A pipeline intelligent inspection system with multi-source signal monitoring

Abstract

The invention discloses a pipeline intelligent inspection system for multi-source signal monitoring, comprising: a four-rotor unmanned aerial vehicle, an optoelectronic pod, a signal acquisition device, a ground signal transmission device and an inspection management system platform; the inspection management system platform comprises a signal storage module, a signal processing module and a signal analysis module; the signal analysis module comprises an environmental anomaly monitoring module, a gas anomaly detection module and a path planning module; the environmental anomaly monitoring module is used to perform environmental anomaly monitoring on signals through an image recognition model based on a graph autoencoder; the gas anomaly detection module is used to perform gas anomaly detection on signals through a gas anomaly detection model based on a sparse autoencoder; the path planning module is used to perform optimal inspection path planning based on a genetic algorithm; an intelligent inspection mode integrating line inspection, signal acquisition and feedback, and anomaly feedback is adopted to improve inspection efficiency and enhance inspection management level.

Description

Technical Field

[0001] The present invention relates to the technical field of pipeline inspection, and in particular to an intelligent pipeline inspection system for multi-source signal monitoring. Background Art

[0002] Oil and gas pipeline inspection is an important safety guarantee measure for the production and operation of the oil system. The inspection work is completed on schedule according to the pre-established inspection plan. The inspection content includes investigating whether there are abnormal situations such as oil and gas leaks, wandering personnel, and oil theft in the pipeline. However, the current oil and gas pipeline inspection and inspection are mostly carried out manually, and there are the following problems: oil and gas pipelines are widely distributed and most of them are located in remote areas. Some sections are even located in mountainous areas or gullies. The communication is not smooth and the climate is changeable, which makes manual inspection time-consuming and laborious, and there are certain safety hazards. Pipeline inspection is a procedural process. Problems found during the inspection need to be fed back to the control system in a timely manner. However, the types of signal sources that need to be monitored at the same time are different. For example, oil and gas leaks require taking photos of the scene, and gas sensors are also needed to detect the type of leaked gas; wandering personnel and oil theft require taking photos and videos. The task of collecting multi-source signals puts higher requirements on the work efficiency of inspectors, and different inspectors have different judgment criteria for abnormal situations. The basic task of pipeline inspection also needs to exclude the influence of subjective factors. The inspection quality and efficiency of pipeline multi-source signal collection tasks in the existing technology need to be improved. Therefore, how to use new equipment and new technologies to build a pipeline intelligent inspection system for multi-source signal monitoring to improve the inspection quality and efficiency is an urgent problem that technical personnel in this field need to solve. Summary of the invention

[0003] The purpose of the present invention is to provide a pipeline intelligent inspection system for multi-source signal monitoring to solve the problem in the prior art that the inspection quality and efficiency of pipeline multi-source signal acquisition tasks need to be improved.

[0004] The present invention provides a pipeline intelligent inspection system for multi-source signal monitoring, comprising: a four-rotor unmanned aerial vehicle, an optoelectronic pod, a signal acquisition device, a ground signal transmission device and an inspection management system platform; the four-rotor unmanned aerial vehicle is provided with an optoelectronic pod at the bottom, the optoelectronic pod is connected to the signal acquisition device, the optoelectronic pod is communicatively connected to the ground signal transmission device, and the ground signal transmission device is communicatively connected to the inspection management system platform;

[0005] The optoelectronic pod is used to control the four-rotor drone to fly to a designated location according to the operation instructions sent by the inspection management system platform, and to perform automatic inspection according to the pre-stored pipeline route map, and can record the location coordinates in real time; the signal acquisition device includes a camera and a gas sensor, and the camera is used to collect inspection videos during the inspection process; the gas sensor is used to collect air gas components during the inspection process; the inspection management system platform includes a signal storage module, a signal processing module and a signal analysis module; the signal storage module is used to store the signals collected by the signal acquisition device; the signal processing module is used to process the stored signals; the signal analysis module includes an environmental anomaly monitoring module, a gas anomaly detection module and a path planning module; the environmental anomaly monitoring module is used to perform environmental anomaly monitoring on the signal through an image recognition model based on a graph autoencoder; the gas anomaly detection module is used to perform gas anomaly detection on the signal through a gas anomaly detection model based on a sparse autoencoder; the path planning module is used to perform optimal inspection path planning based on a genetic algorithm.

[0006] Furthermore, the environmental anomaly monitoring module is used to use normal state pictures to train the image autoencoder, so that the image recognition model learns the normal state characteristics. When an abnormality occurs, the image recognition model can quickly identify it; the input of the offline training of the image recognition model is all normal state pictures, and the input of the online detection is a randomly acquired picture. When an abnormal situation is detected, an alarm is triggered.

[0007] Furthermore, the image recognition model based on the graph autoencoder includes an encoder and a decoder, the encoder uses a graph convolutional neural network GCN, and the decoder uses an inner product;

[0008] In GCN, X is the input image and A is the connection matrix. The latent feature expression of the input image is Z=GCN (X, A). GCN is a function that takes X and A as input and Z as output. ;in , D is the degree matrix, and is the parameter to be trained, RELU is the activation function; in the inner product, , is the reconstructed connection matrix;

[0009] The image recognition model based on the graph autoencoder adopts the cross entropy loss function. When applied online, when the input is a normal state picture, the loss function value is small, otherwise it is large. The loss function threshold can be selected according to the actual requirements of the task. The cross entropy loss function is:

[0010] ;

[0011] Where N is the number of pixels in the input image, y represents the value of an element in the connection matrix A, represent The value of an element in .

[0012] Furthermore, the gas anomaly detection module is used to use the feature learning and reconstruction capabilities of the autoencoder to learn the gas features under normal conditions, so that the gas anomaly detection model based on the sparse autoencoder can quickly identify the abnormal state; the sparse autoencoder is a symmetrical fully connected neural network, and the loss function is:

[0013] ; Among them, function f represents the encoder, function g represents the decoder, W and Z are the parameters of the encoder and decoder respectively; the network is trained offline using normal state data, and the back propagation method is used for parameter optimization; the data for online application is the randomly sampled gas content signal.

[0014] Furthermore, the path planning module is used to perform route planning during the local inspection process in the following manner:

[0015] The drone starts from the end point 0, selects some abnormal points for secondary inspection, and returns to the starting point N+1 after completing the inspection task. The total duration of the drone from takeoff to landing cannot exceed the maximum flight time of the drone. ; 0 and N+1 are used to represent the end point and starting point of the drone respectively. The set T ={1, 2, …,i, …, N} is the set of all positions to be inspected. The set consisting of the end point, starting point and all positions to be inspected of the drone is A = {0,1,…,i ,…,N,N+1}; the weight of the position i to be inspected is (i∈T) indicates that the weight represents the abnormality of position i. The larger the weight, the greater the abnormality, and the higher the priority of the position to be inspected. It is necessary to arrange a drone for a second inspection as soon as possible.

[0016] Furthermore, the path planning module is used to maximize the sum of the weights of the UAV's secondary inspection positions as the objective function of the optimization problem, and establish the following mathematical model, which means that in the secondary inspection process, as many as possible:

[0017] ;

[0018] The constraints are:

[0019] ;

[0020] ;

[0021] ;

[0022] ;

[0023] ;

[0024] Formula (1) indicates that in the secondary inspection, the drone must start from the end of the line and eventually return to the starting point; Formula (2) indicates that each location can only be inspected once at most; Formula (3) is the flow conservation constraint, the in-degree and out-degree of each node are equal; Formula (4) is the drone endurance constraint, represents the power consumption of the drone during the overall inspection process; Formula (5) is the value of the binary decision variable. When the drone inspects position i and flies to position j for inspection, ,otherwise ; After establishing the objective function and constraints, the genetic algorithm is used to optimize the solution.

[0025] Furthermore, the camera includes a visible light camera, an infrared camera and a front camera.

[0026] Furthermore, the visible light camera is used for daytime operation to take video of the patrol route; the infrared camera is used for nighttime operation to detect the temperature of the pipeline and confirm the pipeline condition; the positive camera is mounted on the quadcopter to scan and photograph the pipeline.

[0027] Furthermore, the ground signal transmission equipment includes a ground tower, a wireless private network and a fiber optic network, and the ground tower, the wireless private network and the fiber optic network are used simultaneously to complete the real-time collection of the drone inspection signals and the real-time feedback of the inspection system feedback signals.

[0028] Furthermore, the inspection management system platform is connected to a video image viewing terminal and a signal statistical analysis visualization terminal for real-time viewing of the drone inspection process and detection analysis results, and for visual display.

[0029] The beneficial effects of the present invention are as follows: the present invention provides a multi-source signal monitoring pipeline intelligent inspection system, which has an integrated unmanned inspection process, forming a complete inspection closed loop of data collection, transmission, analysis, and command feedback; the system is equipped with a multi-source signal monitoring and intelligent analysis module, which can perform real-time calculation and analysis on the data transmitted to the management platform; the inspection efficiency is high, cost-saving, and is not affected by environmental factors such as terrain and weather. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solution of the present invention, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0031] Figure 1 It is a schematic diagram of a pipeline intelligent inspection system for multi-source signal monitoring of the present invention;

[0032] Figure 2 It is a signal transmission flow chart of a pipeline intelligent inspection system with multi-source signal monitoring according to the present invention;

[0033] Figure 3 It is a structural diagram of a signal analysis module of a pipeline intelligent inspection system for multi-source signal monitoring of the present invention;

[0034] Figure 4 It is a structural diagram of an image recognition model based on a graph autoencoder of a pipeline intelligent inspection system for multi-source signal monitoring of the present invention;

[0035] Figure 5 It is a structural diagram of a gas anomaly detection model based on a sparse autoencoder in a pipeline intelligent inspection system for multi-source signal monitoring of the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. The technical solutions provided by the embodiments of the present invention are described in detail below in conjunction with the drawings.

[0037] See also Figure 1 The present invention provides a pipeline intelligent inspection system with multi-source signal monitoring, including: a quad-rotor drone 1, an optoelectronic pod 2, a signal acquisition device, a ground signal transmission device 5, and an inspection management system platform 6. The optoelectronic pod 2 is arranged at the bottom of the quad-rotor drone 1, and the lower part of the quad-rotor drone axis is connected to the upper part of the optoelectronic pod; the optoelectronic pod 2 is connected to the signal acquisition device, the optoelectronic pod 2 is connected to the ground signal transmission device 5, and the ground signal transmission device 5 is connected to the inspection management system platform 6.

[0038] The photoelectric pod 2 is used to control the four-rotor drone 1 to fly to the designated location according to the operation instructions sent by the inspection management system platform 6, and to perform automatic inspection according to the pre-stored pipeline route map, and can record the position coordinates in real time. The signal acquisition device includes a camera 3 and a gas sensor 4. The camera 3 is used to collect inspection videos during the inspection process; the gas sensor 4 is used to collect air gas components during the inspection process. The photoelectric pod base can be equipped with a gas sensor and a camera at the bottom; the gas sensor can convert the gas volume fraction into a corresponding electrical signal.

[0039] The inspection management system platform 6 includes a signal storage module, a signal processing module and a signal analysis module; the signal storage module is used to store the signals collected by the signal acquisition device; the signal processing module is used to process the stored signals; Figure 3 As shown, the signal analysis module includes an environmental anomaly monitoring module, a gas anomaly detection module and a path planning module; the inspection management system platform 6 can enable the drone to fly on the optimal inspection path and return home along the optimal path. The coordinates of the abnormal conditions found during the flight can be used as constraints when returning, and secondary sampling and confirmation can be performed at the designated location.

[0040] See also Figure 2 The inspection management system platform sends operation instructions to the quadcopter UAV and conducts comprehensive scheduling. The quadcopter is equipped with an optoelectronic pod and flies to a designated location to obtain UAV system data and UAV inspection data, including air gas composition data and UAV inspection video, which are sent back to the inspection management system platform through ground signal transmission equipment; the inspection management system platform processes the received UAV system data and UAV inspection data, determines the ground situation, and then makes corresponding operation instructions and abnormal alarms.

[0041] Specifically, the environmental anomaly monitoring module is used to monitor environmental anomalies of signals through an image recognition model based on a graph autoencoder. The autoencoder has good feature learning and reconstruction capabilities, and in the task of environmental anomaly detection, normal state pictures are often easier to obtain than abnormal state pictures. The environmental anomaly monitoring module is used to train the graph autoencoder with normal state pictures so that the image recognition model can learn the normal state features. When an anomaly occurs, the image recognition model can quickly identify it. The input of the image recognition model for offline training is all normal state pictures, and the input of the online detection is randomly acquired pictures. When an abnormal situation is detected, an alarm is triggered. Please refer to Figure 4 The image recognition model based on graph autoencoder includes an encoder and a decoder, the encoder uses a graph convolutional neural network GCN, and the decoder uses an inner-product;

[0042] In GCN, X is the input image and A is the connection matrix. The latent feature expression of the input image is Z=GCN (X, A). GCN is a function that takes X and A as input and Z as output. ;in, , D is the degree matrix, and is the parameter to be trained, RELU is the activation function; in the inner product, , is the reconstructed connection matrix;

[0043] The image recognition model based on the graph autoencoder uses the cross entropy loss function. When applied online, when the input is a normal state picture, the loss function value is small, otherwise it is large. The loss function threshold can be selected according to the actual requirements of the task and is adjustable. The cross entropy loss function is:

[0044] ;

[0045] Where N is the number of pixels in the input image, y represents the value of an element in the connection matrix A, represent The value of an element in .

[0046] The gas anomaly detection module is used to detect gas anomalies on signals through a gas anomaly detection model based on a sparse autoencoder; see Figure 5 The gas anomaly detection module is used to use the feature learning and reconstruction capabilities of the autoencoder to learn the gas features under normal conditions, so that the gas anomaly detection model based on the sparse autoencoder can quickly identify abnormal conditions; the use of the sparse autoencoder can highlight important features while weakening unnecessary features, thereby reducing feature redundancy. The sparse autoencoder is a symmetrical fully connected neural network, and the loss function is:

[0047] ;

[0048] Among them, function f represents the encoder, function g represents the decoder, W and Z are the parameters of the encoder and decoder respectively; the network is trained offline using normal state data, and the back propagation method is used for parameter optimization; the data for online application is the randomly sampled gas content signal.

[0049] The path planning module is used to plan the optimal inspection path based on the genetic algorithm. The line inspection of this system is divided into overall inspection and local inspection. The overall inspection traverses the entire line according to the established route, and performs real-time photography and gas collection; the local inspection is performed on the return flight, that is, the abnormal points found during the overall inspection are resampled. This optimal path planning algorithm solves the route planning during the local inspection process. The path planning module is used to plan the route during the local inspection process in the following way:

[0050] The drone starts from the end point 0, selects some abnormal points for secondary inspection, and returns to the starting point N+1 after completing the inspection task. The total duration of the drone from takeoff to landing cannot exceed the maximum flight time of the drone. ; 0 and N+1 are used to represent the end point and starting point of the drone respectively. The set T ={1, 2, …,i, …, N} is the set of all positions to be inspected. The set consisting of the end point, starting point and all positions to be inspected of the drone is A = {0,1,…,i ,…,N,N+1}; the weight of the position i to be inspected is (i∈T) indicates that the weight represents the abnormality of position i. The larger the weight, the greater the abnormality, and the higher the priority of the position to be inspected. It is necessary to arrange a drone for a second inspection as soon as possible.

[0051] For the problem of optimizing the path of the secondary inspection of UAV, the path planning module is used to maximize the sum of the weights of the secondary inspection positions of the UAV as the objective function of the optimization problem, and establish the following mathematical model, which means that in the process of the secondary inspection, as many as possible:

[0052] ;

[0053] The constraints are:

[0054] ;

[0055] ;

[0056] ;

[0057] ;

[0058] ;

[0059] Formula (1) indicates that in the secondary inspection, the drone must start from the end of the line and eventually return to the starting point; Formula (2) indicates that each location can only be inspected once at most; Formula (3) is the flow conservation constraint, the in-degree and out-degree of each node are equal; Formula (4) is the drone endurance constraint, represents the power consumption of the drone during the overall inspection process; Formula (5) is the value of the binary decision variable. When the drone inspects position i and flies to position j for inspection, ,otherwise ; After establishing the objective function and constraints, the genetic algorithm is used to optimize the solution.

[0060] Specifically, the camera 3 includes a visible light camera, an infrared camera and a positive camera. The visible light camera is used for daytime operation to shoot videos of the patrol route; the infrared camera is used for nighttime operation to detect the temperature of the pipeline and confirm the pipeline condition; the positive camera is carried on the quad-rotor drone to scan and shoot the pipeline.

[0061] The ground signal transmission equipment 5 includes a ground tower, a wireless private network and a fiber optic network. At the same time, the ground tower, the wireless private network and the fiber optic network are used to complete the real-time collection of the UAV inspection acquisition signal and the real-time feedback of the inspection system feedback signal. It can realize the real-time access of wired data, wireless private network data, and 4G / 5G data, and realize the real-time feedback of 1080p / 30fps high-definition video images, gas information feedback, image information feedback, location information feedback, and control information feedback.

[0062] The inspection management system platform 6 is connected to the video image viewing terminal and the signal statistical analysis visualization terminal, and is used to view the UAV inspection process and detection analysis results in real time and to perform visual display.

[0063] The above-described embodiments of the present invention do not limit the protection scope of the present invention.

Claims

1. A pipeline intelligent inspection system with multi-source signal monitoring, characterized in that: include: A four-rotor unmanned aerial vehicle (1), an optoelectronic pod (2), a signal collection device, a ground signal transmission device (5), and an inspection management system platform (6); the four-rotor unmanned aerial vehicle (1) is provided with an optoelectronic pod (2) at the bottom, the optoelectronic pod (2) is connected to the signal collection device, the optoelectronic pod (2) is communicatively connected to the ground signal transmission device (5), and the ground signal transmission device (5) is communicatively connected to the inspection management system platform (6); The photoelectric pod (2) is used to control the four-rotor drone (1) to fly to a designated location according to the received operation instructions sent by the inspection management system platform (6), and to perform automatic inspection according to the pre-stored pipeline route map, and can record the location coordinates in real time; The signal acquisition device comprises a camera (3) and a gas sensor (4), wherein the camera (3) is used to collect inspection videos during the inspection process; The gas sensor (4) is used to collect air gas components during the inspection process; The inspection management system platform (6) comprises a signal storage module, a signal processing module and a signal analysis module; the signal storage module is used to store the signals collected by the signal collection device; the signal processing module is used to process the stored signals; the signal analysis module comprises an environmental anomaly monitoring module, a gas anomaly detection module and a path planning module; The environmental anomaly monitoring module is used to monitor the environmental anomaly of the signal through an image recognition model based on a graph autoencoder; the gas anomaly detection module is used to detect gas anomalies on the signal through a gas anomaly detection model based on a sparse autoencoder; the path planning module is used to plan the optimal inspection path based on a genetic algorithm; The gas anomaly detection module is used to learn the gas characteristics under normal conditions by using the feature learning and reconstruction capabilities of the autoencoder, so that the gas anomaly detection model based on the sparse autoencoder can quickly identify the abnormal state; the sparse autoencoder is a symmetrical fully connected neural network, and the loss function is: Among them, function f represents the encoder, function g represents the decoder, W and Z are the parameters of the encoder and decoder respectively; the network is trained offline using normal state data, and the back propagation method is used for parameter optimization; the data for online application is the randomly sampled gas content signal.

2. According to the multi-source signal monitoring pipeline intelligent inspection system of claim 1, it is characterized in that: The environmental anomaly monitoring module is used to train the image autoencoder using normal state pictures, so that the image recognition model can learn the normal state features, and when an anomaly occurs, the image recognition model can quickly identify it; The inputs of the offline training of the image recognition model are all normal state pictures, and the inputs of the online detection are randomly acquired pictures. When an abnormal situation is detected, an alarm is triggered.

3. According to the multi-source signal monitoring pipeline intelligent inspection system of claim 2, it is characterized in that: The image recognition model based on graph autoencoder includes an encoder and a decoder, the encoder uses a graph convolutional neural network GCN, and the decoder uses an inner product; In GCN, X is the input image and A is the connection matrix. The latent feature expression of the input image is Z = GCN (X, A). GCN is a function with X and A as input and Z as output. in, D is the degree matrix, W0 and W1 are the parameters to be trained, and RELU is the activation function; in the inner product, is the reconstructed connection matrix; The image recognition model based on the graph autoencoder adopts the cross entropy loss function. When applied online, when the input is a normal state picture, the loss function value is small, otherwise it is large. The loss function threshold can be selected according to the actual requirements of the task. The cross entropy loss function is: Where N is the number of pixels in the input image, y represents the value of an element in the connection matrix A, represent The value of an element in .

4. The pipeline intelligent inspection system with multi-source signal monitoring according to claim 3 is characterized in that: The path planning module is used to plan the route during the local inspection in the following manner: The drone starts from the end point 0, selects some abnormal points for secondary inspection, and returns to the starting point N+1 after completing the inspection task. The total duration of the drone from takeoff to landing cannot exceed the maximum endurance time T of the drone. max ; 0 and N+1 are used to represent the end point and starting point of the drone respectively. The set T = {1,2,…,i,…,N} is the set of all positions to be inspected. The set consisting of the end point, starting point and all positions to be inspected of the drone is A = {0,1,…,i,…,N,N+1}; the weight of the position i to be inspected is w i (i∈T) indicates that the weight represents the abnormality degree of position i. The larger the weight, the greater the abnormality, and the higher the priority of the inspection of the position. It is necessary to arrange a drone for a second inspection as soon as possible.

5. The pipeline intelligent inspection system with multi-source signal monitoring according to claim 4 is characterized in that: The path planning module is used to maximize the sum of the weights of the UAV's secondary inspection positions as the objective function of the optimization problem and establish the following mathematical model. The meaning is to maximize the number of positions as much as possible during the secondary inspection process: The constraints are: (1) (2) (3) (4)t p ≤T max -T1; (5) Formula (1) indicates that in the secondary inspection, the drone must start from the end of the line and eventually return to the starting point; Formula (2) indicates that each location can only be inspected once at most; Formula (3) is the flow conservation constraint, and the in-degree and out-degree of each node are equal; Formula (4) is the drone endurance constraint, and T1 represents the power consumption of the drone during the overall inspection process; Formula (5) is the value of the binary decision variable. When the drone inspects location i and flies to location j for inspection, then x ij =1, otherwise x ij =0; After establishing the objective function and constraints, a genetic algorithm is used to optimize the solution.

6. The pipeline intelligent inspection system with multi-source signal monitoring according to claim 1 is characterized in that: The camera (3) comprises a visible light camera, an infrared camera and a positive camera.

7. The pipeline intelligent inspection system with multi-source signal monitoring according to claim 6 is characterized in that: The visible light camera is used for daytime operation to take video of the inspection route; the infrared camera is used for nighttime operation to detect the temperature of the pipeline and confirm the pipeline condition; the positive camera is carried on the quad-rotor drone to scan and photograph the pipeline.

8. The pipeline intelligent inspection system with multi-source signal monitoring according to claim 1 is characterized in that: The ground signal transmission equipment (5) includes a ground tower, a wireless private network and an optical fiber network, and simultaneously utilizes the ground tower, the wireless private network and the optical fiber network to complete the real-time collection of the drone inspection collection signals and the real-time return of the inspection system feedback signals.

9. The pipeline intelligent inspection system with multi-source signal monitoring according to claim 1 is characterized in that: The inspection management system platform (6) is connected to a video image viewing terminal and a signal statistical analysis visualization terminal, and is used to view the inspection process and detection analysis results of the drone in real time and to perform visual display.

Images