drainage pipe defect image to be evaluated, specifically: The drainage pipe defect image X to be evaluated is input through the input layer; the defect image X is passed through the first convolution layer to obtain a low-level feature image Z 1 ; Low-level feature image Z 1 After activation function Then pass through the first pooling layer to obtain the pooled feature map; After repeated processing through the convolutional layer, pooling layer, and fully connected layer, the output layer generates the prediction result.

[0028] As a preferred embodiment of this embodiment, in step S4), the improved VGGNet-4 network model optimizes model parameters by back propagation algorithm; and based on the chain rule, the gradient of the loss function relative to each parameter is calculated, and the parameters of the model are updated according to the gradient, which specifically includes the following steps: S41), starting from the output layer, back propagating layer by layer, the output layer error gradient for: In the formula, is the true label; is the predicted probability; S42), the gradient of the second fully connected layer is: In the formula, W 2 is the weight matrix of the second fully connected layer, A 2 is the output of the first fully connected layer; express ReLU The derivative of L represents the loss function; represents the output of fully connected layer 2; represents the error gradient of the fully connected layer 2; T is the transpose operation; S43), the gradient expression of the first fully connected layer is: In the formula, is the weight matrix of the first fully connected layer, A 1 is the output of the flat layer; express ReLU The derivative of represents the output of the fully connected layer 1; represents the error gradient of the fully connected layer 1; S44), the gradient of the fourth convolutional layer is: In the formula, is the output of the third pooling layer. The gradient of the pooling layer is propagated through the "max pooling mask" in back propagation. Continue to calculate the weight gradient of each convolutional layer layer by layer, and use the chain rule to propagate the gradient back to the input layer. In the formula, is the error gradient of the current layer; is the weight matrix of the current layer; is the output of the previous layer; The weight gradient of each layer is obtained by combining the input of the previous layer and the error signal of this layer. δ Calculated; after calculating the gradients of all layers, the model parameters are updated using the gradient descent method: In the formula, represents the model weight; represents the learning rate; Repeat the above steps until the model converges on the training set.

[0029] As a preferred embodiment of this invention, the gap between the predicted probability and the true label is measured by using a cross entropy loss function, and the expression of the cross entropy loss function is: In the formula, C is the number of categories, It is i The true labels of defect categories, It is i The predicted probability of each defect class.

[0030] Table 1 is a parameter table of the improved VGGNet-4 network model in this embodiment S5) Constructing a drainage pipeline health status assessment index system based on the defect feature image of the drainage pipeline itself; like Figure 4 As shown, the drainage pipeline health status assessment index system is divided into two categories: one is applicable to maintenance and replacement, which is the structure of different departments, including drainage pipeline maintenance indicators PMI and drainage pipe replacement indicators PRI ,in, PMI Help determine if the pipeline needs repair, PRI Helps decide whether the entire pipe section needs to be replaced, focusing on the severity of the defect; a type of structure suitable for maintenance and replacement is the same department, using a single indicator PRI Evaluate the overall health of the pipeline and make decisions involving long-term monitoring and maintenance plans. Structural and functional defect evaluation indicators should be calculated separately. Among them, the drainage pipeline maintenance indicator PMI It is expressed as: In the formula, Indicates defects k The level value of Indicates the number of defect levels; The importance of a pipeline can be determined by comprehensively considering factors such as the pipeline volume, location, and nearby important structures.

[0031] The drainage pipe replacement index PRI It is expressed as: In the formula, Indicates defects k The level value of Indicates defects k The area of A Indicates the inner surface area or cross-sectional area of ​​the pipe segment to be calculated.

[0032] PRI The application scenarios and goals are different in the two types of departmental architectures. Among them, in the architecture where maintenance and replacement are in the same department, PRI Responsible for maintenance and update scenarios, used to evaluate the overall health status of pipelines. This indicator is not only applicable to daily maintenance and upkeep needs, but also can provide a basis for decision-making for large-scale updates and renovations. Maintenance and replacement are in the architecture of different departments. PRI It is suitable for scenarios with division of labor and responsibilities, focusing on pipeline replacement needs and providing a basis for replacement decisions.

[0033] In the framework where maintenance and replacement are the same department, the health status indicator of the drainage pipe PRI It can directly reflect the comprehensive status of the pipe section. PRI The lower the value, the better the overall condition. PRI The value is large and the growth rate is fast. The repair cost is much higher than the cost of updating the pipeline. At this time, repair is not considered and the pipeline is directly replaced.

[0034] The following is a demonstration of the calculation method for each defect size when calculating indicators. In this embodiment, the definition and classification of CJJ181 defects are taken as an example. The present invention constructs health indicators based on defect levels and sizes, is not restricted by regions and specifications, and can be applied to different standards such as CJJ181, PACP or WRc. Figure 5 (a) is a schematic diagram of radial cracks in the pipeline. 11 When calculating the PRI, Take the area of ​​the rectangle formed by the farthest end around the rupture. The rupture that is not connected through should be counted as a separate defect. See the attached Figure 5 (b) in the above equation, A Take the inner surface area of ​​the ruptured pipe section; As attached Figure 6 (a) is a schematic diagram of the pipeline deformation. 12 When calculating the PRI, Take the inner surface area of ​​the pipe before deformation where the shape variation occurs, see the attached Figure 6 (b) in the above equation, A Take the inner surface area of ​​the deformed pipe section.

[0035] As attached Figure 7 This is a schematic diagram of pipeline corrosion. 13 When calculating the PRI, Take the inner surface area of ​​the pipe where corrosion occurs, A Take the inner surface area of ​​the pipe section where corrosion occurs.

[0036] As attached Figure 8 This is a schematic diagram of a misaligned pipe. Misaligned pipe C 14 When calculating the PRI, Take the inner surface area of ​​the two pipe sections where the misalignment occurs, A Take the inner surface area of ​​all pipe sections between the two inspection wells where the misalignment occurs.

[0037] As attached Fig. 9 This is a schematic diagram of the pipeline undulating. The undulation C 15 When calculating the PRI, Take the inner surface area of ​​the two pipe sections where the undulation occurs, A Take the inner surface area of ​​all pipe sections between the two inspection wells where the misalignment occurs.

[0038] As attached Fig.10 This is a schematic diagram of a pipeline disconnection.16 When calculating the PRI, Take the inner surface area of ​​the pipe area under the two pipe sections and the unjoined length, A Take the inner surface area of ​​all pipe sections between the two inspection wells where the misalignment occurs.

[0039] As attached Fig.11 This is a schematic diagram of the interface material falling off in the pipeline. 17 When calculating the PRI, Take the inner surface area of ​​the pipe area where the falling object is located at the interface length, A Take the inner surface area of ​​the two pipe sections before and after the detached object.

[0040] As attached Fig.12 This is a schematic diagram of a pipeline with a branch pipe concealed connection. 18 When calculating the PRI, Take the inner surface area of ​​the main pipe where the branch pipe and the main pipe overlap, A Take the inner surface area of ​​the concealed pipe section of the branch pipe.

[0041] As attached Fig.13 This is a schematic diagram of foreign matter penetrating into a pipeline. 19 When calculating the PRI, Take the inner surface area of ​​the pipe where the foreign body and the pipe overlap. A Take the inner surface area of ​​the pipe section where foreign matter penetrates.

[0042] As attached Fig.14 This is a schematic diagram of a pipeline leak. 20 When calculating the PRI, Take the inner surface area of ​​the pipe where the leakage occurs, A Take the inner surface area of ​​the pipe section where leakage occurs.

[0043] As attached Fig.15 A schematic diagram of the deposition in the pipeline is shown in Figure 1. Fig.15 (a), deposition C 21 When calculating the PRI, Take the water section loss area caused by sedimentation and siltation, A Take the cross-sectional area of ​​the water flow section of the pipe section where sedimentation occurs, see the attached Fig.15 (b) in .

[0044] As attached Fig.16 The schematic diagram of scaling in the pipeline is shown in Figure 2. Fig.16 (a) Scaling C 22 When calculating the PRI, Take the water flow section loss area caused by scaling, A Take the cross-sectional area of ​​the water flow section of the pipe section where sedimentation occurs, see the attached Fig.16(b) in the figure.

[0045] As attached Fig.17 This is a schematic diagram of an obstacle in the pipeline, such as Fig.17 (a) in the figure, obstacle C 23 When calculating the PRI, Take the water section loss area caused by obstacles, A Take the cross-sectional area of ​​the water flow section of the pipe section where the obstacle occurs, such as Fig.17 (b) in the figure.

[0046] As attached Fig.18 (a) is a schematic diagram of the residual wall and dam root appearing in the pipeline. The residual wall and dam root C 24 When calculating the PRI, Take the water-passing section loss area caused by the residual wall and dam root, A Take the cross-sectional area of ​​the water-passing section of the residual wall and dam root pipe section, such as Fig.18 As shown in (b) in .

[0047] As attached Fig.19 This is a schematic diagram of the tree roots appearing in the pipeline, such as Fig.19 (a) in the figure, the root C 25 When calculating the PRI, Take the water loss area caused by tree roots, A Take the cross-sectional area of ​​the water-passing section of the root section of the tree, see the attached Fig.19 (b) in .

[0048] S6), based on the defect quantification data and the defined drainage pipe health status assessment index system, calculate the value of each index to obtain the drainage pipe health status value, and classify and store the latest detection results in the drainage pipe history database.

[0049] In this embodiment, there are 50 pipelines. The reinforced concrete pipe sections are connected into straight sections. The basic data of the pipeline are shown in Table 2. The pipeline defect detection results and pipeline data related to index calculation are shown in Table 3.

[0050] This example takes the pipe diameter, a parameter that represents the importance of pipes in CJJ-181, as an example. ,in is the pipe diameter, is the reference diameter, in this example ; Table 2 Basic data of pipeline Table 3 Pipeline defect quantitative data The repair index of structural defects PMIS for: The maintenance index of functional defects PMI F for: Structural Defect Replacement Index PRI S for: Functional defect replacement index PRI F for: The above embodiments and descriptions are only for illustrating the principles and best embodiments of the present invention. Without departing from the spirit and scope of the present invention, the present invention may be subject to various changes and improvements, all of which fall within the scope of the present invention to be protected.

Claims

1. A method for constructing a drainage pipeline health index, characterized in that: The steps include: S1) Establish a historical database of drainage pipelines; S2) Detect the target pipeline according to the operation and maintenance requirements and obtain the detection video of the target pipeline; and perform data cleaning and alignment processing on the target pipeline detection data to obtain the detection video of the pipeline to be evaluated; S3), extracting key frames from the inspection video and screening out high-resolution images containing target pipeline defects; S4), adding the drainage pipe defect image to be evaluated into the defect classification database, and using the convolutional neural network model to extract features of the drainage pipe defect image to be evaluated; obtaining the defect feature image, and obtaining effective defect information, that is, category and size quantitative data through the defect feature image; S5) Constructing a drainage pipeline health status assessment index system based on the defect feature image of the drainage pipeline itself; S6), based on the defect quantification data and the defined drainage pipe health status assessment index system, calculate the value of each index to obtain the drainage pipe health status value, and classify and store the latest detection results in the drainage pipe history database.

2. A method for constructing a drainage pipeline health index according to claim 1, characterized in that: In step S1), the drainage pipeline historical database includes drainage pipeline basic data, test result data and historical operation and maintenance decision data.

3. A method for constructing a drainage pipeline health index according to claim 2, characterized in that: In step S1), the drainage pipe ontology basic database includes a structure database, a function database and a defect classification database.

4. A method for constructing a drainage pipeline health index according to claim 1, characterized in that: In step S4), the convolutional neural network model is an improved VGGNet-4 network model, and the improved VGGNet-4 network model includes an input layer, a convolution block, a fully connected layer and an output layer; each of the convolution blocks has a corresponding convolution layer, and each of the convolution layers uses a 3x3 convolution kernel.

5. A method for constructing a drainage pipeline health index according to claim 4, characterized in that: In step S4), the convolutional neural network model has four convolutional layers and two fully connected layers, and corresponding pooling layers are connected between adjacent convolutional layers; a dropout layer is used after each fully connected layer to prevent overfitting; A flattening layer is also connected between the first fully connected layer and the fourth convolutional layer.

6. A method for constructing a drainage pipeline health index according to claim 5, characterized in that: In step S4), the improved VGGNet-4 network model is used to extract features of the drainage pipe defect image to be evaluated, specifically: The drainage pipe defect image X to be evaluated is input through the input layer; the defect image X is passed through the first convolution layer to obtain a low-level feature image Z 1; Low-level feature images Z 1 After activation function Then pass through the first pooling layer to obtain the pooled feature map; Repeat all the convolutional layers, pooling layers, and fully connected layers, and finally the output layer generates the prediction result.

7. A method for constructing a drainage pipeline health index according to claim 6, characterized in that: In step S4), the improved VGGNet-4 network model optimizes the model parameters by back propagation algorithm; and based on the chain rule, the gradient of the loss function relative to each parameter is calculated, and the parameters of the model are updated according to the gradient, which specifically includes the following steps: S41), starting from the output layer, back propagating layer by layer, the output layer error gradient for: In the formula, is the true label; is the predicted probability; S42), the gradient of the second fully connected layer is: In the formula, W 2 is the weight matrix of the second fully connected layer, A 2 is the output of the first fully connected layer; express ReLU The derivative of L represents the loss function; represents the output of fully connected layer 2; represents the error gradient of the fully connected layer 2; T is the transpose operation; S43), the gradient expression of the first fully connected layer is: in, is the weight matrix of the first fully connected layer, A 1 is the output of the flat layer; express ReLU The derivative of represents the output of the fully connected layer 1; represents the error gradient of the fully connected layer 1; S44), the gradient of the fourth convolutional layer is: In the formula, is the output of the third pooling layer. The gradient of the pooling layer is propagated through the "max pooling mask" in the back propagation. Continue to calculate the weight gradient of each convolutional layer layer by layer, and use the chain rule to propagate the gradient back to the input layer. In the formula, is the error gradient of the current layer; is the weight matrix of the current layer; is the output of the previous layer; The weight gradient of each layer is obtained by combining the input of the previous layer and the error signal of this layer. δ Calculated; after calculating the gradients of all layers, the model parameters are updated using the gradient descent method: In the formula, represents the model weight; represents the learning rate; Repeat the above steps until the model converges on the training set.

8. A method for constructing a drainage pipeline health index according to claim 7, characterized in that: In step S4), the gap between the predicted probability and the true label is measured by using the cross entropy loss function, and the expression of the cross entropy loss function is: In the formula, C is the number of categories, It is i The true labels of defect categories, It is i The predicted probability of each defect class.

9. A method for constructing a drainage pipeline health index according to claim 1, characterized in that: In step S5), the drainage pipeline health status assessment index system is divided into two categories: one is applicable to the maintenance and replacement of different departments, including drainage pipeline maintenance indicators PMI and drainage pipe replacement indicators PRI ,in, PMI Used to determine whether the pipeline needs maintenance. PRI Used to decide whether the entire pipe section needs to be replaced, focusing on the severity of the defect; Another type of structure suitable for maintenance and replacement is the same department, using a single indicator PRI , assessing the comprehensive health status of the pipeline, involving decision-making on long-term monitoring and maintenance plans.

10. A method for constructing a drainage pipeline health index according to claim 9, characterized in that: In step S5), the drainage pipe maintenance index PMI It is expressed as: In the formula, Indicates defects k The level value of Indicates the number of defect levels; Indicates pipeline importance; The drainage pipe replacement index PRI It is expressed as: In the formula, Indicates defects k The level value of Indicates defects k The area of A Indicates the inner surface area or cross-sectional area of ​​the pipe section to be calculated; The drainage pipe replacement index PRI The application scenarios and goals are different in the two types of departmental architectures. Among them, in the architecture where maintenance and replacement are in the same department, PRI Responsible for maintenance and update scenarios, used to evaluate the overall health status of pipelines. This indicator is not only applicable to daily maintenance and upkeep needs, but also can provide a basis for decision-making for large-scale updates and renovations. Maintenance and replacement are in the architecture of different departments. PRI It is suitable for scenarios with division of labor and responsibilities, focusing on pipeline replacement needs and providing a basis for replacement decisions.

Images