A method and system for real-time defect monitoring of a hydraulic tunnel

Abstract

The application discloses a kind of hydraulic tunnel real-time defect monitoring method and system, the method utilizes fuzzy algorithm to calculate out the danger value of the hydraulic tunnel to be detected according to the real-time environmental factors of the hydraulic tunnel to be detected, and according to the danger value of the hydraulic tunnel to be detected obtained by calculation, for the real-time acquisition of the hydraulic tunnel image selects appropriate convolution layer and autonomous intention layer, so that the tunnel image recognition model can accurately identify the defects of the real-time image of the hydraulic tunnel to be detected, improve the accuracy of defect identification, and according to the danger value calculated according to environmental factors and defect identification result, determine the maintenance strategy of the hydraulic tunnel to be detected, improve the accuracy and effectiveness of maintenance strategy, solve the problem that the prior art only tunnel defect identification is carried out to the collected tunnel image data, while lacking the influence of internal environmental factors of hydraulic tunnel, leading to unable to accurately identify the defects of hydraulic tunnel, and then reduce the accuracy and effectiveness of maintenance strategy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of tunnel monitoring, in particular to a real-time defect monitoring method and system for a hydraulic tunnel. BACKGROUND

[0002] Real-time defect monitoring of a hydraulic tunnel is a key link to ensure the safe operation of a water conservancy project. The traditional monitoring method mainly relies on manual inspection, which is low in efficiency, high in labor intensity, and difficult to cover all areas, resulting in missed detection and blind spots. In addition, manual inspection is affected by factors such as light, the inspector's perception, standardization, and professionalism, making it difficult to ensure the accuracy and reliability of the monitoring results.

[0003] With the development of technology, image processing-based monitoring technology has been applied in recent years. A camera is installed inside the tunnel to collect images, and then image processing algorithms are used to identify defects in the images. However, this technology has poor adaptability to the environment. The environment inside the hydraulic tunnel is complex and variable, such as uneven light and large humidity changes, which can severely affect image quality and reduce the accuracy of defect identification. Moreover, traditional image processing algorithms lack flexibility and cannot dynamically adjust the identification strategy according to the actual danger level of the tunnel. The same identification method is often used for defects of different severity, which cannot be treated differently, resulting in unsatisfactory results when identifying minor defects or severe defects. Therefore, the image processing algorithm used in existing tunnel monitoring technology only identifies defects in hydraulic tunnels based on collected tunnel image data, without considering the influence of internal environmental factors of the hydraulic tunnel, resulting in inaccurate identification of defects in the hydraulic tunnel and reducing the accuracy and effectiveness of maintenance strategies. SUMMARY

[0004] The present application provides a real-time defect monitoring method and system for a hydraulic tunnel, which can solve the problem that existing tunnel monitoring technology only identifies defects in hydraulic tunnels based on collected tunnel image data, without considering the influence of internal environmental factors of the hydraulic tunnel, resulting in inaccurate identification of defects in the hydraulic tunnel and reducing the accuracy and effectiveness of maintenance strategies.

[0005] To solve the above technical problems, an embodiment of the present application provides a real-time defect monitoring method for a hydraulic tunnel, comprising:

[0006] obtaining real-time environmental data and real-time images of a hydraulic tunnel to be detected; wherein the real-time environmental data includes real-time temperature data, real-time humidity data, and real-time vibration data;

[0007] The real-time temperature data, real-time humidity data and real-time vibration data of the water tunnel to be detected are input into the trained tunnel danger prediction model based on a fuzzy algorithm, so that the trained tunnel danger prediction model based on the fuzzy algorithm performs danger prediction according to the real-time temperature data, real-time humidity data and real-time vibration data of the water tunnel to be detected, and obtains a danger prediction value of the water tunnel to be detected; the tunnel danger prediction model based on the fuzzy algorithm comprises a fuzzification layer, a fuzzy reasoning layer and a defuzzification layer;

[0008] The danger prediction value and real-time image of the water tunnel to be detected are input into the trained tunnel image recognition model, so that the trained tunnel image recognition model performs defect recognition according to the danger prediction value and real-time image of the water tunnel to be detected, and obtains a defect recognition result of the water tunnel to be detected;

[0009] According to the danger prediction value and defect recognition result of the water tunnel to be detected, a maintenance strategy of the water tunnel to be detected is determined.

[0010] Further, the model training of the tunnel danger prediction model based on the fuzzy algorithm comprises:

[0011] The historical temperature data, historical humidity data and historical vibration data of the water tunnel at each historical moment with an actual danger value label are obtained;

[0012] The historical temperature membership degree, historical humidity membership degree and historical vibration membership degree of the water tunnel at each historical moment are calculated by the fuzzification layer in the tunnel danger prediction model based on the fuzzy algorithm to be trained;

[0013] The historical danger grade fuzzy membership degree of the water tunnel at each historical moment is calculated by the fuzzy reasoning layer in the tunnel danger prediction model based on the fuzzy algorithm to be trained according to the historical temperature membership degree, historical humidity membership degree and historical vibration membership degree of the water tunnel at each historical moment;

[0014] The historical danger prediction value of the water tunnel at each historical moment is calculated by the gravity center method of the defuzzification layer in the tunnel danger prediction model based on the fuzzy algorithm to be trained according to the historical danger grade fuzzy membership degree of the water tunnel at each historical moment;

[0015] The loss value is calculated by comparing the historical danger prediction value and actual danger value of the water tunnel at each historical moment, the parameters of the tunnel danger prediction model based on the fuzzy algorithm to be trained are optimized according to the loss value, and the trained tunnel danger prediction model based on the fuzzy algorithm is obtained until the loss value converges.

[0016] Further, the history temperature membership degree, the history humidity membership degree and the history vibration membership degree of the hydraulic tunnel at each time point are calculated by a fuzzification layer in the tunnel danger prediction model based on the fuzzy algorithm to be trained, comprising:

[0017] The history temperature membership degree of the hydraulic tunnel at each time point is calculated by a preset temperature fuzzy set and a temperature membership function of the fuzzification layer in the tunnel danger prediction model based on the fuzzy algorithm to be trained; wherein the preset temperature fuzzy set comprises a low temperature fuzzy set, a middle temperature fuzzy set and a high temperature fuzzy set;

[0018] The history humidity membership degree of the hydraulic tunnel at each time point is calculated by a preset humidity fuzzy set and a humidity membership function of the fuzzification layer in the tunnel danger prediction model based on the fuzzy algorithm to be trained; wherein the preset temperature fuzzy set comprises a low humidity fuzzy set, a middle humidity fuzzy set and a high humidity fuzzy set;

[0019] The history vibration membership degree of the hydraulic tunnel at each time point is calculated by a preset vibration fuzzy set and a vibration membership function of the fuzzification layer in the tunnel danger prediction model based on the fuzzy algorithm to be trained; wherein the preset temperature fuzzy set comprises a low vibration fuzzy set, a middle vibration fuzzy set and a high vibration fuzzy set.

[0020] Further, the calculation formulas of the temperature membership function, the humidity membership function and the vibration membership function are as follows:

[0021] The calculation formula of the temperature membership function is:

[0022] ;

[0023] Wherein, the represents the membership degree of the temperature data T in the temperature fuzzy set; the A represents the preset temperature fuzzy set, and the preset temperature fuzzy set comprises a low temperature fuzzy set, a middle temperature fuzzy set and a high temperature fuzzy set; the represents the preset temperature mean value of the preset temperature fuzzy set; the represents the preset temperature standard deviation of the preset temperature fuzzy set;

[0024] The calculation formula of the humidity membership function is:

[0025] ;

[0026] Wherein, the represents the membership degree of the humidity data H in the temperature fuzzy set; the B represents the preset humidity fuzzy set, and the preset humidity fuzzy set comprises a low humidity fuzzy set, a middle humidity fuzzy set and a high humidity fuzzy set; the a preset humidity mean value of the preset humidity fuzzy set; the preset humidity mean value of the preset humidity fuzzy set is calculated by the following formula: a preset humidity standard deviation of the preset humidity fuzzy set; the preset humidity standard deviation of the preset humidity fuzzy set is calculated by the following formula:

[0027] a calculation formula of the humidity membership function is as follows:

[0028]

[0029] wherein, the C represents a preset vibration fuzzy set, the preset vibration fuzzy set including a low vibration fuzzy set, a medium vibration fuzzy set and a high vibration fuzzy set; the a preset vibration mean value of the preset vibration fuzzy set; the preset vibration mean value of the preset vibration fuzzy set is calculated by the following formula: a preset vibration standard deviation of the preset vibration fuzzy set; the preset vibration standard deviation of the preset vibration fuzzy set is calculated by the following formula:

[0030] a calculation formula of the gravity method is as follows:

[0031]

[0032] wherein, the x represents a final risk prediction value of the water tunnel to be detected; the a and the b respectively represent prediction interval values of different risk levels; the a risk level fuzzy membership degree.

[0033] further, the trained tunnel image recognition model performs defect recognition according to the risk prediction value of the water tunnel to be detected and the real-time image, and obtains a defect recognition result of the water tunnel to be detected, including:

[0034] according to the risk prediction value of the water tunnel to be detected, determining the convolution layer type and the self-attention layer type of the discriminator in the trained tunnel image recognition model; wherein, the convolution layer type includes a first convolution layer, a second convolution layer and a third convolution layer; the self-attention layer type includes a first self-attention layer, a second self-attention layer and a third self-attention layer;

[0035] extracting features of the real-time image of the water tunnel to be detected through the convolution layer in the trained tunnel image recognition model, to obtain a plurality of tunnel feature maps;

[0036] enhancing features of the plurality of tunnel feature maps through the self-attention layer in the trained tunnel image recognition model, to obtain a plurality of enhanced tunnel feature maps;

[0037] performing linear transformation and nonlinear activation on the plurality of enhanced tunnel feature maps through the full connection layer in the trained tunnel image recognition model, to obtain the defect recognition result.

[0038] ​​​Further, the convolutional layer type and the self-attention layer type of the discriminator in the trained tunnel image recognition model are determined according to the risk prediction value of the water tunnel to be detected, and the determination comprises:

[0039] When the actual risk value is located in the low-risk preset interval, the convolutional layer type and the self-attention layer type of the discriminator in the trained tunnel image recognition model are determined as the first convolutional layer and the first self-attention layer respectively;

[0040] When the actual risk value is located in the medium-risk preset interval, the convolutional layer type and the self-attention layer type of the discriminator in the trained tunnel image recognition model are determined as the second convolutional layer and the second self-attention layer respectively;

[0041] When the actual risk value is located in the high-risk preset interval, the convolutional layer type and the self-attention layer type of the discriminator in the trained tunnel image recognition model are determined as the third convolutional layer and the third self-attention layer respectively.

[0042] Further, the model training of the tunnel image recognition model comprises:

[0043] Obtaining historical water tunnel images with actual risk value labels and defect actual labels;

[0044] According to the actual risk value of the historical water tunnel image, the convolutional layer type and the self-attention layer type of the tunnel image recognition model to be trained are determined;

[0045] The convolutional layer and the self-attention layer of the determined tunnel image recognition model to be trained are used to respectively extract features and enhance features of the historical water tunnel image, so as to obtain a feature map of the historical water tunnel image;

[0046] The feature map of the historical water tunnel image is linearly transformed and nonlinearly activated by the full connection layer of the tunnel image recognition model to be trained, so as to obtain a defect recognition result of the historical water tunnel image;

[0047] The defect recognition result of the historical water tunnel image is compared with the defect actual label to calculate a loss value, and the parameters of the tunnel image recognition model to be trained are optimized according to the loss value until the loss value converges, so as to obtain the trained tunnel image recognition model.

[0048] Further, the maintenance strategy of the water tunnel to be detected is determined according to the risk prediction value and the defect recognition result of the water tunnel to be detected, and the determination comprises:

[0049] The risk prediction value of the water tunnel to be detected is compared with a preset risk level threshold value, the defect recognition result of the water tunnel to be detected is classified according to the maintenance level, and the priority of the maintenance strategy of the water tunnel to be detected is determined according to the maintenance level classification.

[0050] Based on the above method embodiment, the application provides a system embodiment;

[0051] An embodiment of the application provides a real-time defect monitoring system for a hydraulic tunnel, comprising a data acquisition module, a danger prediction module, a defect identification module and a maintenance strategy generation module.

[0052] The data acquisition module is configured to acquire real-time environmental data and real-time images of the hydraulic tunnel to be detected, wherein the real-time environmental data comprises real-time temperature data, real-time humidity data and real-time vibration data.

[0053] The danger prediction module is configured to input the real-time temperature data, the real-time humidity data and the real-time vibration data of the hydraulic tunnel to be detected into a trained tunnel danger prediction model based on a fuzzy algorithm, so that the trained tunnel danger prediction model based on the fuzzy algorithm performs danger prediction according to the real-time temperature data, the real-time humidity data and the real-time vibration data of the hydraulic tunnel to be detected, and obtains a danger prediction value of the hydraulic tunnel to be detected.

[0054] The defect identification module is configured to input the danger prediction value and the real-time images of the hydraulic tunnel to be detected into a trained tunnel image identification model, so that the trained tunnel image identification model performs defect identification according to the danger prediction value and the real-time images of the hydraulic tunnel to be detected, and obtains a defect identification result of the hydraulic tunnel to be detected.

[0055] The maintenance strategy generation module is configured to determine a maintenance strategy of the hydraulic tunnel to be detected according to the defect identification result of the hydraulic tunnel to be detected.

[0056] Compared with the prior art, the embodiment of the present application has the following beneficial effects: the technical scheme of the present application utilizes the tunnel danger prediction model based on the fuzzy algorithm which is constructed and trained, performs danger value prediction calculation on the water conservancy tunnel to be detected according to real-time temperature data, real-time humidity data and real-time vibration data of the water conservancy tunnel to be detected, obtains a danger prediction value of the water conservancy tunnel to be detected, determines the image recognition channel of the tunnel image recognition model which is constructed and trained according to the danger prediction value of the water conservancy tunnel to be detected, that is, determines the convolution layer and the self-attention layer of the tunnel image recognition model, inputs the real-time image of the water conservancy tunnel to be detected into the corresponding image recognition channel, performs defect recognition by utilizing the convolution layer, the self-attention layer and the full connection layer in the tunnel image recognition model, obtains a defect recognition result of the water conservancy tunnel to be detected, and finally determines the maintenance strategy of the water conservancy tunnel to be detected according to the danger prediction value and the defect recognition result of the water conservancy tunnel to be detected. That is, the present application comprehensively calculates the danger value of the water conservancy tunnel to be detected according to real-time environmental factors of the water conservancy tunnel to be detected by utilizing the fuzzy algorithm, selects appropriate convolution layers and self-attention layers for the water conservancy tunnel images obtained in real time according to the calculated danger value of the water conservancy tunnel to be detected, so that the tunnel image recognition model can reasonably and accurately perform defect recognition on the real-time images of the water conservancy tunnel to be detected, and the defect recognition accuracy is improved. Moreover, the danger value calculated according to the environmental factors and the defect recognition result are utilized to determine the maintenance strategy of the water conservancy tunnel to be detected, the accuracy and effectiveness of the maintenance strategy are improved, and the problem that the existing tunnel monitoring technology only performs water conservancy tunnel defect recognition on the collected tunnel image data, lacks consideration of the influence of the internal environmental factors of the water conservancy tunnel, and thus cannot accurately recognize the defects of the water conservancy tunnel, thereby reducing the accuracy and effectiveness of the maintenance strategy, is solved. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 : a step flow chart of a water conservancy tunnel real-time defect monitoring method provided by the embodiment of the present application;

[0058] Figure 2 : a system structure diagram of a water conservancy tunnel real-time defect monitoring system provided by the embodiment of the present application. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0060] In the description of the present application, it should be understood that the terms "first", "second" and "third" are only for descriptive purposes and cannot be understood as indicating or implying relative importance or implying the number of indicated technical features.

[0061] Embodiment 1:

[0062] With reference to Figure 1 A step flow chart of a water tunnel real-time defect monitoring method provided by an embodiment of the present application, which comprises at least the following steps S1 to S4:

[0063] Step S1: obtaining real-time environmental data and real-time images of the water tunnel to be detected; wherein the real-time environmental data comprises real-time temperature data, real-time humidity data and real-time vibration data;

[0064] In this embodiment, the temperature sensor, humidity sensor, vibration sensor and camera equipment are installed at the detection position of the water tunnel to be detected at a certain interval, and a unified collection frequency is set, so that the temperature sensor, humidity sensor, vibration sensor and camera equipment can obtain the real-time environmental data and real-time images of the water tunnel to be detected at the same time, and the real-time environmental data and real-time images collected by the sensors are respectively subjected to data preprocessing and image preprocessing; wherein the data preprocessing includes but is not limited to data cleaning and data transformation; the data cleaning includes but is not limited to data deduplication, missing value processing, abnormal value processing and error value processing; the image preprocessing includes but is not limited to denoising, smoothing and edge optimization;

[0065] Step S2: inputting the real-time temperature data, real-time humidity data and real-time vibration data of the water tunnel to be detected into the trained tunnel danger prediction model based on fuzzy algorithm, so that the trained tunnel danger prediction model based on fuzzy algorithm can make danger prediction according to the real-time temperature data, real-time humidity data and real-time vibration data of the water tunnel to be detected, and obtain the danger prediction value of the water tunnel to be detected; the tunnel danger prediction model based on fuzzy algorithm comprises a fuzzification layer, a fuzzy reasoning layer and a defuzzification layer;

[0066] In this embodiment, the model training of the tunnel danger prediction model based on fuzzy algorithm comprises:

[0067] obtaining historical temperature data, historical humidity data and historical vibration data of the water tunnel with actual danger value label at each time point;

[0068] In the embodiment, the historical temperature data, the historical humidity data and the historical vibration data of the water conservancy tunnel labeled with the actual risk value at each historical moment are preprocessed; the data preprocessing includes but is not limited to data cleaning and data transformation; the data cleaning includes but is not limited to data deduplication, missing value processing, abnormal value processing and error value processing;

[0069] The historical temperature membership degree, the historical humidity membership degree and the historical vibration membership degree of the water conservancy tunnel at each historical moment are calculated by a fuzzy layer in the tunnel risk prediction model based on the fuzzy algorithm to be trained;

[0070] According to the historical temperature membership degree, the historical humidity membership degree and the historical vibration membership degree of the water conservancy tunnel at each historical moment, fuzzy reasoning is performed by a fuzzy reasoning layer in the tunnel risk prediction model based on the fuzzy algorithm to be trained, and the historical risk level fuzzy membership degree of the water conservancy tunnel at each historical moment is calculated;

[0071] In the embodiment, according to the historical temperature membership degree, the historical humidity membership degree and the historical vibration membership degree of the water conservancy tunnel at each historical moment, the historical risk level fuzzy membership degree of the water conservancy tunnel at each historical moment that meets the fuzzy preset risk level fuzzy rule library is calculated by using a risk level fuzzy membership degree algorithm according to the preset risk level fuzzy rule library of the fuzzy reasoning layer in the tunnel risk prediction model based on the fuzzy algorithm; the risk level fuzzy membership degree algorithm includes but is not limited to a maximum membership degree algorithm and a minimum membership degree algorithm;

[0072] In the embodiment, the risk level fuzzy rules and the number of the preset risk level fuzzy rule library can be constructed by a person skilled in the art according to actual conditions; the risk level fuzzy rules can be constructed by using AND and OR rules in fuzzy logic;

[0073] In the embodiment, examples of the risk level fuzzy rules include but are not limited to the following rules:

[0074] When the temperature data is in a high temperature condition, the humidity data is in a high humidity condition and the vibration data is in a high vibration condition, the risk level of the water conservancy tunnel is high;

[0075] When the temperature data is in a medium temperature condition, the humidity data is in a medium humidity condition and the vibration data is in a medium vibration condition, the risk level of the water conservancy tunnel is medium;

[0076] When the temperature data is in a low temperature condition, the humidity data is in a low humidity condition and the vibration data is in a low vibration condition, the risk level of the water conservancy tunnel is low;

[0077] According to the historical risk grade fuzzy membership degree of the water tunnel at each historical moment, the historical risk prediction value of the water tunnel at each historical moment is calculated through the barycentric method of the defuzzification layer in the tunnel risk prediction model to be trained based on the fuzzy algorithm.

[0078] In the embodiment, the calculation formula of the barycentric method is as follows:

[0079] ;

[0080] Wherein, the x represents the final risk prediction value of the water tunnel to be detected; the a and the b respectively represent the prediction interval values of different risk grades; the represents the risk grade fuzzy membership degree;

[0081] The loss value is calculated by comparing the historical risk prediction value of the water tunnel at each historical moment with the actual risk value, the parameters of the tunnel risk prediction model to be trained based on the fuzzy algorithm are optimized according to the loss value, until the loss value converges, and the trained tunnel risk prediction model based on the fuzzy algorithm is obtained.

[0082] In the embodiment, the historical temperature membership degree, the historical humidity membership degree and the historical vibration membership degree of the water tunnel at each historical moment are calculated through the fuzzification layer in the tunnel risk prediction model to be trained based on the fuzzy algorithm, which includes:

[0083] The historical temperature membership degree of the water tunnel at each historical moment is calculated through the preset temperature fuzzy set and the temperature membership function of the fuzzification layer in the tunnel risk prediction model to be trained based on the fuzzy algorithm; wherein, the preset temperature fuzzy set includes a low temperature fuzzy set, a medium temperature fuzzy set and a high temperature fuzzy set;

[0084] In the embodiment, when the detection temperature data is in the temperature value interval of the low temperature fuzzy set, the low temperature membership degree of the detection temperature data in the low temperature fuzzy set is calculated through the temperature membership function, when the detection temperature data is in the temperature value interval of the medium temperature fuzzy set, the medium temperature membership degree of the detection temperature data in the medium temperature fuzzy set is calculated through the temperature membership function, and when the detection temperature data is in the temperature value interval of the high temperature fuzzy set, the high temperature membership degree of the detection temperature data in the high temperature fuzzy set is calculated through the temperature membership function;

[0085] The historical humidity membership degree of the water tunnel at each historical moment is calculated through the preset humidity fuzzy set and the humidity membership function of the fuzzification layer in the tunnel risk prediction model to be trained based on the fuzzy algorithm; wherein, the preset temperature fuzzy set includes a low humidity fuzzy set, a medium humidity fuzzy set and a high humidity fuzzy set;

[0086] In the embodiment, when the detected humidity data is in the humidity value interval of the low humidity fuzzy set, the low humidity membership degree of the detected humidity data in the low humidity fuzzy set is calculated through the humidity membership function; when the detected humidity data is in the humidity value interval of the medium humidity fuzzy set, the medium humidity membership degree of the detected humidity data in the medium humidity fuzzy set is calculated through the humidity membership function; and when the detected humidity data is in the humidity value interval of the high humidity fuzzy set, the high humidity membership degree of the detected humidity data in the high humidity fuzzy set is calculated through the humidity membership function.

[0087] The historical vibration membership degrees of the water conservancy tunnel at each historical moment are calculated through the preset vibration fuzzy set and the vibration membership function of the fuzzification layer in the tunnel danger prediction model based on the fuzzy algorithm to be trained; wherein the preset temperature fuzzy set includes a low vibration fuzzy set, a medium vibration fuzzy set and a high vibration fuzzy set.

[0088] In the embodiment, when the detected vibration data is in the vibration value interval of the low vibration fuzzy set, the low vibration membership degree of the detected vibration data in the low vibration fuzzy set is calculated through the vibration membership function; when the detected vibration data is in the vibration value interval of the medium vibration fuzzy set, the medium vibration membership degree of the detected vibration data in the medium vibration fuzzy set is calculated through the vibration membership function; and when the detected vibration data is in the vibration value interval of the high vibration fuzzy set, the high vibration membership degree of the detected vibration data in the high vibration fuzzy set is calculated through the vibration membership function.

[0089] In the embodiment, the calculation formulas of the temperature membership function, the humidity membership function and the vibration membership function are as follows:

[0090] The calculation formula of the temperature membership function is:

[0091] ;

[0092] Wherein, the represents the membership degree of the temperature data T in the temperature fuzzy set; the A represents the preset temperature fuzzy set, and the preset temperature fuzzy set includes a low temperature fuzzy set, a medium temperature fuzzy set and a high temperature fuzzy set; the represents the preset temperature mean value of the preset temperature fuzzy set; the represents the preset temperature standard deviation of the preset temperature fuzzy set.

[0093] The calculation formula of the humidity membership function is:

[0094] ;

[0095] Wherein, the The degree of membership of humidity data H in the temperature fuzzy set is represented by ; B represents a preset humidity fuzzy set, which includes a low humidity fuzzy set, a medium humidity fuzzy set, and a high humidity fuzzy set; The preset humidity mean value represents the preset humidity fuzzy set; The standard deviation of the preset humidity set represents the preset humidity standard deviation.

[0096] The formula for calculating the humidity membership function is as follows:

[0097] ;

[0098] Among them, the The degree of membership of vibration data V in the temperature fuzzy set is represented by C; C represents a preset vibration fuzzy set, which includes a low vibration fuzzy set, a medium vibration fuzzy set, and a high vibration fuzzy set; The preset vibration mean value represents the preset vibration fuzzy set; the This represents the preset vibration standard deviation of the preset vibration fuzzy set.

[0099] Step S3: Input the hazard prediction value and real-time image of the hydraulic tunnel to be detected into the trained tunnel image recognition model, so that the trained tunnel image recognition model can perform defect identification based on the hazard prediction value and real-time image of the hydraulic tunnel to be detected, and obtain the defect identification result of the hydraulic tunnel to be detected.

[0100] In this embodiment, the trained tunnel image recognition model performs defect identification based on the predicted hazard value and real-time image of the hydraulic tunnel to be detected, and obtains the defect identification result of the hydraulic tunnel to be detected, including:

[0101] Based on the predicted hazard value of the hydraulic tunnel to be detected, the convolutional layer type and self-attention layer type of the discriminator in the trained tunnel image recognition model are determined; wherein, the convolutional layer type includes a first convolutional layer, a second convolutional layer, and a third convolutional layer; the self-attention layer type includes a first self-attention layer, a second self-attention layer, and a third self-attention layer;

[0102] In the embodiment, when the actual risk value is located in the low-risk preset interval, it is determined that the convolution layer type and the self-attention layer type of the discriminator in the trained tunnel image recognition model are respectively a first convolution layer and a first self-attention layer; when the actual risk value is located in the medium-risk preset interval, it is determined that the convolution layer type and the self-attention layer type of the discriminator in the trained tunnel image recognition model are respectively a second convolution layer and a second self-attention layer; when the actual risk value is located in the high-risk preset interval, it is determined that the convolution layer type and the self-attention layer type of the discriminator in the trained tunnel image recognition model are respectively a third convolution layer and a third self-attention layer; wherein the low-risk preset interval, the medium-risk preset interval and the high-risk preset interval can be determined by dividing according to the actual situation by those skilled in the art.

[0103] The real-time image of the water tunnel to be detected is subjected to feature extraction by the convolution layer in the trained tunnel image recognition model, and a plurality of tunnel feature maps are obtained.

[0104] In the embodiment, there are differences between different types of convolution layers; wherein the differences include but are not limited to the receptive field size, the number of convolution kernels, the size of convolution kernels and the convolution step length of the convolution layer.

[0105] The plurality of tunnel feature maps are subjected to feature enhancement by the self-attention layer in the trained tunnel image recognition model, and a plurality of enhanced tunnel feature maps are obtained.

[0106] In the embodiment, the feature weight values of different types of self-attention layers are different.

[0107] The plurality of enhanced tunnel feature maps are subjected to linear transformation and nonlinear activation by the fully connected layer in the trained tunnel image recognition model, and a defect recognition result is obtained.

[0108] In the embodiment, the plurality of enhanced tunnel feature maps are subjected to linear transformation and nonlinear activation by the weight matrix multiplication and the activation function of the fully connected layer in the trained tunnel image recognition model, and a defect recognition result is obtained; wherein the activation function includes but is not limited to ReLU activation function, Sigmoid activation function and Tanh activation function.

[0109] In the embodiment, the model training of the tunnel image recognition model comprises:

[0110] Obtain historical water tunnel images with actual risk value labels and defect actual labels;

[0111] Image preprocessing is performed on the historical water tunnel images with actual risk value labels and defect actual labels; wherein the image preprocessing includes but is not limited to denoising, smoothing and edge optimization;

[0112] According to the actual risk value of the historical water tunnel image, the convolution layer type and the self-attention layer type of the to-be-trained tunnel image recognition model are determined;

[0113] The convolution layer and the self-attention layer of the to-be-trained tunnel image recognition model are determined, and the historical water tunnel image is subjected to feature extraction and feature enhancement respectively, so as to obtain a feature map of the historical water tunnel image;

[0114] The feature map of the historical water tunnel image is subjected to linear transformation and nonlinear activation through the full connection layer of the to-be-trained tunnel image recognition model, so as to obtain a defect recognition result of the historical water tunnel image;

[0115] The defect recognition result of the historical water tunnel image is compared with an actual defect label to calculate a loss value, and the parameters of the to-be-trained tunnel image recognition model are optimized according to the loss value until the loss value converges, so as to obtain a trained tunnel image recognition model;

[0116] In the embodiment, the loss value between the defect recognition result of the historical water tunnel image and the actual defect label is calculated through a loss function, the parameters of the to-be-trained tunnel image recognition model are optimized according to the loss value until the loss value converges, so as to obtain a trained tunnel image recognition model; wherein the loss function includes but is not limited to a mean square error loss function, a mean absolute error loss function and a binary cross entropy loss function; the optimization processing includes but is not limited to a stochastic gradient descent optimization algorithm and an Adam optimizer.

[0117] Step S4: determining a maintenance strategy of the to-be-detected water tunnel according to the risk prediction value and the defect recognition result of the to-be-detected water tunnel.

[0118] In the embodiment, the risk prediction value of the to-be-detected water tunnel is compared with a preset risk level threshold, the defect recognition result of the to-be-detected water tunnel is classified according to a maintenance level, and the priority of the maintenance strategy of the to-be-detected water tunnel is determined according to the maintenance level classification.

[0119] Embodiment 2:

[0120] Reference Figure 2 A system structure diagram of a water tunnel real-time defect monitoring system provided by the embodiment of the application, the system at least includes a data acquisition module, a risk prediction module, a defect recognition module and a maintenance strategy generation module;

[0121] The data acquisition module is used to acquire real-time environment data and real-time images of the to-be-detected water tunnel; wherein the real-time environment data includes real-time temperature data, real-time humidity data and real-time vibration data;

[0122] The danger prediction module is configured to input real-time temperature data, real-time humidity data and real-time vibration data of the water tunnel to be detected into the trained tunnel danger prediction model based on a fuzzy algorithm, so that the trained tunnel danger prediction model based on the fuzzy algorithm predicts danger according to the real-time temperature data, the real-time humidity data and the real-time vibration data of the water tunnel to be detected, and obtains a danger prediction value of the water tunnel to be detected; the tunnel danger prediction model based on the fuzzy algorithm comprises a fuzzification layer, a fuzzy reasoning layer and a defuzzification layer.

[0123] The defect identification module is configured to input the danger prediction value and real-time images of the water tunnel to be detected into the trained tunnel image recognition model, so that the trained tunnel image recognition model identifies defects according to the danger prediction value and the real-time images of the water tunnel to be detected, and obtains a defect identification result of the water tunnel to be detected.

[0124] The maintenance strategy generation module is configured to determine a maintenance strategy of the water tunnel to be detected according to the danger prediction value and the defect identification result of the water tunnel to be detected.

[0125] The above specific embodiments further specifically describe the purposes, technical solutions and advantages of the present application, and it should be understood that the above specific embodiments are merely specific embodiments of the present application and are not used to limit the protection scope of the present application. It should be particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for real-time defect monitoring of a hydraulic tunnel, characterized by, The method comprises the steps of: acquiring real-time environmental data and real-time images of a water tunnel to be detected, wherein the real-time environmental data comprises real-time temperature data, real-time humidity data and real-time vibration data; inputting the real-time temperature data, real-time humidity data and real-time vibration data of the water tunnel to be detected into a trained tunnel danger prediction model based on a fuzzy algorithm, so that the trained tunnel danger prediction model based on the fuzzy algorithm performs danger prediction according to the real-time temperature data, real-time humidity data and real-time vibration data of the water tunnel to be detected, and obtains a danger prediction value of the water tunnel to be detected; the tunnel danger prediction model based on the fuzzy algorithm comprises a fuzzification layer, a fuzzy reasoning layer and a defuzzification layer; inputting the danger prediction value and the real-time images of the water tunnel to be detected into a trained tunnel image recognition model, so that the trained tunnel image recognition model performs defect recognition according to the danger prediction value and the real-time images of the water tunnel to be detected, and obtains a defect recognition result of the water tunnel to be detected; determining a maintenance strategy of the water tunnel to be detected according to the danger prediction value and the defect recognition result of the water tunnel to be detected.

2. A method for real-time defect monitoring of a hydraulic tunnel according to claim 1, characterized in that, The model training of the tunnel danger prediction model based on the fuzzy algorithm comprises the steps of: acquiring historical temperature data, historical humidity data and historical vibration data of water tunnels at historical moments with actual danger values; calculating historical temperature membership degrees, historical humidity membership degrees and historical vibration membership degrees of the water tunnels at the historical moments through the fuzzification layer of the tunnel danger prediction model based on the fuzzy algorithm to be trained; performing fuzzy reasoning through the fuzzy reasoning layer of the tunnel danger prediction model based on the fuzzy algorithm to be trained according to the historical temperature membership degrees, historical humidity membership degrees and historical vibration membership degrees of the water tunnels at the historical moments, and calculating historical danger grade fuzzy membership degrees of the water tunnels at the historical moments; calculating historical danger prediction values of the water tunnels at the historical moments through the gravity center method of the defuzzification layer of the tunnel danger prediction model based on the fuzzy algorithm to be trained according to the historical danger grade fuzzy membership degrees of the water tunnels at the historical moments; comparing and calculating loss values of the historical danger prediction values and the actual danger values of the water tunnels at the historical moments, optimizing parameters of the tunnel danger prediction model based on the fuzzy algorithm to be trained according to the loss values, until the loss values converge, and obtaining the trained tunnel danger prediction model based on the fuzzy algorithm.

3. A method of real-time defect monitoring of a hydraulic tunnel according to claim 2, characterized in that, The calculation of the historical temperature membership degrees, historical humidity membership degrees and historical vibration membership degrees of the water tunnels at the historical moments through the fuzzification layer of the tunnel danger prediction model based on the fuzzy algorithm to be trained comprises the steps of: calculating the historical temperature membership degrees of the water tunnels at the historical moments through a preset temperature fuzzy set and a temperature membership function of the fuzzification layer of the tunnel danger prediction model based on the fuzzy algorithm to be trained; wherein the preset temperature fuzzy set comprises a low temperature fuzzy set, a medium temperature fuzzy set and a high temperature fuzzy set. Calculate the historical humidity membership degree of the hydraulic tunnel at each time point in history through the preset humidity fuzzy set and humidity membership function of the fuzzification layer in the tunnel danger prediction model to be trained based on a fuzzy algorithm, wherein the preset humidity fuzzy set comprises a low humidity fuzzy set, a medium humidity fuzzy set and a high humidity fuzzy set; Calculate the historical vibration membership degree of the hydraulic tunnel at each time point in history through the preset vibration fuzzy set and vibration membership function of the fuzzification layer in the tunnel danger prediction model to be trained based on a fuzzy algorithm, wherein the preset vibration fuzzy set comprises a low vibration fuzzy set, a medium vibration fuzzy set and a high vibration fuzzy set.

4. A method of real-time defect monitoring of a hydraulic tunnel according to claim 3, characterized in that, The calculation formulas of the temperature membership function, the humidity membership function and the vibration membership function are as follows: The calculation formula of the temperature membership function is: ; Wherein, the represents the membership degree of temperature data T in the temperature fuzzy set; the A represents a preset temperature fuzzy set, which includes a low temperature fuzzy set, a medium temperature fuzzy set and a high temperature fuzzy set; the represents a preset temperature mean value of the preset temperature fuzzy set; the represents a preset temperature standard deviation of the preset temperature fuzzy set; The calculation formula of the humidity membership function is: ; Wherein, the represents the membership degree of humidity data H in the temperature fuzzy set; B represents a preset humidity fuzzy set, which includes a low humidity fuzzy set, a medium humidity fuzzy set and a high humidity fuzzy set; and represents the preset humidity mean value of the preset humidity fuzzy set; and represents the preset humidity standard deviation of the preset humidity fuzzy set. The calculation formula of the vibration membership function is: ; Wherein, the represents the membership degree of the vibration data V in the temperature fuzzy set; the C represents a preset vibration fuzzy set, the preset vibration fuzzy set including a low vibration fuzzy set, a medium vibration fuzzy set and a high vibration fuzzy set; the represents a preset vibration mean value of the preset vibration fuzzy set; the represents a preset vibration standard deviation of the preset vibration fuzzy set.

5. A method of real-time defect monitoring of a hydraulic tunnel according to claim 4, characterized in that, The calculation formula of the vibration membership function is: ； Wherein, the x represents the final dangerous prediction value of the water tunnel to be detected; the a and b respectively represent the prediction interval values of different dangerous levels; the represents the dangerous level fuzzy membership degree.

6. A method of real-time defect monitoring of a hydraulic tunnel according to claim 5, characterized in that, The calculation formula of the barycentric method is as follows: The trained tunnel image recognition model performs defect recognition according to the danger prediction value and the real-time image of the hydraulic tunnel to be detected, and obtains a defect recognition result of the hydraulic tunnel to be detected, comprising: According to the danger prediction value of the hydraulic tunnel to be detected, determine the convolution layer type and the self-attention layer type of the discriminator in the trained tunnel image recognition model; wherein the convolution layer type comprises a first convolution layer, a second convolution layer and a third convolution layer; the self-attention layer type comprises a first self-attention layer, a second self-attention layer and a third self-attention layer; Extract features of the real-time image of the hydraulic tunnel to be detected through the convolution layer in the trained tunnel image recognition model, and obtain a plurality of tunnel feature maps; Enhance the features of the plurality of tunnel feature maps through the self-attention layer in the trained tunnel image recognition model, and obtain a plurality of enhanced tunnel feature maps; 7. A method of real-time defect monitoring of a hydraulic tunnel according to claim 6, characterized in that, Perform linear transformation and nonlinear activation on the plurality of enhanced tunnel feature maps through the full connection layer in the trained tunnel image recognition model, and obtain the defect recognition result. According to the danger prediction value of the hydraulic tunnel to be detected, determine the convolution layer type and the self-attention layer type of the discriminator in the trained tunnel image recognition model, comprising: When the actual danger value is located in the low danger preset interval, the convolution layer type and the self-attention layer type of the discriminator in the trained tunnel image recognition model are determined to be the first convolution layer and the first self-attention layer, respectively; When the actual danger value is located in the medium danger preset interval, the convolution layer type and the self-attention layer type of the discriminator in the trained tunnel image recognition model are determined to be the second convolution layer and the second self-attention layer, respectively; 8. A method of real-time defect monitoring of a hydraulic tunnel according to claim 7, characterized in that, When the actual danger value is located in the high danger preset interval, the convolution layer type and the self-attention layer type of the discriminator in the trained tunnel image recognition model are determined to be the third convolution layer and the third self-attention layer, respectively. The model training of the tunnel image recognition model comprises: Obtain historical hydraulic tunnel images with actual danger value labels and defect actual labels; According to the actual danger value of the historical hydraulic tunnel image, determine the convolution layer type and the self-attention layer type of the tunnel image recognition model to be trained; The convolutional layer and the self-attention layer of the determined to-be-trained tunnel image recognition model are used to respectively perform feature extraction and feature enhancement on the historical water conservancy tunnel image, so as to obtain a feature map of the historical water conservancy tunnel image; The feature map of the historical water conservancy tunnel image is linearly transformed and nonlinearly activated by the full connection layer of the to-be-trained tunnel image recognition model, so as to obtain a defect recognition result of the historical water conservancy tunnel image; The defect recognition result of the historical water conservancy tunnel image is compared with an actual label of the defect to calculate a loss value, and parameters of the to-be-trained tunnel image recognition model are optimized according to the loss value until the loss value converges, so as to obtain a trained tunnel image recognition model.

9. A method of real-time defect monitoring of a hydraulic tunnel according to claim 8, characterized in that, The maintenance strategy of the to-be-detected water conservancy tunnel is determined according to the risk prediction value and the defect recognition result of the to-be-detected water conservancy tunnel, and the maintenance strategy of the to-be-detected water conservancy tunnel includes: The risk prediction value of the to-be-detected water conservancy tunnel is compared with a preset risk level threshold value, the defect recognition result of the to-be-detected water conservancy tunnel is classified according to a maintenance level, and the priority of the maintenance strategy of the to-be-detected water conservancy tunnel is determined according to the maintenance level classification.

10. A real-time defect monitoring system for a hydraulic tunnel, characterized in that, It includes: a data acquisition module, a risk prediction module, a defect recognition module and a maintenance strategy generation module; The data acquisition module is used to acquire real-time environmental data and real-time images of the to-be-detected water conservancy tunnel; wherein the real-time environmental data includes real-time temperature data, real-time humidity data and real-time vibration data; The risk prediction module is used to input the real-time temperature data, the real-time humidity data and the real-time vibration data of the to-be-detected water conservancy tunnel into the trained tunnel risk prediction model based on the fuzzy algorithm, so that the trained tunnel risk prediction model based on the fuzzy algorithm performs risk prediction according to the real-time temperature data, the real-time humidity data and the real-time vibration data of the to-be-detected water conservancy tunnel, and obtains a risk prediction value of the to-be-detected water conservancy tunnel; the tunnel risk prediction model based on the fuzzy algorithm includes a fuzzification layer, a fuzzy reasoning layer and a defuzzification layer; The defect recognition module is used to input the risk prediction value and the real-time image of the to-be-detected water conservancy tunnel into the trained tunnel image recognition model, so that the trained tunnel image recognition model performs defect recognition according to the risk prediction value and the real-time image of the to-be-detected water conservancy tunnel, and obtains a defect recognition result of the to-be-detected water conservancy tunnel; The maintenance strategy generation module is used to determine the maintenance strategy of the to-be-detected water conservancy tunnel according to the risk prediction value and the defect recognition result of the to-be-detected water conservancy tunnel.

Images