Intelligent state monitoring method for low-power-consumption logging cable winch

Abstract

The invention discloses an intelligent state monitoring method for a low-power-consumption logging cable winch, and the method comprises the steps: carrying out the overall deployment of a logging cable winch state monitoring network WM-YOLO which is completely pre-trained in edge computing equipment, and providing good hardware support for video monitoring through the low-power-consumption and high-computing-power characteristics of the equipment; the method comprises the following steps of: simultaneously acquiring video data of four paths of cameras placed above a wellhead and in front of a cable main winch, simultaneously performing frame extraction on the four paths of video stream data and combining the four paths of video stream data into an image vector consisting of four paths of video stream single-frame image data, and further, sending the image vector comprising state information of the wellhead and the cable winch into a neural network; the edge device executes a forward reasoning process of the WM-YOLO network, performs target detection on an input image vector to obtain a detection result, and finally calculates and analyzes an output result; normal results are displayed in real time, and abnormal results are displayed in real time while warning information is sent out in time.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent logging winch video monitoring. More specifically, it relates to an automated low-power intelligent state monitoring method for logging cable winches and an edge-side implementation method. Background Art

[0002] Cable logging is a key technology in modern oilfield development. During its actual application process, a logging tool is mainly transported downhole through a cable with a length of several kilometers, so as to scientifically and comprehensively obtain the formation data of the oil well. In the specific cable logging operation, depth is the most fundamental and core parameter during the logging operation. Therefore, depth monitoring is the most core work content, and depth accuracy is the basic quality control index to measure the success of the logging operation. Therefore, ensuring the accuracy of the logging depth is the top priority during the cable logging process.

[0003] During the actual work process, a hydraulic winch is used as the power source to drive the entire cable logging system. During open-hole logging, due to human fatigue, inattentiveness, too fast upper body speed of the hydraulic winch, excessive impact force, abrasion and failure of the outer steel wire of the logging cable, etc., serious problems such as the logging tool hitting up and down, cable breakage, and the logging tool falling into the wellbore occur, which affect the subsequent cable and tool fishing work and bring unpredictable adverse consequences. Therefore, there is an urgent need for an intelligent state monitoring system for logging cable winches to realize intelligent and automated intelligent state monitoring of logging cable winches, timely and accurately send warning information for abnormal states of cable winches, and assist on-site workers to take quick treatment measures, which is of great significance for on-site safe logging operations.

[0004] In recent years, with the continuous progress of science and technology, logging technology is developing towards the direction of "intelligentization" and "informationization". Computer vision technology has developed rapidly in recent years and has been widely applied in the logging field. However, in the intelligent state monitoring of logging cable winches at the ground end, there is no mature technology or product available. There is an urgent need for an automated and intelligent logging cable winch state intelligent monitoring system or device based on video or image to realize comprehensive monitoring of all aspects of the cable car state such as logging cable depth estimation and warning of excessive lifting impact of the wellhead instrument, so as to reduce labor costs, improve the risk prevention and control ability, and ensure operation safety. Summary of the Invention

[0005] The object of the present invention is to overcome the deficiencies of the prior art and provide an intelligent monitoring method for the state of a low-power logging cable winch. By deploying a computer vision detection algorithm on an edge computing device, based on the low-power and high-computing-power working conditions of the edge computing device, the abnormal stress state of the logging cable above the wellhead and the abnormal state information on the winch side are automatically detected in real time, and at the same time, the preliminary estimation function of the logging depth is realized, so as to realize the warning and monitoring functions of the abnormal state information of the logging cable winch and the wellhead.

[0006] To achieve the above object of the invention, an intelligent state monitoring method for a low-power logging cable winch of the present invention is characterized by including the following steps:

[0007] (1) Integrally deploy the trained and complete logging cable winch state monitoring network WM-YOLO on a low-power edge computing device; automatically initialize the network parameters when the system is powered on;

[0008] (2) Obtain the state monitoring image data of the wellhead and the winch;

[0009] (3) The edge computing device executes the forward inference process of the WM-YOLO network to obtain the target detection result in the single-frame image vector;

[0010] (4) The edge computing device outputs the real-time state information of the logging cable winch according to the real-time detection target result D f ;

[0011] (5) The edge computing device displays the real-time state information of the logging cable winch. For abnormal state information, it issues a warning message in real time to prompt and notify the operator to take corresponding control strategies, and feeds back to the actuator via the edge computing device to terminate the current operation.

[0012] The object of the invention of the present invention is realized as follows:

[0013] An intelligent state monitoring method for a low-power logging cable winch of the present invention integrally deploys the pre-trained and complete logging cable winch state monitoring network WM-YOLO on an edge computing device, providing good hardware support for video monitoring with the low-power and high-computing-power characteristics of the device; at the same time, obtaining the video data of four cameras placed above the wellhead and in front of the main cable winch, simultaneously extracting frames from the four video streams and combining them into an image vector composed of single-frame image data of the four video streams. Further, the image vector containing the wellhead and cable winch state information is sent into the neural network, and the edge device executes the forward inference process of the WM-YOLO network to perform target detection on the input image vector to obtain the detection result, and finally calculates and analyzes the output result; the normal result is displayed in real time, while for the abnormal result, while being displayed in real time, a warning message is issued in a timely manner.

[0014] Meanwhile, the intelligent status monitoring method for the low-power logging cable winch of the present invention also has the following beneficial effects:

[0015] (1) The present invention adopts computer vision technology and edge computing technology to realize real-time intelligent status monitoring of the logging cable winch and wellhead operations, and issues alarm information in real time for abnormal states, which helps the on-site staff to make corresponding strategies in a timely manner for abnormal states, saves labor costs to a certain extent, and improves work efficiency.

[0016] (2) Based on low-power edge computing devices, intelligent monitoring of the status of the logging cable winch and wellhead is realized, which has extremely strong low-power characteristics. At the same time, it improves the automation and intelligence levels of related operation scenarios. The miniaturization and microminiaturization of edge computing devices further reduce the occupied volume space of related devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flowchart of an intelligent status monitoring method for a low-power logging cable winch of the present invention;

[0018] Figure 2 is an architecture diagram of a specific implementation manner of an intelligent status monitoring method for a low-power logging cable winch of the present invention;

[0019] Figure 3 is an example of wellhead monitoring results; DETAILED DESCRIPTION OF THE INVENTION

[0020] The following describes the specific implementation manner of the present invention in conjunction with the drawings, so that those skilled in the art can better understand the present invention. It should be particularly noted that in the following description, when the detailed description of known functions and designs may dilute the main content of the present invention, these descriptions will be omitted here.

[0021] Embodiment

[0022] Figure 1 is a flowchart of an intelligent status monitoring method for a low-power logging cable winch of the present invention.

[0023] In this embodiment, as Figure 1 shown, an intelligent status monitoring method for a low-power logging cable winch of the present invention includes the following steps:

[0024] S1. Deploy the trained logging cable winch status monitoring network WM-YOLO as a whole on an edge computing device with low-power characteristics; automatically initialize the network parameters when the system is powered on;

[0025] S2. Obtain the wellhead and winch status monitoring image data;

[0026] S2.1. Select the target well for logging operations, and deploy a total of four network cameras at key positions on the logging derrick and the logging cable winch. The four cameras are directly connected to the switch to obtain video data in real time;

[0027] Among them, install No. 1 and No. 2 network cameras at symmetrical positions on the derrick above the wellhead, so that the shooting ranges of the two cameras cover the wellhead and part of the logging cable connected to the instrument, in order to obtain the status information at the wellhead and the image of the logging cable status;

[0028] Install No. 3 and No. 4 network cameras in front of the logging cable winch, so that the shooting ranges of the two cameras cover the front range of the main winch, in order to obtain the image of the main winch status;

[0029] S2.2. The low-power edge computing device is directly connected to the switch. By reading different network IP addresses, it simultaneously obtains the four-way video data and enters it into the cache, and saves the video data to the storage hard disk array.

[0030] In this embodiment, the intelligent status monitoring architecture of the low-power logging cable winch is as Figure 2 shown. Its entire system consists of on-site data acquisition, an edge computing terminal, and a visualization detection terminal. The four network cameras are connected to the edge computing terminal via the switch to simultaneously obtain real-time images of the wellhead and the winch side. The edge computing device is directly connected to the storage hard disk to store video data and monitoring results. The visualization terminal directly connected to the edge device displays the monitoring results in real time.

[0031] S2.3. The edge device converts the video stream data of each network camera into single-frame image data through video frame extraction technology, denoted as i represents the i-th camera, and f represents the f-th frame image formed by frame extraction.

[0032] S2.4. Repeat step S2.3 until all the four-way video data is converted into image data, and splice it to form an image vector composed of the f-th frame images of the four cameras at a certain moment

[0033]

[0034] S3. The edge computing device executes the forward inference process of the WM-YOLO network to obtain the target detection results in the single-frame image vector;

[0035] S3.1. The edge computing device sends the image frames obtained at a certain moment sequentially into the status monitoring network WM-YOLO, and obtains feature maps of different scales through continuous convolution and downsampling in the backbone network, denoted as

[0036]

[0037] Among them, j = 1, 2, 3 represents the ruler number, RC represents a convolutional block with a residual connection structure, and its convolutional kernel size is 3×3; CBS represents a network structure block composed of convolution-batch normalization-Silu activation function in sequence, SPPF represents a fast spatial pyramid pooling module, and the subscript * ×1 represents the number of repeated executions.

[0038] S3.2. Through the feature fusion network, fuse the obtained deep feature maps to obtain the fused feature map

[0039]

[0040] S3.2.1. Deep-level feature maps are fused with the sub-deep feature maps after upsampling and magnification to form the fused feature map

[0041]

[0042] Among them, Concat represents the tensor splicing operation, and Upsample represents the upsampling operation;

[0043] S3.2.2. Fused feature maps are fused with the feature map F 1 i again to form the fused feature map

[0044]

[0045] To better extract the status information of the logging cable winch, in the WM-YOLO algorithm, the feature maps obtained by upsampling are reversely fused with the deep-level feature maps again to construct a two-way feature fusion branch. Specifically as follows:

[0046] S3.2.3. Fuse the fused feature map with the fused feature map to obtain the fused feature map

[0047]

[0048] S3.2.4. Fuse the fused feature map with the deep-level feature map to obtain the fused feature map

[0049]

[0050] S3.3. Perform a convolution operation on the obtained fused feature map ;

[0051]

[0052] Among them, Conv represents a convolution operation with a convolution kernel size of 3×3;

[0053] S3.4. Perform object detection on the convolution result through the soft non-maximum suppression algorithm;

[0054]

[0055] Among them, Softnms represents the soft non-maximum suppression algorithm;

[0056] S4. The edge computing device outputs the real-time status information of the logging cable winch according to the real-time detected target result D f ;

[0057] S4.1. Read from the real-time detected target result D f and set a target detection box and corresponding status information for the detected targets in respectively. Among them, the heights of the target detection boxes are respectively denoted as In this embodiment, according to the detected targets in

[0058] , identify whether there are abnormalities such as cable bending caused by instrument jamming at the wellhead. If so, set it to the abnormal state; otherwise, it is the normal state. An example of the wellhead status monitoring result is shown in and . When the cable is in the normal state, it is in a vertical stress state and is marked as the normal state. When an instrument jamming accident occurs, the cable is abnormally stressed and bends, and it is marked as the abnormal monitoring result. Figure 3 As shown, when the cable is in the normal state, it is in a vertical stress state and is marked as the normal state. When an instrument jamming accident occurs, the cable is abnormally stressed and bends, and it is marked as the abnormal monitoring result.

[0059] According to the detected targets in , identify whether there are problems such as cable slack and winding disorder on one side of the winch. If so, set it to the abnormal state; otherwise, it is the normal state. In the normal state, we can identify the number of layers and turns of the cable wound on the main winch, so as to achieve the purpose of real-time estimating the cable winding length, and can output the main winch status information and the cable winding length information at the same time, as follows:

[0060] S4.2. Read the status information corresponding to the detected targets in . If all the status information is in the normal state, the edge computing device outputs the real-time status of the logging cable winch as normal, and then enters step (4.3); otherwise, the algorithm ends;

[0061] S4.3. Calculate and Average height of the target detection frame set in

[0062] S4.4. Calculate the remaining height of the cable side rail k is the camera calibration coefficient;

[0063] S4.5. Calculate the thickness of the cable wound on the drum R is the height of the side rails on both sides of the drum, and r is the diameter of the logging cable;

[0064] S4.6. Calculate the number of layers of the cable wound on the drum r is the diameter of the logging cable;

[0065] S4.7. Calculate the length L of the cable wound on the drum;

[0066] S4.7.1. Let the number of turns of the cable evenly wound on each layer of the drum be m, and the diameter corresponding to each layer of the cable wound on the drum be d l , l = 1, 2,..., N ed ;

[0067] S4.7.2. Calculate the length L of the cable evenly wound on each layer l = πd l m; In this embodiment, when calculating the outermost layer of the wound cable, the value of m needs to be determined according to the actual number of turns of the wound cable, because there may be a situation where the outermost layer is not fully wound.

[0068] S4.7.3. Calculate the length of the cable wound on the drum

[0069] S4.8. The edge computing device outputs the length L of the logging cable.

[0070] S5. Receive the output result in step S4. The edge computing device displays the normal state information of the cable winch. For abnormal state information, it issues a warning message in real time to prompt and notify the operator to take corresponding control strategies, and feeds back to the actuator via the edge computing device to terminate the current operation.

[0071] Although the above describes the illustrative specific embodiments of the present invention for the convenience of those skilled in the art to understand the present invention, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions made using the concept of the present invention are within the scope of protection.

Claims

1. A low-power logging cable winch intelligent state monitoring method, characterized in that: The steps include: (1) Deploy the fully trained logging cable winch status monitoring network WM-YOLO on a low-power edge computing device; automatically initialize network parameters when the system is powered on; (2) Obtaining wellhead and winch status monitoring image data; (3) The edge computing device executes the forward reasoning process of the WM-YOLO network to obtain the target detection results in the single-frame image vector; (4) The edge computing device detects the target result in real time D f Output real-time status information of logging cable winch; (5) The edge computing device displays the real-time status information of the logging cable winch. For abnormal status information, it issues a warning message in real time and notifies the operator to adopt the corresponding control strategy. The edge computing device then feeds back to the actuator to terminate the current operation.

2. The low-power logging cable winch intelligent state monitoring method according to claim 1 is characterized in that: The method for acquiring the image data for monitoring the wellhead and winch status is as follows: (2.1) Select the target well for logging operation, deploy a total of four network cameras on the logging derrick and logging cable winch, and connect the four cameras directly to the switch to obtain video data in real time; Among them, No. 1 and No. 2 network cameras are symmetrically installed above the logging derrick, so that the shooting range of the two cameras covers the wellhead and the logging cable connected to the instrument, so as to obtain the status information at the wellhead and the status image of the logging cable; Install No. 3 and No. 4 network cameras in front of the logging cable winch so that the shooting range of the two cameras covers the front range of the main winch to obtain the status image of the main winch; (2.2) The low-power edge computing device is directly connected to the switch, and simultaneously obtains four-way video data into the cache by reading different network IP addresses, and saves the video data to the storage hard disk array; (2.3) The edge computing device converts the video stream data of each network camera into single-frame image data through video frame extraction technology, denoted as X i f , i represents the i-th camera, f represents the f-th frame image formed by frame extraction; (2.4) Repeat step (2.3) until all the data of the four-channel video are converted into image data and spliced ​​to form an image vector composed of a certain moment, where the image vector composed of the image frames of the four cameras at the f-th frame is 3. The low-power logging cable winch intelligent state monitoring method according to claim 1 is characterized in that: The method for obtaining the target detection result in the single-frame image vector is: (3.1) The edge computing device takes the image frame obtained at a certain moment The images are sent to the state monitoring network WM-YOLO in sequence, and feature maps of different scales are obtained through continuous convolution and downsampling in the backbone network, which are recorded as Where j = 1, 2, 3 represents the scale number, RC represents the convolution block with residual connection structure, and its convolution kernel size is 3×3; CBS represents the network structure block composed of convolution-batch normalization-Silu activation function in sequence, SPPF represents the fast spatial pyramid pooling module, and the subscript * ×1 Indicates the number of repetitions; (3.2) The deep feature map obtained is fused through the feature fusion network to obtain a fused feature map (3.2.1) Deep feature map And the sub-deep feature map after upsampling Fusion to form a fusion feature map Among them, Concat represents the tensor concatenation operation, and Upsample represents the upsampling operation; (3.2.2) Fusion feature map With feature map Fusion again to form a fusion feature map (3.2.3) The fusion feature map And fusion feature map Fusion is performed to obtain a fusion feature map (3.2.4) The fusion feature map With deep feature maps Fusion is performed to obtain a fusion feature map (3.3) Obtain fusion feature map Perform convolution operation; Among them, Conv represents the convolution operation with a convolution kernel size of 3×3; (3.4) Target detection is performed on the convolution result through the soft non-maximum suppression algorithm; Among them, Softnms represents the soft non-maximum suppression algorithm; (3.5) Repeat step (3.1) to loop through each frame of the image vector to obtain real-time target detection results 4. The low-power logging cable winch intelligent state monitoring method according to claim 1 is characterized in that: In step (4), according to the real-time detection target result D f The method to output the real-time status information of the logging cable winch is: (4.1) From the real-time detection target result D f Read They are The detection target in sets the target detection frame and the corresponding status information, where the height of the target detection frame is recorded as (4.2), read The state information corresponding to the detection target is obtained. If the state information is normal, the edge computing device outputs the real-time state of the logging cable winch as normal, and then enters step (4.3); otherwise, the algorithm ends; (4.3), calculation and The mean height of the target detection box set in (4.4) Calculate the remaining height of the cable side guardrail k is the camera calibration coefficient; (4.5) Calculate the thickness of the cable wound on the drum R is the height of the side rails on both sides of the drum, and r is the diameter of the logging cable; (4.6) Calculate the number of layers of cable wound on the drum r is the diameter of the logging cable; (4.7) Calculate the length L of the cable wound on the drum; (4.7.1) Assume that the number of turns of each layer of cable wound evenly on the drum is m, and the diameter of each layer of cable wound on the drum is d l ,l=1,2,…,N ed ; (4.7.2) Calculate the length L of each layer of uniformly wound cable l =πd l m; (4.7.3) Calculate the length of the cable wound on the drum (4.8) The edge computing device outputs the length L of the logging cable.

Images