Pipeline inspection method

Abstract

The invention discloses a pipeline inspection method. The method comprises the following steps: forming an abnormal hidden danger image data set, and dividing into a training set and a test set; performing target detection regression training on the training set by using a single-stage target detection technology and an unsupervised anomaly monitoring technology, and then performing evaluation by using the test set to obtain a pipeline inspection model; and inputting a to-be-detected pipeline inspection image into the pipeline inspection model for detection, and obtaining an abnormal detection result. The method has the advantages that the abnormal phenomena around the oil and gas pipeline can be accurately recognized, the abnormal event recognition efficiency is greatly improved, and the false alarm situation existing in target recognition is reduced. The influence caused by environmental factors can be reduced, the labor intensity is reduced, the safety of line patrol operators is improved, and the cost is reduced. The risk factors can be quickly and accurately searched and identified, so that the positive effects of reducing cost, improving efficiency, reducing economic loss and avoiding casualties are achieved.

Description

Technical Field

[0001] The invention belongs to the technical field of oil field pipeline maintenance, and in particular relates to a pipeline inspection method. Background Art

[0002] Oil and gas pipelines have the characteristics of large geographical span, long construction span, frequent safety accidents, etc., and the total mileage of pipelines is showing a trend of continuous growth. Therefore, the need to strengthen the safety inspection and management of pipelines is very urgent. At present, conventional manual inspections can no longer meet the requirements of timely, fast and accurate inspections. More scientific technical means are needed to detect oil and gas pipelines. Therefore, the use of drones for pipeline inspections has gradually become an application trend in the oil and gas industry at home and abroad. The use of drone technology equipped with algorithm recognition modules to capture data and establish inspection algorithm models for oil and gas pipelines can provide remote diagnosis and optimization management methods for digital oilfield construction and production maintenance, which can effectively realize long-distance and refined inspections of oil and gas pipelines.

[0003] Chinese Patent Application No. 202310045569.9 discloses a "BIM+MR-based non-opening penetrating underground pipeline monitoring and early warning method". This method builds a three-dimensional visual BIM model based on the completion drawings of the underground pipeline and adapts the MR scene with Unity3D. Wearing MR equipment on site, the underground pipeline model of the current area is called up by identifying the identity of the target pipe well to view the pipeline information and perform non-opening penetrating underground pipeline monitoring and early warning. However, this method requires the use of advanced virtual reality technology and high-precision underground pipeline positioning equipment. The cost of these technologies and equipment is high, and professional skills and experience are also required for use and maintenance. Therefore, the implementation of this method requires high investment and maintenance costs. At the same time, it needs to rely on the integrity and accuracy of the building information model (BIM). If the relevant information in the BIM is incomplete or inaccurate, it may affect the monitoring and early warning effect of the pipeline. Although this method can achieve real-time monitoring and early warning of pipelines, for some special cases, such as sudden breakage or damage of pipelines, field inspection and repair are still required.

[0004] Chinese patent application No. 202211307168.8 discloses "an intelligent pipeline inspection method", which divides various types of instruments into end-point pipeline task instruments, ordinary task instruments, and start-point pipeline task instruments according to the task type; establishes a database of various types of instruments and associates each instrument with the inspection point it executes and the end task point of the pipeline inspection type instrument, and configures the pan-tilt angle of the instrument execution and the pan-tilt angle of the inspection point; during the inspection, the instrument type at the inspection point is determined, the instrument type is confirmed, and the pipeline inspection task of the inspection point is confirmed according to the instrument type; according to the continuous inspection results of each type of instrument on the pipeline, a warning judgment is made for the pipeline site; when the pipeline inspection task of the inspection point is completed, the pan-tilt angle is adjusted, and the pipeline inspection task is performed at the next inspection point until all inspection tasks are completed. However, since the amount of data generated by this method during the pipeline inspection process is large, effective processing and analysis are required, but the existing data processing and analysis technology still has limitations, so problems such as data misjudgment or missed inspection may occur. Summary of the invention

[0005] In order to solve the above problems, an object of the present invention is to provide a pipeline inspection method.

[0006] In order to achieve the above object, the pipeline inspection method provided by the present invention comprises the following steps performed in sequence:

[0007] 1) Define abnormal hidden dangers, and use the drone-mounted image acquisition device to vertically look down and process the original abnormal hidden danger images within the pipeline inspection range according to the type, number and actual scale range of abnormal hidden dangers. All processed abnormal hidden danger images constitute an abnormal hidden danger image dataset, which is divided into a training set and a test set;

[0008] 2) The training set is trained with single-stage target detection technology and unsupervised anomaly monitoring technology for target detection regression, and then the test set is used for evaluation to obtain the pipeline inspection model;

[0009] 3) Use a drone to inspect the pipeline inspection area, and use an airborne image acquisition device to vertically look down at the pipeline to obtain an inspection image of the pipeline to be inspected, and then input the above pipeline inspection model for detection to obtain abnormal detection results.

[0010] In step 1), the abnormal hidden dangers are defined, and according to the type, number and actual scale range of the abnormal hidden dangers, the original abnormal hidden danger images within the pipeline inspection range are taken vertically by using the image acquisition device on the drone and processed, and all the processed abnormal hidden danger images constitute an abnormal hidden danger image data set, and the specific steps of dividing it into a training set and a test set are as follows:

[0011] 1.1) Abnormal situations including pipeline leakage, suspicious persons and vehicles around pipelines, occupation, tree planting, and earthwork are defined as abnormal hidden dangers;

[0012] 1.2) According to the actual situation, the above abnormal hidden dangers are divided into multiple types including equipment failure, building structure problems, and safety hazards;

[0013] 1.3) Determine the number and shooting range of abnormal hidden dangers that need to be photographed based on the type of abnormal hidden dangers and the area within the pipeline inspection range;

[0014] 1.4) Select the appropriate drone type and airborne image acquisition devices including high-definition cameras and thermal imagers according to the shooting range and the image accuracy required;

[0015] 1.5) Develop a shooting plan, determine the shooting route and shooting time according to the number of abnormal hidden dangers and the actual scale range;

[0016] 1.6) Control the drone to fly along the shooting route, and use the onboard image acquisition device to take vertical bird's-eye shots to obtain the original abnormal hidden danger images, and then pre-process the original abnormal hidden danger images by cropping, correcting and marking them for subsequent use;

[0017] 1.7) The pre-processed abnormal hidden danger images are classified and labeled, and all the processed abnormal hidden danger images constitute an abnormal hidden danger image dataset, which is divided into a training set and a test set.

[0018] In step 2), the training set is trained for target detection regression using single-stage target detection technology and unsupervised anomaly monitoring technology, and then evaluated using the test set. The specific steps for obtaining the pipeline inspection model are as follows:

[0019] 2.1) Preprocess the abnormal hidden danger images in the training set and test set, including data cleaning and labeling, to ensure the accuracy and consistency of the data;

[0020] 2.2) The UNet model in the single-stage target detection technology is used to train the abnormal hidden danger images in the training set to obtain abnormal conditions including suspicious persons and vehicles around the pipeline. The model parameters are appropriately adjusted during the training process to obtain an abnormal situation recognition model;

[0021] 2.3) The PaDiM model in the unsupervised anomaly monitoring technology is used to train the abnormal hidden danger images in the training set to obtain abnormal points including pipeline leakage, occupation, tree planting, and earthwork. The model parameters are appropriately adjusted during the training process to prevent overfitting and underfitting problems, improve the performance and stability of the model, and obtain an abnormal point recognition model;

[0022] 2.4) The test set is used to evaluate the accuracy and performance of the abnormal situation recognition model and the abnormal point recognition model. The abnormal situation recognition model and the abnormal point recognition model with good evaluation constitute the pipeline inspection model.

[0023] In step 3), the pipeline inspection area is inspected by using a drone, and the pipeline is photographed vertically by using an airborne image acquisition device to obtain an inspection image of the pipeline to be inspected, and then the pipeline inspection model is input for inspection. The specific steps for obtaining the abnormal detection result are as follows:

[0024] 3.1) Set the pipeline inspection route according to the pipeline coordinate information, control the drone to fly along the pipeline inspection route for inspection, and use the airborne image acquisition device to take vertical bird's-eye view photos to obtain the pipeline inspection image to be inspected, and then perform pre-processing including cropping, correction and annotation;

[0025] 3.2) The preprocessed pipeline inspection image to be inspected is input into the pipeline inspection model for data analysis and global detection of abnormal hidden dangers. The pipeline inspection model is used to identify abnormal hidden danger targets. When the abnormal hidden danger targets reach the preset critical value, the geographic coordinates of the abnormal points are marked on the pipeline inspection image to be inspected to generate an alarm map as the abnormal detection result, which is used to guide the work of the inspection personnel.

[0026] The pipeline inspection method provided by the present invention has the following beneficial effects:

[0027] (1) This method can significantly improve the inspection speed and efficiency of key pipeline inspection areas. Compared with manual inspection, it can avoid the misjudgment of abnormal points caused by subjective factors or empiricism. It uses single-stage target detection technology and unsupervised anomaly monitoring technology to establish a more accurate and complete pipeline inspection model, which can realize the accurate identification of abnormal phenomena around oil and gas pipelines, greatly improve the efficiency of abnormal event identification, reduce the false alarm of target identification, and shorten the inspection time by 27% compared with conventional inspections, with an accuracy rate of 93%.

[0028] (2) The use of drones to inspect oilfield pipeline areas can reduce the impact of environmental factors, reduce labor intensity, improve the safety of patrol workers, and reduce costs.

[0029] (3) In a relatively complex pipeline environment, it can quickly and accurately find and identify dangerous factors, including people, vehicles and other targets, to ensure the safety of the pipeline environment, and effectively ensure the high quality, high efficiency and safety of the inspection. The identification of pipelines and surrounding abnormal areas is more accurate, and it can crack down on bad behaviors in oil fields, long trunk lines, rivers and other scenes, and warn of abnormal situations, thereby achieving the positive effects of reducing costs, increasing efficiency, reducing economic losses and avoiding casualties. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A flow chart of the pipeline inspection method provided by the present invention. DETAILED DESCRIPTION

[0031] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] like Figure 1 As shown, the pipeline inspection method provided by the present invention comprises the following steps performed in sequence:

[0033] 1) Define abnormal hidden dangers, and use the drone-mounted image acquisition device to vertically look down and process the original abnormal hidden danger images within the pipeline inspection range according to the type, number and actual scale range of abnormal hidden dangers. All processed abnormal hidden danger images constitute an abnormal hidden danger image dataset, which is divided into a training set and a test set;

[0034] The specific steps are as follows:

[0035] 1.1) Abnormal situations including pipeline leakage, suspicious persons and vehicles around pipelines, occupation, tree planting, and earthwork are defined as abnormal hidden dangers;

[0036] 1.2) According to the actual situation, the above abnormal hidden dangers are divided into multiple types including equipment failure, building structure problems, and safety hazards;

[0037] 1.3) Determine the number and shooting range of abnormal hidden dangers that need to be photographed based on the type of abnormal hidden dangers and the area within the pipeline inspection range;

[0038] 1.4) Select the appropriate drone type and airborne image acquisition devices including high-definition cameras and thermal imagers according to the shooting range and the image accuracy required;

[0039] 1.5) Develop a shooting plan, determine the shooting route and shooting time according to the number of abnormal hidden dangers and the actual scale range;

[0040] 1.6) Control the drone to fly along the shooting route, and use the onboard image acquisition device to take vertical bird's-eye shots to obtain the original abnormal hidden danger images, and then pre-process the original abnormal hidden danger images by cropping, correcting and marking them for subsequent use;

[0041] 1.7) The pre-processed abnormal hidden danger images are classified and labeled, and all the processed abnormal hidden danger images constitute an abnormal hidden danger image dataset, which is divided into a training set and a test set.

[0042] 2) The training set is trained with single-stage target detection technology and unsupervised anomaly monitoring technology for target detection regression, and then the test set is used for evaluation to obtain the pipeline inspection model;

[0043] The specific steps are as follows:

[0044] 2.1) Preprocess the abnormal hidden danger images in the training set and test set, including data cleaning and labeling, to ensure the accuracy and consistency of the data;

[0045] 2.2) The UNet model in the single-stage target detection technology is used to train the abnormal hidden danger images in the training set to obtain abnormal situations including suspicious persons and vehicles around the pipeline. The model parameters are appropriately adjusted during the training process to prevent overfitting and underfitting problems, improve the performance and stability of the model, and obtain an abnormal situation recognition model;

[0046] 2.3) The PaDiM model in the unsupervised anomaly monitoring technology is used to train the abnormal hidden danger images in the training set to obtain abnormal points including pipeline leakage, occupation, tree planting, and earthwork. The model parameters are appropriately adjusted during the training process to prevent overfitting and underfitting problems, improve the performance and stability of the model, and obtain an abnormal point recognition model;

[0047] 2.4) The test set is used to evaluate the accuracy and performance of the abnormal situation recognition model and the abnormal point recognition model. The abnormal situation recognition model and the abnormal point recognition model with good evaluation constitute the pipeline inspection model.

[0048] 3) Use a drone to inspect the pipeline inspection area, and use an airborne image acquisition device to vertically look down at the pipeline to obtain an inspection image of the pipeline to be inspected, and then input the above pipeline inspection model for detection to obtain abnormal detection results.

[0049] The specific steps are as follows:

[0050] 3.1) Set the pipeline inspection route according to the pipeline coordinate information, control the drone to fly along the pipeline inspection route for inspection, and use the airborne image acquisition device to take vertical bird's-eye view photos to obtain the pipeline inspection image to be inspected, and then perform pre-processing including cropping, correction and annotation;

[0051] 3.2) The preprocessed pipeline inspection image to be inspected is input into the pipeline inspection model for data analysis and global detection of abnormal hidden dangers. The pipeline inspection model is used to identify abnormal hidden danger targets. When the abnormal hidden danger targets reach the preset critical value, the geographic coordinates of the abnormal points are marked on the pipeline inspection image to be inspected to generate an alarm map as the abnormal detection result, which is used to guide the work of the inspection personnel.

Claims

1. A pipeline inspection method, characterized in that: The pipeline inspection method comprises the following steps performed in sequence: 1) Define abnormal hidden dangers, and use the drone-mounted image acquisition device to vertically look down and process the original abnormal hidden danger images within the pipeline inspection range according to the type, number and actual scale range of abnormal hidden dangers. All processed abnormal hidden danger images constitute an abnormal hidden danger image dataset, which is divided into a training set and a test set; 2) The training set is trained with single-stage target detection technology and unsupervised anomaly monitoring technology for target detection regression, and then the test set is used for evaluation to obtain the pipeline inspection model; 3) Use a drone to inspect the pipeline inspection area, and use an airborne image acquisition device to vertically look down at the pipeline to obtain an inspection image of the pipeline to be inspected, and then input the above pipeline inspection model for detection to obtain abnormal detection results.

2. The pipeline inspection method according to claim 1, characterized in that: In step 1), the abnormal hidden dangers are defined, and according to the type, number and actual scale range of the abnormal hidden dangers, the original abnormal hidden danger images within the pipeline inspection range are taken vertically by using the image acquisition device on the drone and processed, and all the processed abnormal hidden danger images constitute an abnormal hidden danger image data set, and the specific steps of dividing it into a training set and a test set are as follows: 1.1) Abnormal situations including pipeline leakage, suspicious persons and vehicles around pipelines, occupation, tree planting, and earthwork are defined as abnormal hidden dangers; 1.2) According to the actual situation, the above abnormal hidden dangers are divided into multiple types including equipment failure, building structure problems, and safety hazards; 1.3) Determine the number and shooting range of abnormal hidden dangers that need to be photographed based on the type of abnormal hidden dangers and the area within the pipeline inspection range; 1.4) Select the appropriate drone type and onboard image acquisition devices including high-definition cameras and thermal imagers according to the shooting range and the image accuracy required; 1.5) Develop a shooting plan, determine the shooting route and shooting time according to the number of abnormal hidden dangers and the actual scale range; 1.6) Control the drone to fly along the shooting route, and use the onboard image acquisition device to take vertical bird's-eye shots to obtain the original abnormal hidden danger images, and then pre-process the original abnormal hidden danger images by cropping, correcting and marking them for subsequent use; 1.7) The pre-processed abnormal hidden danger images are classified and labeled, and all the processed abnormal hidden danger images constitute an abnormal hidden danger image dataset, which is divided into a training set and a test set.

3. The pipeline inspection method according to claim 1, characterized in that: In step 2), the training set is trained for target detection regression using single-stage target detection technology and unsupervised anomaly monitoring technology, and then evaluated using the test set. The specific steps for obtaining the pipeline inspection model are as follows: 2.1) Perform preprocessing including data cleaning and labeling on abnormal hidden danger images in the training set and test set to ensure the accuracy and consistency of the data; 2.2) The UNet model in the single-stage target detection technology is used to train the abnormal hidden danger images in the training set to obtain abnormal conditions including suspicious persons and vehicles around the pipeline. The model parameters are appropriately adjusted during the training process to obtain an abnormal situation recognition model; 2.3) The PaDiM model in the unsupervised anomaly monitoring technology is used to train the abnormal hidden danger images in the training set to obtain abnormal points including pipeline leakage, occupation, tree planting, and earthwork. The model parameters are appropriately adjusted during the training process to prevent overfitting and underfitting problems, improve the performance and stability of the model, and obtain an abnormal point recognition model; 2.4) The test set is used to evaluate the accuracy and performance of the abnormal situation recognition model and the abnormal point recognition model. The abnormal situation recognition model and the abnormal point recognition model with good evaluation constitute the pipeline inspection model.

4. The pipeline inspection method according to claim 1, characterized in that: In step 3), the pipeline inspection area is inspected by using a drone, and the pipeline is photographed vertically by using an airborne image acquisition device to obtain an inspection image of the pipeline to be inspected, and then the pipeline inspection model is input for inspection. The specific steps for obtaining the abnormal detection result are as follows: 3.1) Set the pipeline inspection route according to the pipeline coordinate information, control the drone to fly along the pipeline inspection route for inspection, and use the airborne image acquisition device to take vertical bird's-eye view photos to obtain the pipeline inspection image to be inspected, and then perform pre-processing including cropping, correction and annotation; 3.2) The preprocessed pipeline inspection image to be inspected is input into the pipeline inspection model for data analysis and global detection of abnormal hidden dangers. The pipeline inspection model is used to identify abnormal hidden danger targets. When the abnormal hidden danger targets reach the preset critical value, the geographic coordinates of the abnormal points are marked on the pipeline inspection image to be inspected to generate an alarm map as the abnormal detection result, which is used to guide the work of the inspection personnel.

Images