Pipe obstacle clearing robot, pipe obstacle clearing device and sewer pipe defect detection method

Abstract

The application relates to the field of urban drainage pipeline repair, in particular to a pipeline obstacle cleaning robot, a pipeline obstacle cleaning device and a drainage pipeline defect detection method. The pipeline obstacle cleaning robot comprises a mechanical arm unit and a running unit arranged in a first direction. The running unit is used for running in the first direction. The running unit comprises a body. The mechanical arm unit comprises an action head and a mechanical arm driving assembly. One end of the body in the first direction is a connecting end. The connecting end is connected with the mechanical arm unit. The mechanical arm driving assembly is used for driving the action head to rotate relative to the body. The application aims at solving some problems in pipeline obstacle cleaning and provides the pipeline obstacle cleaning robot, the pipeline obstacle cleaning device and the drainage pipeline defect detection method.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of urban drainage pipeline repair, in particular to a pipeline obstacle clearing robot, a pipeline obstacle clearing device and a drainage pipeline defect detection method. BACKGROUND

[0002] In urban sewage and rainwater drainage pipelines, there are soft obstacles such as silt and organic garbage, and hard obstacles such as foreign matter accumulation, building mortar and tree root intrusion. In addition, there are phenomena such as illegal construction in pipelines and old pipelines. When the above situations are serious, pipeline blockage is likely to occur, causing a larger range of waterlogging, causing serious loss to people's lives and property.

[0003] In order to ensure the smoothness of urban rainwater and sewage drainage pipelines, facilitate the timely emptying of urban road waterlogging, and solve the situation of waterlogging caused by drainage outlet blockage due to sudden increase of domestic sewage or rainstorm weather, it is necessary to clean various obstacles in the pipeline and repair the pipeline. For soft obstacles such as silt in the pipeline, high-pressure water flushing and other methods can be used to complete the desilting. Hard obstacles such as tree roots and cemented boards cannot be removed by the above methods. In particular, for large-diameter sewage and rainwater drainage pipelines, manual entry for desilting and obstacle removal is generally possible, but for relatively small-diameter pipelines, workers generally cannot enter for obstacle removal, and there is currently no effective trenchless obstacle removal equipment. SUMMARY

[0004] The present application aims to solve at least one technical problem involved in the background art, and provides a pipeline obstacle clearing robot, a pipeline obstacle clearing device and a drainage pipeline defect detection method.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0006] One aspect of the present application provides a pipeline obstacle clearing robot, comprising a mechanical arm unit and a traveling unit arranged in a first direction, the traveling unit being used for traveling in the first direction, the traveling unit comprising a body, the mechanical arm unit comprising an action head and a mechanical arm driving assembly, one end of the body being a connection end in the first direction, the connection end being connected with the mechanical arm unit, and the mechanical arm driving assembly being used for driving the action head to rotate relative to the body.

[0007] Optionally, the mechanical arm unit further comprises a first swing arm and a second swing arm, one end of the second swing arm is a pivot joint end in the length direction of the second swing arm, the other end of the second swing arm is a mounting end, the mounting end is used for mounting an action head, one end of the first swing arm is pivotally connected to the connecting end in the length direction of the first swing arm, the other end of the first swing arm is pivotally connected to the pivot joint end, the mechanical arm driving assembly comprises a first driving member and a second driving member, the first driving member is used for driving the first swing arm to swing relative to the body, and the second driving member is used for driving the second swing arm to swing relative to the first swing arm.

[0008] Optionally, the pipeline obstacle clearing robot is arranged in a strip shape extending in the first direction, or the pipeline obstacle clearing robot is arranged in a wedge shape gradually shrinking in cross-sectional size from the running unit to the mechanical arm unit; the mechanical arm driving assembly further comprises a rotation driving member, the rotation driving member is mounted on the connecting end, the first swing arm is connected to the connecting end through the rotation driving member, and the rotation driving member is used to drive the first swing arm to rotate about a straight line parallel to the first direction as an axis.

[0009] Optionally, the mechanical arm unit further comprises a connecting joint, the connecting joint is connected to the rotation driving member, the first swing arm comprises a first connecting plate and a second connecting plate, and two support rods arranged in parallel to each other, the two support rods are connected through the first connecting plate and the second connecting plate, one end of the support rod is pivotally connected to the pivot joint end in the length direction of the support rod, the other end of the support rod is pivotally connected to the connecting joint, the first connecting plate is arranged close to the pivot joint end in the length direction of the support rod, and the second connecting plate is arranged close to the connecting joint; the first driving member and the second driving member are both electric push rods, one end of the first driving member is pivotally connected to the connecting joint in the length direction of the first driving member, the other end of the first driving member is pivotally connected to the first connecting plate, one end of the second driving member is pivotally connected to the second connecting plate in the length direction of the second driving member, the other end of the second driving member is pivotally connected to the second swing arm, and the length direction of the first driving member intersects the length direction of the second driving member.

[0010] Optionally, the driving unit further comprises a jacking assembly, the jacking assembly comprises a jacking table, a connecting rod and a linear drive, the connecting rod and the jacking table are both above the body, one end of the connecting rod is pivoted to the body in the length direction of the connecting rod, the other end of the connecting rod is pivoted to the jacking table, one end of the linear drive is pivoted to the body in the length direction of the linear drive, the other end of the linear drive is pivoted to the connecting rod, so that the jacking table moves upward relative to the body as the linear drive extends, and the jacking table moves downward relative to the body as the linear drive shortens, and the jacking table is used to contact the inner wall of the pipeline.

[0011] Another aspect of the present application provides a pipeline obstacle removing device, comprising a front camera, a rear camera, an image processing device, a control system and the pipeline obstacle removing robot provided by the present application, the image processing device and the pipeline obstacle removing robot are both in communication connection with the control system, the image processing device is provided with an image recognition system, the driving unit comprises a jacking assembly installed above the body, the front camera and the rear camera are both in communication connection with the image processing device, the front camera is installed on the mechanical arm unit, and the rear camera is installed on the top surface of the body, the front camera is used to observe the pipeline environment and the running state of the action head in front of the pipeline obstacle removing robot, and the rear camera is used to observe the pipeline environment and the running state of the jacking assembly behind the pipeline obstacle removing robot.

[0012] Optionally, the control system comprises a host computer and a monitor, the monitor is in communication connection with the image processing device, the pipeline obstacle removing device further comprises a power supply system and a cable car, a coaxial cable connected to the monitor, a communication cable connected to the host computer and a power supply cable connected to the power supply system are converged into a total cable at the cable car, and the total cable is connected with the pipeline obstacle removing robot.

[0013] A third aspect of the present application provides a method for detecting defects of a drainage pipeline, which is executed by an image recognition system provided in an image processing device in the pipeline obstacle removing device, and the method comprises the following steps:

[0014] receiving drainage pipeline image data currently collected by the front camera or the rear camera;

[0015] inputting the drainage pipeline image data into a preset drainage pipeline defect detection model, so that the drainage pipeline defect detection model correspondingly outputs corresponding drainage pipeline defect recognition result data;

[0016] The sewer pipeline defect identification result data is sent to the control system, so that a user views the sewer pipeline defect identification result data from the control system, and corresponding control instructions are issued to the pipeline obstacle removing robot by using the control system.

[0017] Optionally, the sewer pipeline defect detection model is obtained by pre-training a preset YOLOv5 optimization model using historical sewer image data; the YOLOv5 optimization model is a neural network formed by adding a FasterNet block, a CA attention mechanism module and a CBAM attention mechanism module in a YOLOv5 model.

[0018] Optionally, before the sewer pipeline image data is input into the preset sewer pipeline defect detection model, the method further comprises:

[0019] The sewer pipeline image data sets of different sources are merged to obtain an initial data set containing each historical sewer image data and a corresponding label, wherein the label includes respective identifiers for indicating different defect types of the sewer pipeline, and the defect types include: no defect, bending, crack, debris, hole, joint offset, obstacle, utility invasion and tree root.

[0020] Each of the historical sewer image data in the initial data set is subjected to data enhancement processing to expand the initial data set and form a corresponding training data set.

[0021] The YOLOv5 optimization model is trained based on the training data set to obtain a sewer pipeline defect detection model for identifying the corresponding defect type of the sewer pipeline image data.

[0022] The technical scheme provided by the present application can achieve the following beneficial effects:

[0023] The pipeline obstacle removing robot, the pipeline obstacle removing device and the sewer pipeline defect detection method provided by the present application have the following beneficial effects: in use, the first direction, i.e. the running direction of the pipeline obstacle removing robot in the pipeline, is usually the length direction of the pipeline, the mechanical arm unit and the running unit in the pipeline obstacle removing robot are arranged in the length direction of the pipeline, which is more suitable than the arrangement of the mechanical arm unit and the running unit in the width direction of the pipeline or other arrangement manners, and the height and the width are smaller, so that the pipeline obstacle removing robot is less limited by the diameter of the pipeline in the application process, has stronger applicability, has a wider application range, and can be particularly applied to relatively small-diameter pipelines in which workers cannot enter to remove obstacles.

[0024] The additional technical features and advantages of the present application will be more apparent in the following description or can be understood through the specific implementation of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0025] To more clearly illustrate the technical solutions in the specific embodiments of the present application, the accompanying drawings needed to be used in the description of the specific embodiments will be briefly described in the following. Obviously, the accompanying drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those of ordinary skill in the art without any creative work on the premise of the accompanying drawings.

[0026] Figure 1 The left view structural schematic diagram of a first state of an embodiment of the pipeline obstacle removing robot provided by the present application;

[0027] Figure 2 The left view structural schematic diagram of a second state of an embodiment of the pipeline obstacle removing robot provided by the present application;

[0028] Figure 3 The left view structural schematic diagram of a second state of an embodiment of the pipeline obstacle removing robot provided by the present application;

[0029] Figure 4 The left view structural schematic diagram of a third state of an embodiment of the pipeline obstacle removing robot provided by the present application, wherein the position of the secondary swing arm 113 and the position of the primary swing arm 114 are both the limit positions of swinging upward relative to the body 121, and the position of the jacking table 122a is the limit position of lifting relative to the body 121;

[0030] Figure 5 The left view structural schematic diagram of a fourth state of an embodiment of the pipeline obstacle removing robot provided by the present application, wherein the position of the secondary swing arm 113 and the position of the primary swing arm 114 are both the limit positions of swinging downward relative to the body 121, and the position of the jacking table 122a is the limit position of lifting relative to the body 121;

[0031] Figure 6 The front view structural schematic diagram of the first state of an embodiment of the pipeline obstacle removing robot provided by the present application applied to a pipeline with an inner diameter of Φ250mm;

[0032] Figure 7 The front view structural schematic diagram of the first state of an embodiment of the pipeline obstacle removing robot provided by the present application applied to a pipeline with an inner diameter of Φ330mm;

[0033] Figure 8 The rear view structural schematic diagram of the second state of an embodiment of the pipeline obstacle removing robot provided by the present application;

[0034] Figure 9aAn embodiment structure schematic diagram of an action head provided by the embodiment of the present application;

[0035] Figure 9b An embodiment structure schematic diagram of an action head provided by the embodiment of the present application;

[0036] Figure 10 An embodiment structure schematic diagram of a pipeline obstacle removing device provided by the embodiment of the present application combined with an application scenario;

[0037] Figure 11 An electric control system schematic diagram of an embodiment of a pipeline obstacle removing device provided by the embodiment of the present application;

[0038] Figure 12 A flowchart schematic diagram of an embodiment of a drainage pipeline defect detection method provided by the embodiment of the present application;

[0039] FIG. 13(a) is an example schematic diagram of an architecture of a drainage pipeline defect detection model in a drainage pipeline defect detection method provided by the embodiment of the present application;

[0040] FIG. 13(b) is a structure schematic diagram of a convolution module in the drainage pipeline defect detection model shown in FIG. 13(a);

[0041] FIG. 13(c) is a structure schematic diagram of a C3-faster target detection module in the drainage pipeline defect detection model shown in FIG. 13(a);

[0042] FIG. 13(d) is a structure schematic diagram of a bottleneck convolution block in the C3-faster target detection module shown in FIG. 13(c);

[0043] FIG. 13(e) is a structure schematic diagram of a fast spatial pyramid pooling module in the drainage pipeline defect detection model shown in FIG. 13(a);

[0044] Figure 14 An example schematic diagram of a whole flow of a drainage pipeline defect detection provided by the embodiment of the present application.

[0045] Reference signs:

[0046] 100-pipeline obstacle removing robot;

[0047] 110-mechanical arm unit;

[0048] 111-action head;

[0049] 112-action motor;

[0050] 113-secondary swing arm;

[0051] 114-primary swing arm;

[0052] 114a-supporting rod

[0053] 114b-first connecting plate

[0054] 114c-second connecting plate

[0055] 115-connecting joint

[0056] 116-secondary driving member

[0057] 117-primary driving member

[0058] 120-traveling unit

[0059] 121-body

[0060] 122-jacking assembly

[0061] 122a-jacking platform

[0062] 122b-connecting rod

[0063] 122c-linear driving member

[0064] 123-in-wheel motor

[0065] 130-lifting ring

[0066] 140-front camera

[0067] 150-rear camera

[0068] 160-lifting column

[0069] 170-24V aviation plug

[0070] 180-signal aviation plug

[0071] 190-48V aviation plug

[0072] 200-power supply system

[0073] 210-power supply cable

[0074] 300-upper computer

[0075] 310-communication cable

[0076] 400-monitor

[0077] 410-coaxial cable

[0078] 500-lifting machine

[0079] 510-pulling steel wire rope

[0080] 600-rinsing equipment;

[0081] 610 - Water pipe;

[0082] 700-Cable Car;

[0083] 710 - Main cable. Detailed Implementation

[0084] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0085] In the description of this application, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0086] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0087] like Figures 1 to 8 As shown, one aspect of this application provides a pipeline clearing robot 100, including a robotic arm unit 110 and a travel unit 120 arranged in a first direction. The travel unit 120 is used for travel in the first direction. The travel unit 120 includes a body 121. The robotic arm unit 110 includes an action head 111 and a robotic arm drive assembly. One end of the body 121 in the first direction is a connecting end, which is connected to the robotic arm unit 110. The robotic arm drive assembly is used to drive the action head 111 to rotate relative to the body 121.

[0088] In the embodiment of the present application, the mechanical arm driving assembly is used to drive the action head 111 to rotate relative to the body 121, including at least one of the rotation of the action head 111 with its own shaft as the rotation shaft, and the rotation of the action head 111 with a certain position on the mechanical arm unit 110 or the traveling unit 120 as the shaft. In the embodiment of the present application, preferably, the action head 111 is detachable, and the action head 111 is matched with multiple types, which can be selected according to the working conditions. For example, the action head 111 can be a cutting head (such as shown in FIG. 1), a grinding disc (such as shown in FIG. 2), a saw blade (such as shown in FIG. 3), and the like. Figures 1 to 8 Figure 9a Figure 9b

[0089] In use, the pipeline obstacle clearing robot 100 provided in the present application is in the first direction, that is, the traveling direction of the pipeline obstacle clearing robot 100 traveling in the pipeline by the traveling unit 120, which is usually the length direction of the pipeline. The mechanical arm unit 110 (including the whole of the action head 111 and the mechanical arm driving assembly) and the traveling unit 120 in the pipeline obstacle clearing robot 100 are arranged in the length direction of the pipeline, which is smaller in height and width than the arrangement of the mechanical arm unit 110 and the traveling unit 120 in the width direction of the pipeline or other arrangement modes. In the application process, the pipeline obstacle clearing robot 100 is not easily limited by the diameter of the pipeline, has strong applicability, a wide application range, and can be particularly applied to some relatively small-diameter pipelines in which workers cannot enter to clear obstacles. (For example, for sewage and rainwater drainage pipelines with a diameter of 800 mm or more, workers can generally enter to clear silt and obstacles, but for pipelines with a diameter of less than 800 mm, especially a diameter of less than 600 mm, there is currently no effective trenchless obstacle clearing equipment in China.)

[0090] In addition, in the pipeline trenchless repair construction, the general process method is to first clean the pipeline of foreign matters, then lay a repair material such as epoxy resin, high-density polyethylene, and the like close to the inner wall of the pipeline, and finally use a method such as ultraviolet lamp irradiation to harden the repair material according to the characteristics of the repair material, to complete the repair. According to the current process, the repair method of laying a resin material and the like requires that the pipeline to be repaired is complete and has no branches. If the pipeline has branches, only segmented processing can be performed between the branches of the pipeline. In the repair construction of the pipeline with branches, especially the pipeline with a small diameter in which workers cannot enter, in order to realize one-time repair of the long-distance pipeline with branches, a special device is used to cut the hardened repair material at the branch outlet from the inside of the repaired pipeline after the repair material is laid on the inner wall of the long-distance pipeline and hardened. The pipeline obstacle clearing robot 100 provided in the embodiment of the present application can be used to cut the hardened repair material in the pipeline repair construction in which some workers cannot enter.

[0091] ​​​Optionally, the mechanical arm unit 110 further comprises a first swing arm 114 and a second swing arm 113, one end of the second swing arm 113 in the length direction of the second swing arm 113 is a pivot joint end, the other end of the second swing arm 113 is a mounting end, the mounting end is used for mounting the action head 111, one end of the first swing arm 114 in the length direction of the first swing arm 114 is pivotally connected to the connecting end, the other end of the first swing arm 114 is pivotally connected to the pivot joint end, the mechanical arm driving assembly comprises a first driving member 117 and a second driving member 116, the first driving member 117 is used for driving the first swing arm 114 to swing relative to the body 121, and the second driving member 116 is used for driving the second swing arm 113 to swing relative to the first swing arm 114. Preferably, the action head 111 is mounted on the rotor of the action motor 112, and the stator of the action motor 112 is mounted on the second swing arm 113. In this way, the two ends in the length direction of the first swing arm 114 and the second swing arm 113 are connected, and when the pipeline obstacle cleaning robot 100 is used, the length direction of the first swing arm 114 and the length direction of the second swing arm 113 can coincide with the first direction or be parallel to the first direction, and the first swing arm 114 and the second swing arm 113 are connected in other directions (for example, the width direction or the height direction of the first swing arm 114, etc.), so that the mechanical arm unit 110 occupies a smaller cross-sectional area of the pipeline, the mechanical arm unit 110 is more easily stretched into the pipeline, and the pipeline obstacle cleaning robot 100 is more suitable for obstacle cleaning and / or pollution cleaning work in a pipeline with a smaller diameter; at the same time, the pivot joint of the first swing arm 114 and the body 121 and the pivot joint of the second swing arm 113 and the first swing arm 114 realize the swinging of the first swing arm 114 and the second swing arm 113 within a certain range, and further make the action head 111 swing within a certain range and act on different pipeline positions, thereby improving the flexibility and applicability of the pipeline obstacle cleaning robot 100.

[0092] Optionally, the pipeline obstacle removing robot 100 is arranged in a strip shape extending in the first direction, or the pipeline obstacle removing robot 100 is arranged in a wedge shape gradually shrinking in cross-sectional size from the traveling unit 120 to the mechanical arm unit 110; the mechanical arm driving assembly further comprises a rotating driving member, the rotating driving member is installed on the connecting end, the primary swing arm 114 is connected with the connecting end through the rotating driving member, and the rotating driving member is used to drive the primary swing arm 114 to rotate around a straight line parallel to the first direction as an axis. The pipeline obstacle removing robot 100 is arranged in a strip shape or a wedge shape, so that the pipeline obstacle removing robot 100 is relatively regular and linear, and is more easily to travel in the pipeline. Since the rotating driving member is arranged, after the primary swing arm 114 and / or the secondary swing arm 113 swing, the linear structure is bent, the rotating driving member is used to drive the primary swing arm 114, the secondary swing arm 113 and the action head 111 to rotate together, and then the flexibility and the application range of the pipeline obstacle removing robot 100 are increased.

[0093] Optionally, the mechanical arm unit 110 further comprises a connecting joint 115 connected with the rotating driving member, the first swing arm 114 comprises a first connecting plate 114b and a second connecting plate 114c, and two support rods 114a arranged in parallel with each other, the two support rods 114a are connected through the first connecting plate 114b and the second connecting plate 114c, one end of the support rod 114a is pivoted to the pivoting end in the length direction of the support rod 114a, the other end of the support rod 114a is pivoted to the connecting joint 115, the first connecting plate 114b is arranged close to the pivoting end in the length direction of the support rod 114a, and the second connecting plate 114c is arranged close to the connecting joint 115; the first driving member 117 and the second driving member 116 are both electric push rods, one end of the first driving member 117 is pivoted to the connecting joint 115 in the length direction of the first driving member 117, the other end of the first driving member 117 is pivoted to the first connecting plate 114b, one end of the second driving member 116 is pivoted to the second connecting plate 114c in the length direction of the second driving member 116, and the other end of the second driving member 116 is pivoted to the second swing arm 113, and the length direction of the first driving member 117 intersects with the length direction of the second driving member 116. It can be understood that the length direction of the first driving member 117 does not intersect with the length direction of the second driving member 116, for example, when the pipeline obstacle removing robot 100 is arranged horizontally, assuming that the first connecting plate 114b and the second connecting plate 114c are both located above the support rod 114a, the first driving member 117 extends downwardly and obliquely from the first connecting plate 114b to the connecting joint 115, one end of the second driving member 116 is pivoted to the second connecting plate 114c, and the other end of the second driving member 116 is pivoted to the lower end of the second swing arm 113, so that the first driving member 117 and the second driving member 116 are arranged in a cross manner. In the embodiment, the first driving member 117 is preferably two, and the two first driving members 117 are located on both sides of the second driving member 116; during the movement of the first swing arm 114, the first connecting plate 114b and the second connecting plate 114c can both move above the first driving member 117 and the second driving member 116. The first swing arm 114 adopts two support rods 114a as the main skeleton structure, which appropriately reduces the weight of the first swing arm 114 while ensuring the appropriate strength of the first swing arm 114. Of course, the first driving member 117 and the second driving member 116 can also be electric motors and the like, and drive the first swing arm 114 and the second swing arm 113 to swing in combination with a transmission mechanism, but this form occupies a larger space, which can cause the pipeline obstacle removing robot 100 to increase in size.

[0094] Optionally, the driving unit 120 further comprises a jacking assembly 122, the jacking assembly 122 comprises a jacking platform 122a, a connecting rod 122b and a linear drive 122c, the connecting rod 122b and the jacking platform 122a are both above the body 121, one end of the connecting rod 122b is pivoted to the body 121 in the length direction of the connecting rod 122b, the other end of the connecting rod 122b is pivoted to the jacking platform 122a, one end of the linear drive 122c is pivoted to the body 121 in the length direction of the linear drive 122c, the other end of the linear drive 122c is pivoted to the connecting rod 122b, so that the jacking platform 122a moves upward relative to the body 121 as the linear drive 122c extends, and the jacking platform 122a moves downward relative to the body 121 as the linear drive 122c shortens, and the jacking platform 122a is used to contact the inner wall of the pipeline. In this way, before the pipeline obstacle removing robot 100 removes obstacles and / or cleans dirt, the jacking platform 122a can be raised by extending the linear drive 122c, the jacking platform 122a contacts and presses the inner wall of the pipeline, and then the pipeline obstacle removing robot 100 is fixed more firmly in the pipeline, so that the pipeline obstacle removing robot 100 is not prone to overturning during operation. Preferably, the jacking assembly 122 comprises four connecting rods 122b and two linear drives 122c, the two linear drives 122c are located on both sides of the body 121, and two connecting rods 122b are correspondingly pivoted to the two linear drives 122c; in the embodiment of the application, the jacking platform 122a, the connecting rod 122b and the linear drive 122c are detachably connected, the linear drive 122c and the body 121 are detachably connected, and the connecting rod 122b has multiple sizes and can be replaced according to the diameter of the pipeline, so as to increase or decrease the height that the jacking assembly 122 can jacking; preferably, the driving unit 120 comprises four hub motors 123, two groups of the four hub motors 123 are detachably installed on both sides of the body 121, the driving unit 120 can comprise multiple groups of hub extension shafts with different sizes, each group has four hub extension shafts, and the four hub extension shafts are correspondingly connected between the four hub motors 123 and the body 121, in use, the pipeline obstacle removing robot 100 can select appropriate hub extension shafts according to the diameter of the pipeline, so that the pipeline obstacle removing robot 100 can move more smoothly in the pipeline, and preferably, the pipeline obstacle removing robot 100 can complete actions such as advancing, retreating and turning in the pipeline through the hub motor 123.

[0095] The pipeline obstacle cleaning robot 100 adopts a 4-DOF mechanical arm unit 110, i.e. swing DOF L1, L2 and rotation DOF R1, R2, and can realize cleaning operation in any direction and at any position in the whole working range of the pipeline wall. The pipeline obstacle cleaning robot 100 has three types of action heads 111 (cutters), i.e. cutting head, saw blade and grinding blade, and satisfies three types of installation interfaces, and can perform cleaning operation according to different obstacles. The pipeline obstacle cleaning robot 100 has a jacking assembly 122, and in order to prevent accidental overturning, the pipeline obstacle cleaning robot 100 can realize compression and fixation of the body and the pipeline wall through the jacking assembly 122 during traveling and obstacle cleaning operation in the pipeline wall. Generally, the pipeline obstacle cleaning robot 100 can adapt to the pipeline operation with an inner diameter of 250-330 mm through the lifting and lowering of the jacking assembly 122. If it is necessary to increase the working pipe diameter, the jacking assembly 122 height can be expanded and the wheel hub extension shaft can be added, so that the pipeline obstacle cleaning robot 100 can adapt to the pipeline operation with a maximum inner diameter of 780 mm. The jacking assembly 122 can increase the jacking height by 231.5 mm, and the length of the wheel hub extension shaft can be selected as 237 mm. The pipeline obstacle cleaning robot 100 has hoisting and traction functions, so as to facilitate the pipeline obstacle cleaning robot 100 to go down into the well and be recovered. The pipeline obstacle cleaning robot 100 has a waterproof function, and can be conveniently opened for maintenance operation.

[0096] As shown in Figure 10 Another aspect of the present application provides a pipeline obstacle cleaning device, which comprises a front camera 140, a rear camera 150, an image processing device, a control system and the pipeline obstacle cleaning robot 100 provided in the embodiments of the present application. The image processing device and the pipeline obstacle cleaning robot 100 are in communication connection with the control system. An image recognition system is arranged in the image processing device. The traveling unit 120 comprises a jacking assembly 122 arranged above the body 121. The front camera 140 and the rear camera 150 are in communication connection with the image processing device. The front camera 140 is arranged on the mechanical arm unit 110, and the rear camera 150 is arranged on the top surface of the body 121. The front camera 140 is used for observing the pipeline environment in front of the pipeline obstacle cleaning robot 100 and the running state of the action head 111. The rear camera 150 is used for observing the pipeline environment behind the pipeline obstacle cleaning robot 100 and the running state of the jacking assembly 122.

[0097] The pipeline obstacle removing device provided in the embodiment of the application adopts the pipeline obstacle removing robot 100 provided in the embodiment of the application, the mechanical arm unit 110 (including the whole of the action head 111 and the mechanical arm driving assembly) and the running unit 120 in the pipeline obstacle removing robot 100 are arranged in the length direction of the pipeline, and the height and the width are smaller compared with the arrangement of the mechanical arm unit 110 and the running unit 120 in the width direction of the pipeline or other arrangement modes, the pipeline obstacle removing robot 100 is not easily limited by the diameter of the pipeline in the application process, the pipeline obstacle removing robot 100 has stronger applicability and wider application range, and in particular, the pipeline obstacle removing robot 100 can be applied to some relatively small-diameter pipelines in which the staff cannot enter to remove obstacles; and through the setting of the front camera 140, the rear camera 150 and the image processing device, the situation in the pipeline can be obtained, and the operator can operate the pipeline obstacle removing robot 100 according to the situation in the pipeline through the control system, and the control system can specifically operate the movement of the mechanical arm driving assembly, the hub motor 123, the front camera 140 and the rear camera 150 and the like. In the embodiment of the application, the front camera 140 and the rear camera 150 can be installed on the body 121 through a holder, and then the rotation of the front camera 140 and the rear camera 150 relative to the body 121 is realized, and the holder can be driven by a rudder engine. The image processing device can adopt a video board, and the video board is integrated with an image recognition system. The pipeline obstacle removing device provided in the embodiment of the application preferably adopts a high-definition night vision camera for the front camera 140 and the rear camera 150, and the pipeline environment in front of and behind the pipeline obstacle removing robot 100 and the working state of the pipeline obstacle removing robot 100 can be observed in real time.

[0098] Optionally, the pipeline obstacle removing device provided by the embodiment of the application, the control system comprises a host computer 300 and a monitor 400, the monitor 400 is in communication connection with the image processing device, the pipeline obstacle removing device further comprises a power supply system 200 and a cable car 700, a coaxial cable 410 connected to the monitor 400, a communication cable 310 connected to the host computer 300 and a power supply cable 210 connected to the power supply system 200 are converged into a total cable 710 at the cable car 700, and the total cable 710 is connected with the pipeline obstacle removing robot 100. Preferably, the control system further comprises a control board and a manipulator, the manipulator is in communication connection with the host computer 300, and the control board is in communication connection with the host computer 300 and the pipeline obstacle removing robot 100. In the embodiment of the application, the hoisting ring 130 is installed at the tail end of the body 121, the hoisting column 160 is arranged on both sides of the body 121, and the body 121 has a waterproof function; when the pipeline obstacle removing device provided by the embodiment of the application is used, the traction steel wire rope 510 is bound at the hoisting ring 130 and the hoisting column 160, and the pipeline obstacle removing robot 100 is sent into the shaft through the hoisting machine 500, the total cable 710 is continuously released by the cable car 700 as the pipeline obstacle removing robot 100 extends into the shaft, the total cable 710 is continuously elongated, the pipeline obstacle removing robot 100 reaches the horizontal pipe entrance and drives into the horizontal pipe to start the obstacle removing and / or pollution removing operation; preferably, water is continuously flushed into the horizontal pipe through the flushing device 600 and the matched water pipe 610 at the other end of the horizontal pipe, so that the cleaned garbage and pollution are flushed away. Preferably, the 48V navigation plug 190, the 24V navigation plug 170 and the signal navigation plug 180 for connecting with the total cable 710 are further installed at the tail end of the body 121.

[0099] In the embodiments of the present application, preferably, the pipeline obstacle removing robot 100 body adopts a cable power supply and communication mode. The cable power supply voltage adopts 48V and 24V power supply, and the pipeline obstacle removing robot 100 contains a 24V-to-12V DC / DC converter. To ensure the safety of the system and the operator. The communication line contains two groups of twisted pair for communication, which respectively transmit 485 and CAN signals, contains two groups of coaxial cables, which transmit video signals; the pipeline obstacle removing robot 100 adopts wheel hub motor to drive forward and backward, adopts electric push rod to drive the lifting platform 122a to lift, adopts electric push rod to drive the first swing arm 114 and the second swing arm 113 to lift, uses waterproof motor to drive the action head 111 to rotate, adopts hollow joint module (composed of connecting joint 115 and rotating driving part) to drive the first swing arm 114, the second swing arm 113 and the action head 111 to roll, and adopts a rudder to drive the pan-tilt of the front camera 140 and the rear camera 150 to rotate. The wheel hub motor 123 adopts a wheel hub servo motor, the power supply voltage is 24V, the driver adopts a digital servo double-wheel hub motor one-to-two driver, there are 4 motors and 2 drivers. RS485 bus communication and modbus-RTU protocol are adopted. The electric push rod adopts a stainless steel waterproof high-performance push rod, the power supply voltage is 24V, the driver adopts a digital one-to-one and one-to-two controller, which can realize push rod 0-full range fixed point control, single push rod fixed point control and double push rod synchronous fixed point control, and has overcurrent protection function. RS485 bus communication and modbus-RTU protocol are adopted. The waterproof motor driving the action head 111 adopts a brushless DC inner rotor high-power waterproof motor, adopts brushless electronic control, and is adjustable with 48V / 24V power supply. Preferably, the motor (rotating driving part) shaft in the hollow joint module has a circular hole for controlling the wire to pass through the inside and outside of the body. The power supply voltage is 48V, and CAN bus communication is adopted. The pan-tilt driving rudder of the front camera 140 and the rear camera 150 adopts a micro servo motor, the power supply voltage is 24V, RS485 bus communication and modbus-RTU protocol are adopted. The lens of the front camera 140 and the rear camera 150 adopts a vehicle-mounted white light color camera, which is configured with a 12V power supply LED light belt.

[0100] In order to provide a more effective sewer defect detection method for the pipe obstacle clearing device on the basis of the above-mentioned pipe obstacle clearing device, the application also considers automatic defect recognition of the sewer images collected by the front camera or the rear camera. With the increase of the installation and operation time of the sewer network, due to factors such as the basic structure of the pipe, the interface quality, etc., there may be structural defects such as leakage, corrosion, rupture, deformation, misalignment, disconnection, shedding of sealing materials, penetration of foreign matters, and dark connection of branch pipes, as well as functional defects such as sedimentation, scaling, obstacles, residual wall dam roots and tree roots. Structural defects refer to damage to the structure of the pipe and the inspection well, affecting the strength, stiffness and service life of the pipe; functional defects refer to changes in the water section caused by non-structural defects of the pipe, which weaken the smoothness of the pipe network. Drainage is not smooth due to pipe defects, especially during heavy rain.

[0101] Regular detection and maintenance of sewer pipes can timely detect pipe defects, avoid or reduce accidents caused by pipe defects, provide scientific basis for pipe maintenance and repair, and prolong the service life of the pipe. The traditional detection method is to enter the pipe for detection by workers carrying simple tools, but due to the poor internal environment of the sewer pipe, workers entering the internal environment of the sewer pipe for work are dangerous. Due to the seriousness of the sewer pipe defect problem, relevant researchers or workers have developed various new technologies to detect the internal environment of the sewer pipe, including periscope method, sonar method, CCTV pipe endoscopy method, etc. Compared with the traditional method, the safety and work efficiency of the new technology are significantly improved, which can more clearly and intuitively show the problems inside the pipe.

[0102] The current sewer defect detection method mainly uses the closed-circuit television (CCTV) detection method, which uses a pipe detection robot to enter the sewer pipe to record real scene videos. The videos are browsed and identified by professional technicians to determine the type and grade of pipe defects. This process takes a long time, and during this time, the technicians will experience visual fatigue, and the accuracy of sewer defect determination is prone to fluctuation.

[0103] Therefore, in order to further solve the above problems, the application considers using a deep learning method to design a sewer defect detection method in combination with the aforementioned pipe obstacle clearing device. The method is executed by an image recognition system provided in an image processing device in the pipe obstacle clearing device, as shown in Figure 12 The sewer defect detection method specifically includes the following contents:

[0104] Step 10: receiving the sewer pipe image data currently collected by the front camera or the rear camera.

[0105] Step 20: inputting the sewer pipeline image data into a preset sewer pipeline defect detection model to correspondingly output corresponding sewer pipeline defect identification result data.

[0106] Step 30: sending the sewer pipeline defect identification result data to the control system, so that the user views the sewer pipeline defect identification result data from the control system, and uses the control system to issue corresponding control instructions to the pipeline obstacle removing robot.

[0107] However, with the development of technologies such as deep learning, more and more methods can automatically detect sewer pipeline defects instead of manual detection. Some scholars have proposed an image recognition algorithm applying feature extraction and machine learning method, using support vector machine method (SVM) to classify sewer pipeline defects. Some scholars have proposed a machine learning method applying a set of random forest classifiers to identify fault types, but the above two machine learning methods need to manually extract features and cannot adapt to the complex and variable environment of sewer pipelines in the real world. Some scholars use convolutional neural network (CNN) in sewer pipeline defect classification, and some scholars use InceptionV3 to build a classifier to identify sewer pipeline defects. Since the above two methods are single-label classifiers, and the sewer pipeline defects in the real world are often multiple defects.

[0108] Some researchers also use semantic segmentation models to identify multiple defects. Some scholars have proposed a semantic segmentation network named PipeUNet to identify typical defects such as cracks, leaks, joint offsets, and foreign object invasions. Some scholars have proposed a neural network DilaSeg-CRF that combines convolutional neural network (CNN) and dense conditional random field (CRF) to improve segmentation accuracy to identify three common sewer pipeline defects. However, the above two methods are easily affected by lighting conditions and video resolution.

[0109] At the same time, some researchers use target detection models for sewer pipeline defect detection tasks. Some scholars have proposed a sewer pipeline defect detection model based on Faster R-CNN algorithm using clustering analysis method, and use VGG, AlexNet, GoogleNet, and ResNet to replace the feature extraction layer in Faster R-CNN network, effectively improving the accuracy of sewer pipeline defect detection and identification. Some scholars have proposed an automatic method based on convolutional neural network (R-CNN), which is suitable for high-precision and fast sewer pipeline defect detection after training. The target detection model obtained by the above network training indeed has high accuracy, but it ignores the robustness of the detection model, training cost, detection FPS, etc.

[0110] Based on this, in another embodiment of the drainage pipeline defect detection method provided in the present application, in view of the problem that the original YOLOv5 algorithm is not accurate in detecting in a dark and complex background of a drainage pipeline, the drainage pipeline defect detection model in the drainage pipeline defect detection method is pre-trained on a preset YOLOv5 optimization model after training.

[0111] The YOLOv5 optimization model is a neural network formed after adding FasterNet blocks, CA attention mechanism modules and CBAM attention mechanism modules in the YOLOv5 model.

[0112] The present application optimizes the network architecture of the original YOLO v5 in combination with the characteristics of the pipeline detection robot, proposes a drainage pipeline defect detection algorithm based on improved YOLO v5, uses the image data obtained by the cooperation of the on-board camera and the aperture of the pipeline detection robot as the input of the pipeline defect detection model, realizes the detection of defects in a drainage pipeline with a diameter of more than 300 mm, and adds (convolutional block attention module, CBAM) and (Coordinate Attention, CA) attention mechanisms in the network in view of the characteristics of the dark environment and complex background in the drainage pipeline, improves the accuracy of the detection model in dark conditions, reduces the influence of complex background on the detection result, makes the model more focused on the extraction of drainage pipeline defect features, and strengthens the learning of drainage pipeline defect image region features; referring to the idea of the FasterNet Block network module, replacing part of the convolutional modules in the C3 module in the original network with PConv, reducing the requirement for GPU on the basis of lightening the detection model, speeding up the training speed and not losing the detection accuracy of the model.

[0113] The specific description of the drainage pipeline defect detection algorithm is as follows:

[0114] (1) YOLO v5 network structure

[0115] YOLO (You Only Look Once) is an end-to-end detection method proposed by Redmon et al., which is now updated to version 8. It converts the detection problem into a regression problem and is a lightweight and high-precision target detection model based on deep learning. According to the comparison of training speed, parameter amount and accuracy of different YOLO versions, YOLOv5 is more suitable for drainage pipeline defect detection tasks.

[0116] In the YOLOv5 network, according to the different depth multiples and width multiples, YOLOv5s, YOLOv5m, YOLOv5l, YOLOv5x four structures are realized, with the increase of depth and width, the detection accuracy of the target detection model is continuously improved, but the training speed is continuously slowed down, in order to make the sewer defect detection task have a lightweight model that can realize on-site detection, YOLOv5s is selected, and the pictures in the open source data set of sewer are used for training, a total of 8 defects are detected, which are flexion, crack, debris, hole, joint offset, obstacle, utility invasion and tree root.

[0117] The backbone segment is the main part of the network, including the convolution module Conv (CBS), C3, SPPF and CA module. A CBS module is composed of Conv2d+Batch Normalization+sigmoid linear unit, which sequentially performs two-dimensional convolution, regularization operation and activation operation on the input; the C3 module is composed of the CBS module and the residual structure module, the input of the residual structure module is subjected to two convolution modules and addition operation with the original value, which completes the residual feature transfer without increasing the output depth; the SPPF module first uses three multi-scale maximum pooling modules in series on the input, then performs multi-scale fusion, and finally restores the output to the same size as the input; the CA attention module considers the direction-related position information while obtaining the information between channels, which helps the model to better locate and identify.

[0118] The neck segment part has the characteristics of feature fusion from top to bottom and from bottom to top, which fuses the feature information output by the Backbone with high-level feature information, and then performs down-sampling to aggregate shallow feature information, enriching the image feature information.

[0119] The head segment part outputs three scales of prediction maps at the same time, which are suitable for detecting small, medium and large targets respectively. GIOU_loss is used as the loss function of the image rectangular frame, and then the target frame is screened through non-maximum suppression (NMS), finally the prediction class with the highest confidence is output, and the frame coordinates are returned.

[0120] (2) YOLO v5 loss function

[0121] In the head segment, the most important thing is to select the appropriate loss function. The loss function in the YOLO v5 target detection algorithm is composed of three parts: rectangular frame loss (lossrect), confidence loss (lossobj) and classification loss (lossclc), which are used to determine the distance between the predicted information and the real information. The closer the predicted information and the real information, the smaller the value of the loss function.

[0122]

[0123] YOLO v5(v6.1) uses the loss function (Generalized intersection over union loss, GIoU) to calculate the rectangular frame loss, which is based on (Intersection over union loss, IoU) and adds the area S3 of the smallest rectangular frame surrounding rectangular frame A and rectangular frame B to the calculation.

[0124] That is:

[0125]

[0126] From the above formula, GIoU adds (S3-S2) / S3 to IoU.

[0127]

[0128] Where S1 is the overlapping area, S2 is the area of the figure composed of rectangular frame A and rectangular frame B, so IoU is also called intersection ratio.

[0129] From the above, S3-S2 is the area of the white area in the dashed line box, that is, the blank area in the dashed line box that does not belong to A or B, so (S3-S2) / S3 is the proportion of the blank area to the area of the dashed line box. The larger this proportion, the farther A and B are, the smaller the overlap is, and vice versa.

[0130] The value range of GIoU is-1~1, when A and B have no overlapping area, IoU is 0, then GIoU takes a negative value, in the extreme case, when A and B have no overlapping area and the distance is infinite, (S3-S2) / S3 is equal to 1, then GIoU takes-1; Another extreme case, when A and B completely overlap, (S3-S2) / S3 is equal to 0, IoU is 1, then GIoU takes 1.

[0131] Therefore, GIoU solves the problem that IoU is always 0 when A and B have no overlapping area.

[0132] Finally, the calculation formula of GIoU loss is obtained:

[0133]

[0134] (3) Improved YOLO v5 network structure

[0135] In recent years, attention mechanisms have been increasingly applied to the field of image processing. For the YOLOv5 algorithm, although it can obtain feature information of objects at different resolutions and enhance model performance through feature fusion based on network structure design, it is prone to inaccurate detection of pipe defects in the background of drainage pipes and changing environments.

[0136] 3.1) CBAM attention mechanism

[0137] By incorporating the CBAM attention mechanism, the influence of background changes on the detection results can be weakened, making the model more focused on feature learning and extraction of pipe defects. Meanwhile, differences between different pixel categories, channel features, and context associations are taken into account based on the already obtained pipe defect feature information at different resolutions.

[0138] CBAM includes two independent sub-modules: channel attention module and spatial attention module, which perform attention operations on channels and spaces respectively. In the channel attention module, the weights between feature maps are redistributed to increase attention to key feature maps and weaken the influence of redundant feature maps on recognition results. First, the input feature map is duplicated into two parts, which are subjected to global maximum pooling and global average pooling based on width and height, respectively, to obtain two feature maps. Then, they are sent to a two-layer multi-layer perceptron (MLP), which is shared by the two feature maps. Finally, the generated features are transmitted to the spatial attention module.

[0139] In the spatial attention module, the output feature map of the channel attention module is used as the input feature map of the spatial attention module. First, global maximum pooling and global average pooling based on channels are performed to obtain two feature maps. Then, these two feature maps are concatenated based on channels, followed by convolution operation to reduce dimension to one channel. Finally, the spatial attention feature is generated through the sigmoid activation function, and the final output feature of the spatial attention module is multiplied with the input feature to obtain the feature map output by the CBAM attention mechanism.

[0140] For the complex pipe environment and the feature of dim distance, by introducing the attention mechanism module, not only can the learning of pipe defect image region features be enhanced, but also the model size can be reduced, the computing power expenditure can be reduced, and the model can be easily integrated into existing network architectures.

[0141] The performance evaluation results of the CBAM attention mechanism are shown in Table 2.

[0142] Table 2

[0143]

[0144] 3.2) CA attention mechanism

[0145] Some scholars pointed out that SE attention ignored the spatial feature problem, and CBAM attention ignored the long-distance dependence problem. A CA (Coordinate Attention) attention mechanism was proposed, which not only considered channel information, but also incorporated direction-related position information. The model can better locate and identify targets. At the same time, the CA attention mechanism is lightweight and can be inserted into existing network structures, which improves performance in detection and segmentation tasks.

[0146] In order to retain spatial information and obtain long-distance channel-dependent position information, the CA attention mechanism decomposes the global average pooling into X-direction and Y-direction pooling to generate C H 1and C 1 Wfeature maps, which is different from the single feature vector generated by the SE attention method. This allows the attention module to obtain long-distance dependence along one spatial direction while preserving position information along the other spatial direction, which helps the network better locate the target. Further information fusion of the feature maps obtained in the previous step is performed, and the Concate operation is performed to obtain the feature maps. The convolution kernel size is 1 1, and the activation operation is performed, and then the split operation is performed along the spatial dimension to divide it into C / r H 1and C / r 1 W, which is further combined with the Sigmoid activation function after dimensionality reduction using a convolution kernel size of 1 1to obtain the final attention vector. Finally, the original feature map is weighted by multiplication to obtain the final feature map with attention weights in the X and Y directions.

[0147] The performance evaluation experiment results of introducing the CA attention mechanism are shown in Table 3.

[0148] Table 3

[0149]

[0150] 3-3) FasterNet

[0151] In the field of sewer defect detection, defect detection is carried out in real time in embedded devices or on-site computing devices. The computing performance of large models often cannot meet the demand, so the defect detection model needs to be lightweight. The backbone network of YOLOv5 (v6.1) adopts the CSP structure. Due to its complex network structure and huge parameter quantity, although it can effectively extract feature information, it slows down the detection speed. The present application uses the fast lightweight network module FasterNet Block as the feature extraction network of the entire model, reduces GPU occupation, speeds up the detection speed and reduces the parameter quantity by 17%.

[0152] FasterNet Block is a new neural network module further proposed using a new convolution (PConv), which runs faster while reducing the parameter quantity.

[0153] PConv only applies a regular convolution to part of the input channels for spatial feature extraction, and retains the remaining channels, reducing memory access and redundant calculations. In order to fully and effectively utilize all channel information, a pointwise convolution (PWConv) is added after PConv, which has a receptive field on the input feature map like a T-shaped convolution. Compared with ordinary convolution, it pays more attention to the center position. According to the Frobenius norm of the intermediate position, this position is often more important in the feature map.

[0154] The performance evaluation experiment results of introducing the FasterNet Block module are shown in Table 4.

[0155] Table 4

[0156]

[0157] (4) Sewer defect detection model

[0158] Due to the limited computing power of the sewer detection on-site server, considering the model operation overhead, the present application proposes a sewer defect detection algorithm based on the improved YOLO v5 algorithm, selects YOLO v5s as the basis, and performs algorithm improvement, as shown in FIGS. 13 (a) to 13 (e).

[0159] The FasterNet Block module is inserted into the C3 module in the YOLOv5s model to reduce the algorithm calculation pressure and reduce the model size; the CA attention mechanism is inserted between the backbone layer and the neck layer to obtain the information between channels and also consider the direction-related position information, which is helpful for better positioning and identification of the model; the CBAM attention mechanism is added to the C3 module in the neck layer to automatically obtain the importance of each feature channel and feature space, and the obtained importance is used to enhance or suppress the current features to form a sewer defect detection model.

[0160] The performance evaluation experiment results of the sewer defect detection model are shown in Table 5.

[0161] Table 5

[0162]

[0163] As shown in Figure 14 , the on-board camera and aperture of the sewer detection robot (i.e., the aforementioned pipe obstacle removal robot) are matched to collect images of the situation in the sewer pipe. The video line and video acquisition device are connected with the field pipe robot service host computer to establish a communication connection, realize the transmission of image data, and are used for real-time detection of the sewer defect detection algorithm in the service host computer. Finally, the detection results are displayed through the man-machine interaction interface to realize real-time inspection and automatic identification and archiving of the sewer defect detection image data.

[0164] On the basis of the above, in order to further improve the detection accuracy of the sewer defect detection, in another embodiment of the sewer defect detection method provided in the application, referring to Figure 10 , the step 200 in the sewer defect detection method further specifically comprises the following contents before the step 200:

[0165] Step 01: merging and processing different sources of sewer pipe image data sets to obtain an initial data set containing each sewer pipe historical image data and corresponding labels, wherein the labels include respective labels for indicating different defect types of the sewer pipe, and the defect types include: no defect, bending, crack, debris, hole, joint offset, obstacle, utility intrusion and tree root;

[0166] Step 02: performing data enhancement processing on each of the sewer pipe historical image data in the initial data set to expand the initial data set and form a corresponding training data set;

[0167] Step 03: training the YOLOv5 optimization model based on the training data set to obtain a sewer defect detection model for identifying the defect type corresponding to the sewer pipe image data.

[0168] We merged two open source datasets Storm drain model Dataset and Pipe root Dataset from the Robotflow Universe website, and after merging, there were 1317 original images, and there were 8 detection targets, namely flexure, crack, debris, hole, joint offset, obstacle, utility intrusion, and tree root.

[0169] According to the different stages before and after loading the image by the deep learning model, the data enhancement method is divided into offline enhancement and online enhancement. For the offline enhancement method, geometric transformation (flipping, translation, etc.) that does not change the content of the image itself and color transformation (noise, blur, etc.) that changes the content of the image itself are included. For the online enhancement method, taking YOLOv5(v6.1) as an example, including Mosaic operation (9 picture splicing), Copy paste operation (paste part of the target in the picture), Random affine operation (scaling and translation), MixUp operation (two pictures are superimposed after adjusting the transparency), etc.

[0170] Online enhancement is more suitable for large datasets because it uses GPU for calculation. Compared with online enhancement, small dataset samples are more suitable for offline enhancement. In the data preprocessing stage, the imgaug library is used to randomly horizontally flip the original image with a probability of 50%, vertically flip with a probability of 20%, sharpen, adjust brightness, and randomly adjust the hue to obtain the enhanced images of the dataset. The dataset is expanded from 1317 original images to 9012 pictures. Among them, the training set is 7788, and the verification set is 1224.

[0171] That is, the two open source datasets Storm drain model Dataset and Pipe root Dataset are merged, and after merging, there are 1317 original images, and there are 8 detection targets, namely flexure, crack, debris, hole, joint offset, obstacle, utility intrusion, and tree root.

[0172] In the data preprocessing stage, the dataset is expanded and the robustness of the detection model is improved by changing the color temperature, brightness, adding noise, and randomly removing pixels, etc. The new drainage pipe defect map obtained after the detection task expands the dataset and is trained to adapt to different drainage pipe environments.

[0173] In summary, due to the installation age, installation method and pipe material of the drainage pipe, various functional defects and structural defects often occur in the drainage pipe, leading to poor drainage of the drainage pipe, and even affecting the ground. The application provides a drainage pipe defect detection method based on improved YOLOv5. The imgaug library function is used to expand the original data set through rotation, sharpening, brightness, hue, and pixel removal operations, increase the data learning sample, and improve the model robustness. By integrating the CA and CBAM attention mechanism modules, the feature extraction and feature learning of the drainage pipe defect image are strengthened, and the influence of the dark environment of the drainage pipe on the detection result is reduced. On this basis, the FasterNetBlock fast lightweight network module based on Pconv is used for feature extraction, reducing the GPU occupation and speeding up the detection speed. The parameter amount is 84.05% of the original algorithm parameter amount. The experimental results show that the improved lightweight algorithm proposed in the application performs well in real samples, with an average accuracy of 98.8%, which is 2.6% higher than the original YOLOv5 target detection algorithm, and the false positive rate and the false negative rate in the dark background of the drainage pipe are significantly reduced, which can be effectively applied to actual detection work.

[0174] As shown in Figure 11 The total power supply of the pipe clearing device adopts 220V alternating voltage and provides voltage power for each part of the pipe clearing robot 100. The pipe clearing robot 100 adopts low-voltage 12V, 24V and 48V power supply. The upper computer realizes communication and control with each part of the pipe clearing robot 100 through the control board. The video board transmits the image information observed by the front camera 140 and the rear camera 150 to the monitor through a coaxial cable for real-time viewing and automatic detection. In combination with the manipulator, the pipe clearing robot 100 can be controlled accordingly. When the equipment is abnormal, the upper computer interface program can alarm and prompt the related information.

[0175] The upper computer monitoring part of the pipe clearing robot 100 is composed of a power box, a monitoring computer and its software. The main equipment of the power box includes:

[0176] (1) 220VAC-48VDC high-power converter. Providing 48V DC power supply for the robot;

[0177] (2) 220VAC-24VDC high-power converter. Providing 24V DC power supply for the robot;

[0178] (3) USB-485 communication converter. Connecting the monitoring computer and the 485 communication drivers in the robot;

[0179] (4) USB-CAN communication converter. Connecting the monitoring computer and the rotating motor of the robot;

[0180] (5) Cutting motor start-stop button.

[0181] The monitoring computer installs monitoring software, the interface of which is provided for the upper computer operator to use, to realize monitoring of the robot working environment and the robot body, the main components and functions of which are as follows:

[0182] (1) The interface includes buttons for hub motor forward and reverse rotation, acceleration and deceleration, and stop control, the background program includes command word generation programs corresponding to the buttons, and programs sent to the hub motor through 485 communication;

[0183] (2) The interface includes three types of push rods for pressing and lifting control, cutting arm pitching control, and cutting head pitching control, lifting control and reset control buttons, and corresponding command word generation programs. and programs sent to the hub motor through 485 communication;

[0184] (3) The interface includes rudder pitching control buttons, and corresponding command word generation programs. and programs sent to the hub motor through 485 communication;

[0185] (4) The interface includes cutting arm rotating motor roll control buttons, and corresponding command word generation programs. and programs sent to the hub motor through CAN communication;

[0186] (5) The interface includes two camera image display windows fed back by video lines on the robot;

[0187] (6) The interface includes a neural network-based sewer pipe defect detection algorithm that can automatically identify and locate pipe defects through the feedback pipe wall image, and display the defects to the operator through the interactive interface.

[0188] In an application example of the present application:

[0189] (1) The pipe clearing robot and related equipment are transported to the site, and the robot is prepared to be lowered into the well. Connect the pipe clearing robot cables to the traction steel wire rope, hoist the pipe clearing robot to the top of the vertical shaft through the lifting hook, release the electric hoist to vertically send the pipe clearing robot into the well, and release the cable car. Control the posture of the pipe clearing robot to place it horizontally and align it with the working pipe, and remove the lifting hook.

[0190] (2) Start the power supply of the pipe clearing robot, control the pipe clearing robot to climb into the pipe, and observe the front environment through the front camera to find foreign matter.

[0191] (3) When encountering foreign matter, the relative position of the cutter and the obstacle can be observed, the climbing or movement of the robot arm of the pipe clearing robot is controlled, the cutter is aligned with the foreign matter, and repeated cleaning is performed until the obstacle is completely removed.

[0192] (4) In order to ensure that the debris cleaned by the pipeline obstacle-removing robot is quickly and smoothly discharged, uninterrupted water flow flushing needs to be carried out downstream at the position of the terminal shaft well at the beginning of the pipeline obstacle-removing operation of the pipeline obstacle-removing robot.

[0193] (5) The pipeline obstacle-removing robot recovery program is started, and the return action is performed through the cooperation of the traction steel wire rope and the pipeline obstacle-removing robot climbing until the pipeline obstacle-removing robot completely exits the operation pipeline, and the return program is stopped.

[0194] (6) The hook is manually installed with the pipeline obstacle-removing robot, and the pipeline obstacle-removing robot is pulled out of the shaft well by the hoisting machine.

[0195] (7) The discharge condition of the cleaned debris at the pipeline outlet is observed, and if no debris is discharged for a long time, the flushing device is closed, and the current cleaning is completed.

[0196] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for detecting defects in drainage pipes, characterized in that, The process is performed by the image recognition system installed in the image processing unit of the pipeline clearing equipment. The pipeline clearing equipment includes a front-facing camera, a rear-facing camera, an image processing device, a control system, and a pipeline clearing robot. The pipeline clearing robot includes a robotic arm unit and a traveling unit arranged in a first direction. The traveling unit is used to travel in the first direction. The traveling unit includes a body. The robotic arm unit includes an action head and a robotic arm drive assembly. One end of the body in the first direction is a connecting end, which is connected to the robotic arm unit. The robotic arm drive assembly is used to drive the action head to rotate relative to the body. Both the image processing device and the pipeline clearing robot are communicatively connected to the control system. The image processing device is equipped with the image recognition system. The driving unit includes a lifting assembly mounted on top of the main body. Both the front camera and the rear camera are communicatively connected to the image processing device. The front camera is mounted on the robotic arm unit, and the rear camera is mounted on the top surface of the main body. The front camera is used to observe the pipeline environment and the operating status of the lifting head in front of the pipeline clearing robot, and the rear camera is used to observe the pipeline environment and the operating status of the lifting assembly behind the pipeline clearing robot. The method for detecting defects in drainage pipes includes: Receive the drainage pipe image data currently captured by the front or rear camera; The drainage pipe image data is input into a preset drainage pipe defect detection model so that the drainage pipe defect detection model outputs the corresponding drainage pipe defect identification result data. The defect identification result data of the drainage pipe is sent to the control system so that the user can view the defect identification result data of the drainage pipe through the control system and issue corresponding control commands to the pipe clearing robot through the control system. The drainage pipeline defect detection model is obtained by training a preset YOLOv5s optimized model with historical image data of drainage pipelines. The YOLOv5s optimized model is a neural network formed by adding FasterNet blocks, CA attention mechanism modules and CBAM attention mechanism modules to the YOLOv5 model.

2. The method for detecting defects in drainage pipes according to claim 1, characterized in that, The robotic arm unit further includes a primary swing arm and a secondary swing arm. One end of the secondary swing arm is a pivot end along its length, and the other end is a mounting end for mounting the actuator head. One end of the primary swing arm is pivotally connected to the connecting end along its length, and the other end is pivotally connected to the pivot end. The robotic arm drive assembly includes a primary drive component and a secondary drive component. The primary drive component drives the primary swing arm to swing relative to the body, and the secondary drive component drives the secondary swing arm to swing relative to the primary swing arm.

3. The method for detecting defects in drainage pipes according to claim 2, characterized in that, The pipeline clearing robot is configured as a strip extending in the first direction, or the pipeline clearing robot is configured as a wedge shape with the cross-sectional dimensions gradually decreasing from the driving unit to the robotic arm unit; the robotic arm drive assembly further includes a rotation drive component, which is installed at the connecting end, and the first-stage swing arm is connected to the connecting end through the rotation drive component, and the rotation drive component is used to drive the first-stage swing arm to rotate about a straight line parallel to the first direction as an axis.

4. The method for detecting defects in drainage pipes according to claim 3, characterized in that, The robotic arm unit further includes a connecting joint connected to the rotation drive component. The primary swing arm includes a first connecting plate and a second connecting plate, as well as two parallel support rods connected to each other via the first and second connecting plates. One end of each support rod is pivotally connected to the pivot end along its length, and the other end is pivotally connected to the connecting joint. The first connecting plate is positioned near the pivot end along the length of the support rod, and the second connecting plate is positioned near the connecting joint. Both the primary and secondary drive components are electric push rods. One end of the primary drive component is pivotally connected to the connecting joint along its length, and the other end is pivotally connected to the first connecting plate. One end of the secondary drive component is pivotally connected to the second connecting plate along its length, and the other end is pivotally connected to the secondary swing arm. The length directions of the primary and secondary drive components intersect.

5. The method for detecting defects in drainage pipes according to any one of claims 1-4, characterized in that, The traveling unit further includes a lifting assembly, which includes a lifting platform, a connecting rod, and a linear drive. The connecting rod and the lifting platform are both located above the main body. One end of the connecting rod is pivotally connected to the main body along its length, and the other end is pivotally connected to the lifting platform. One end of the linear drive is pivotally connected to the main body along its length, and the other end is pivotally connected to the connecting rod, so that the lifting platform moves upward relative to the main body as the linear drive extends, and moves downward relative to the main body as the linear drive shortens. The lifting platform is used to contact the inner wall of the pipe.

6. The method for detecting defects in drainage pipes according to claim 1, characterized in that, The control system includes a host computer and a monitor. The monitor is communicatively connected to the image processing device. The pipeline clearing equipment also includes a power supply system and a cable car. The coaxial cable connected to the monitor, the communication cable connected to the host computer, and the power supply cable connected to the power supply system converge at the cable car to form a main cable. The main cable is connected to the pipeline clearing robot.

7. The method for detecting defects in drainage pipes according to claim 6, characterized in that, Before inputting the drainage pipe image data into the preset drainage pipe defect detection model, the method further includes: Image datasets of drainage pipes from different sources are merged to obtain an initial dataset containing historical image data of each drainage pipe and corresponding labels. The labels include identifiers for representing different defect types of drainage pipes, including: no defects, buckling, cracks, debris, holes, joint misalignment, obstructions, utility intrusion, and tree roots. Data augmentation processing is performed on each of the historical image data of the drainage pipes in the initial dataset to expand the initial dataset and form a corresponding training dataset; The YOLOv5s optimized model is trained based on the training dataset to obtain a drainage pipe defect detection model for identifying the defect types corresponding to drainage pipe image data.

Images