CN115871901B - An Imitation Sturgeon Robot and Submarine Cable Fault Detection Method - Google Patents

Abstract

The invention discloses a sturgeon-like robot and a submarine cable fault detection method, and relates to the technical field of marine cable detection. The novel fishing head comprises a head part, a trunk unit, a fish tail part and a control system, wherein the head part comprises a head shell, two pectoral fins are symmetrically arranged on two sides of the head shell, and a dorsal fin is arranged on the top of the head shell. The tail portion is coupled to the head housing by a plurality of torso units, including a tail housing and a tail fin. The trunk unit comprises a trunk shell and a steering driving mechanism, and the trunk units are connected end to end in sequence. The trunk units are matched with the fish heads to realize the integral swing of the sturgeon-like robot and provide the advancing power for the sturgeon-like robot. The control system comprises a controller, a sonar, an image acquisition module and a signal emission module, wherein the image acquisition module is arranged at the front end of the head shell. The invention adopts sonar to find the position of the submarine cable, and identifies the fault type of the submarine cable through image acquisition and convolutional neural network, so that the invention has the advantages of high searching speed and accurate position judgment, and is suitable for the submarine complex environment.

Description

A sturgeon-like robot and a submarine cable fault detection method

Technical Field

The present invention relates to the technical field of marine cable fault detection, and in particular to a sturgeon-like robot and a submarine cable fault detection method.

Background Art

Submarine cables refer to cables wrapped in insulating materials and laid on the seabed, mainly used for telecommunication transmission. Submarine cables are divided into underwater communication cables and power cables. Their application areas include international communications, power and communications between coastal islands and cities, offshore wind power generation, power and information transmission on offshore oil platforms, etc. Among them, underwater communication cables are the main mode of international information transmission, accounting for 97% of the world's data transmission; submarine cables are an important means of transmission of power and communications between coastal islands and cities. my country has vast sea areas, and telecommunication transmission between islands, between coastal cities, and between land and islands requires a huge amount of submarine cables; in the field of energy development, submarine cables have broad application prospects in offshore platforms and offshore wind power generation and transmission. Therefore, submarine cables are of great value to social life, economy, military and other aspects.

However, even if submarine cables are shielded and buried, more than 200 underwater cable failures still occur every year. Damage to submarine cables will not only cause significant economic losses, but may also lead to a series of problems in the resource extraction process, further causing marine environmental pollution. Damage to submarine cables mainly includes natural disasters and man-made damage. Natural disasters include submarine earthquakes, landslides, ocean currents and waves, tsunamis, huge waves, sea level rise, extreme weather (hurricanes) and volcanic activities. These natural disasters may eventually cause wear, stress fatigue and failure, and breakage of submarine cables. Man-made damage includes accidental man-made damage and intentional man-made damage. Accidental man-made threats include man-made behaviors that cause unexpected damage to cables, such as trawling or anchoring in fishing, and seabed operations such as clam dredgers and scallop dredgers are very likely to damage submarine cables. Intentional damage to cables is mainly reflected in the theft of submarine cables. The above man-made damages mainly cause damage and breakage of submarine cables.

In summary, there are many threats to submarine cables, which may cause different degrees of damage. In order to detect submarine cable faults in time and carry out repairs, a lot of related research has been carried out. From the perspective of detection technology, existing submarine cable detection can be divided into methods based on optics, electricity, acoustics, magnetism and multiple sensors. Visual detection based on optical detection has the characteristics of being insensitive to noise data and can obtain good environmental information, but it is easily affected by light and water quality; electrical detection does not require close-range detection, but requires loading electrical signals on the submarine cable, which is generally used to detect electrical faults, such as copper core short circuits and open circuits; acoustic detection technology has the characteristics of long detection distance, but is easily interfered by external noise, and its detection accuracy is limited; magnetic detection technology has the advantage of long detection distance, but active detection requires current injection into the cable, and passive detection is easily interfered by the surrounding magnetic field; multi-sensor fusion technology based on the combination of optical, electrical, acoustic, magnetic and other sensors has the characteristics of high reliability of detection results, but the data processing is complex, the calculation is large, and it takes a long time. From the above analysis, it can be seen that various submarine cable fault detection methods in the prior art are limited by corresponding drawbacks and have not yet been able to effectively solve the submarine cable detection problem in actual engineering. Therefore, the prior art needs to be further improved and enhanced.

Summary of the invention

In view of the above-mentioned deficiencies in the prior art, one object of the present invention is to propose a sturgeon-like robot to solve the problem that the submarine cable fault detection means in the prior art are affected by the submarine environment, resulting in corresponding disadvantages, low accuracy and reliability of the detection results, and inability to actually and effectively determine the fault point and fault type of the submarine cable.

In order to solve the above technical problems, the technical solution adopted by the present invention is:

A sturgeon-like robot comprises a fish head part, a trunk unit, a fish tail part and a control system. The fish head part comprises a sturgeon-like head shell, two pectoral fins are symmetrically arranged on the left and right sides of the head shell, and a dorsal fin is arranged on the top.

The fish tail part is located at the rear side of the fish head part, and is connected to the rear side of the head shell through a plurality of trunk units arranged in sequence. The fish tail part includes a sturgeon-like tail shell, and a tail fin is arranged at the rear end of the tail shell.

The trunk unit includes a trunk shell and a steering drive mechanism arranged inside the trunk shell. The trunk units are connected head to tail in sequence. The front end of the first trunk unit is fixedly connected to the rear end of the head shell, and the rear end of the last trunk unit is fixedly connected to the tail shell.

The trunk units cooperate with the fish head part to realize the overall swing of the sturgeon-like robot, providing the robot with driving force to move forward.

The control system includes a controller, a sonar, an image acquisition module and a signal transmission module. The image acquisition module is arranged at the front end of the head shell. The sonar, the image acquisition module and the signal transmission module are all connected to the controller signal.

Furthermore, the trunk shell is a section of annular shell, and there is a gap between any two adjacent trunk shells and between the first trunk shell and the head shell.

The steering drive mechanism includes a dual-axis servo motor and a steering bracket. The dual-axis servo motor is fixed inside the trunk shell. The steering bracket is located at the front side of the dual-axis servo motor and is fixedly connected to its output shaft. A connecting frame is fixed to the rear side of each dual-axis servo motor.

Furthermore, the front end of the steering bracket in the first order of the trunk unit is fixedly connected to the rear end of the head shell, and the front end of the steering bracket in the remaining orders of the trunk units is fixedly connected to the rear end of the connecting frame in the trunk unit adjacent to the front side.

The rear end of the connecting frame in the last trunk unit is fixedly connected to the front end of the tail shell.

Furthermore, a motor bracket is fixedly provided on the inner side of each of the trunk shells, and a dual-axis servo motor is fixed on the motor bracket, and its two output shafts are arranged vertically.

Two float blocks made of buoyancy material are symmetrically arranged on the left and right sides of the dual-axis servo motor, and the two float blocks are fixedly connected to the inner side wall of the trunk shell.

Furthermore, each trunk shell includes two shell monomers arranged opposite to each other, a hinge block is fixed to the upper end of each shell monomer, and the two hinge blocks of the same shell monomer are staggered and hinged by a rotating shaft.

Bolt blocks are fixed to adjacent sides of the bottoms of the two housing monomers, and the two bolt blocks at the bottom of the same housing monomer are opposite to each other and are fixedly connected by bolts.

Furthermore, the motor bracket includes two bracket monomers, the bracket monomer includes a T-shaped structure formed by a transverse rod and a longitudinal rod fixedly connected, and the two bracket monomers are arranged in an axisymmetric manner.

One end of the two cross bars that is away from each other is fixedly connected to the inner side wall of the trunk shell, and the other end is fixedly plugged with the corresponding bracket single body longitudinal rod to form a closed area for placing the dual-axis servo motor.

Furthermore, each of the connecting frames is provided with connecting terminals at the top and the bottom, and each connecting terminal is connected to the rear end of the head shell via a rib wire made of elastic material.

A side of each pectoral fin close to the head shell is fixedly connected with a transverse axis, and the transverse axis is rotationally sealed with the side wall of the head shell. One end of each transverse axis away from the pectoral fin is connected to a first servo motor.

The dorsal fin is arranged vertically, and a vertical shaft is fixedly connected to its bottom. The vertical shaft is rotatably sealed with the top wall of the head shell, and the lower end of the vertical shaft is connected to the output end of the second servo motor fixed in the head shell.

Another object of the present invention is to provide a method for detecting submarine cable faults using the sturgeon-like robot.

A method for detecting submarine cable faults, using the above-mentioned sturgeon-like robot, comprises the following steps:

Step 1: Establish a convolutional neural network prediction model. The convolutional neural network includes an input layer, a convolution layer, a pooling layer, a fully connected layer, and an output layer. After learning and training, a convolutional neural network mathematical model with the best performance for submarine cable fault detection is obtained.

Step 2: Place the sturgeon-like robot in water and dive into deep water. During the swimming process, the sonar sends a signal to detect the location of the submarine cable. After determining the location of the submarine cable, the sonar sends a signal to the controller, and the sturgeon-like robot reaches the location of the submarine cable.

Step three: The sturgeon-like robot swims along the extension direction of the submarine cable. The image acquisition module collects image information of the submarine cable in real time and sends it to the controller. The convolutional neural network is used to determine the status and fault type of the submarine cable and store the data information.

Step 4: After the sturgeon-like robot floats up to the surface, it sends the data information to the receiving terminal on the ship or on the ground through the signal transmission module, and locates the position where the signal is sent through the base station to determine the location coordinates of the fault point.

Step 5: Repeat the process from step 1 to step 4 to determine the fault points and fault types of the submarine cable in turn. The staff will formulate a fault handling plan based on the data information.

Furthermore, the establishment of the submarine cable detection prediction model includes the following steps:

S1: Collect images and perform horizontal flipping, random subtraction, scale transformation, rotation and image expansion of submarine cable data to establish a data set including normal submarine cables and those caused by external forces, suspension and electrical faults.

The convolutional neural network dataset is randomly and evenly divided into training subset, validation subset, and test subset. The training set is used to learn the model parameters of the convolutional neural network, the validation subset selects hyperparameters by evaluating network performance, and the test subset measures network performance through generalization error.

S2: Use the particle swarm optimization algorithm to train the hyperparameters of the convolutional neural network, namely the number of convolutional layers, convolution kernel size, and the number of convolution kernels, and optimize them.

S3: Use the optimized number of convolutional layers, convolution kernel size, and number of convolutions to train the convolutional neural network, and import the data into the input layer, pooling layer, and fully connected layer. The convolutional neural network can extract the trained fault features.

S4: Calculate the loss function after output, update the weights through the gradient descent method, and implement multiple iterations of the convolutional neural network to make the training model converge.

S5: After multiple iterations, the test set is used to judge the classification performance of the convolutional neural network based on particle swarm optimization.

S6: According to the performance evaluation results, adjust the parameters of the number of nodes, number of iterations, and loss function value in the convolutional neural network.

S7: Repeat the above steps to finally obtain the convolutional neural network mathematical model for submarine cable fault detection with the best performance.

Furthermore, the image acquisition module includes a camera equipped with a light source. The input layer in step one is a submarine cable image. The input layer expands the submarine cable data image and further performs an enhancement operation on the image.

The convolution layer convolves the feature vector with the convolution kernel and realizes seabed image feature extraction through activation function response.

The mathematical expression of the convolutional layer neuron is:

(1)

(1)

为卷积神经网络卷积层

的第

个通道的输出；

为卷积神经网络卷积层

的第

个通道的输出；

为用于计算第

个通道净激活的输入特征图子集；

为卷积运算符号；

为卷积层

输入向量

与神经元

连接的权重矩阵；

为卷积层

第

个特征图的偏差值，

为激活函数。in,

Convolutional layer of convolutional neural network

No.

The output of each channel;

Convolutional layer of convolutional neural network

No.

The output of each channel;

To calculate the

A subset of input feature maps with net activations of channels;

is the convolution operator symbol;

For the convolutional layer

Input Vector

With neurons

The weight matrix of the connection;

For the convolutional layer

No.

The deviation value of the feature map,

is the activation function.

The mathematical expression of the pooling layer is:

(2)

(2)

为卷积神经网络池化层

第

个通道的权重系数；

为池化函数；

为池化层

第

个通道的偏置项。in,

Pooling layer for convolutional neural network

No.

The weight coefficient of each channel;

is the pooling function;

For the pooling layer

No.

The offset term for each channel.

The mathematical expression of the fully connected layer is as follows:

(3)

(3)

为全连接层

的输出；

为全连接层

的网络权重系数；

为全连接层

的输出；

为全连接层

的偏置项。in,

Fully connected layer

Output:

Fully connected layer

The network weight coefficient of

Fully connected layer

Output:

Fully connected layer

The bias term.

The output layer gives specific classification results to determine the normal state of the submarine cable and the specific categories of failures caused by external forces, suspension, and electrical faults.

By adopting the above technical scheme, the beneficial technical effect of the present invention is: the sonar detection distance of the sturgeon-like robot of the present invention is long, the target distance is accurately determined, the position of the submarine cable is quickly found, and it reaches the vicinity of the submarine cable. The camera collects the image information of the submarine cable in real time, detects the condition of the submarine cable through the convolutional neural network and determines the fault point. After the sturgeon-like robot surfaces, it sends the data information to the receiving terminal of the ship or the ground, and the bottom base station determines the coordinates of the fault position through the signal source. The present invention has the advantages of fast fault point search speed and accurate position judgment. The convolutional neural network prediction model has strong generalization ability, can adapt to the complex environment of the seabed, has good recognition and classification capabilities, effectively identifies the fault type of the submarine cable, and provides an accurate basis for the formulation of fault handling solutions.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic structural diagram of a sturgeon-like robot according to the present invention.

FIG. 2 is a schematic structural diagram of the present invention in FIG. 1 after removing one side of the housing monomer.

FIG3 is a schematic structural diagram of the present invention in FIG1 after all the trunk shells are removed.

FIG. 4 is a schematic structural diagram of the present invention in FIG. 3 after all the float blocks are further removed.

FIG. 5 is a schematic structural diagram of the pectoral fin and related components of the present invention in FIG. 1 .

FIG. 6 is a partial schematic diagram of the present invention, showing the fishtail portion, the steering drive mechanisms and the ribs.

FIG. 7 is a schematic diagram of a certain part of the structure of the present invention in FIG. 1 , showing a trunk unit.

FIG8 is a schematic diagram of a portion of the structure in FIG7 , showing a trunk housing and a motor bracket.

FIG9 is a flow chart of submarine cable fault detection based on a convolutional neural network according to the present invention.

DETAILED DESCRIPTION

The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

In the description of the present invention, unless otherwise specified, "plurality" means two or more than two; the terms "upper", "lower", "left", "right", "inner", "outer", "front end", "rear end", "head", "tail" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the mechanism or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", "third" and the like are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "connected" and "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

Embodiment 1, in combination with Figures 1 to 8, a sturgeon-like robot comprises a fish head part, a trunk unit, a fish tail part and a control system, wherein the fish head part comprises a sturgeon-like head shell 1, two pectoral fins 11 are symmetrically arranged on the left and right sides of the head shell 1, and a dorsal fin 12 is arranged on the top.

Specifically, a side of each pectoral fin 11 close to the head shell 1 is fixedly connected to a transverse shaft 14 , and the transverse shaft 14 is rotationally sealed with a side wall of the head shell 1 . An end of each transverse shaft away from the pectoral fin 11 is connected to a first servo motor 13 .

The dorsal fin 12 is arranged vertically, and a vertical shaft is fixedly connected to its bottom. The vertical shaft is rotatably sealed with the top wall of the head shell 1 , and the lower end of the vertical shaft is connected to the output end of the second servo motor fixed in the head shell 1 .

The fish tail portion is located at the rear side of the fish head portion and is connected to the rear side of the head shell 1 through a plurality of trunk units arranged in sequence. The fish tail portion includes a sturgeon-shaped tail shell 2, and a tail fin 21 is provided at the rear end of the tail shell 2.

The trunk unit includes a trunk shell 3 and a steering drive mechanism 4 arranged inside the trunk shell 3. The trunk shell 3 is a section of annular shell. Each trunk shell 3 includes two shell monomers 31 arranged opposite to each other. A hinge block 32 is fixed to the upper end of each shell monomer 31. The two hinge blocks 32 of the same shell monomer 31 are staggered and hinged through a rotating shaft.

A bolt block 33 is fixed to one side adjacent to the bottom of the two housing monomers 31 , and the two bolt blocks 33 at the bottom of the same housing monomer 31 face each other and are fixedly connected by bolts.

There is a gap between any two adjacent trunk shells 3 and between the first trunk shell 3 and the head shell 1. The trunk units are connected end to end in sequence, the front end of the first trunk unit is fixedly connected to the rear end of the head shell 1, and the rear end of the last trunk unit is fixedly connected to the tail shell 2.

Specifically, the steering drive mechanism 4 includes a dual-axis servo motor 41 and a steering bracket 42. The dual-axis servo motor 41 is fixed inside the trunk shell 3. Specifically, a motor bracket 5 is fixed on the inner side of each trunk shell 3. The dual-axis servo motor 41 is fixed on the motor bracket 5, and its two output shafts are arranged vertically.

Preferably, the motor bracket 5 includes two bracket monomers 51, and the bracket monomers 51 include a T-shaped structure formed by a cross bar and a longitudinal bar fixedly connected, and the two bracket monomers 51 are arranged in an axisymmetric manner. The ends of the two cross bars that are away from each other are fixedly connected to the inner side wall of the trunk shell 3, and the other ends are fixedly plugged with the longitudinal bars of the corresponding bracket monomers 51, forming a closed area for placing the dual-axis servo motor 41.

Two float blocks 6 made of buoyancy material are symmetrically arranged on the left and right sides of the dual-axis servo motor 41 , and the two float blocks 6 are fixedly connected to the inner side wall of the trunk shell 3 .

The steering bracket 42 is located at the front side of the dual-axis servo motor 41 and is fixedly connected to its output shaft, and a connecting frame 7 is fixed to the rear side of each dual-axis servo motor 41. A connecting terminal 71 is provided at the top and bottom of each connecting frame 7, and each connecting terminal 71 is connected to the rear end of the head shell 1 through a rib wire 72 made of elastic material.

The front end of the steering bracket 42 in the first trunk unit is fixedly connected to the rear end of the head shell 1, and the front end of the steering bracket 42 in the remaining trunk units is fixedly connected to the rear end of the connecting bracket 7 in the trunk unit adjacent to the front side. The rear end of the connecting bracket 7 in the last trunk unit is fixedly connected to the front end of the tail shell 2. Each of the trunk units cooperates with the fish head part to realize the overall swing of the sturgeon-like robot and provide its forward power.

The control system includes a controller, a sonar, an image acquisition module and a signal transmission module. The image acquisition module is arranged at the front end of the head shell 1. The sonar, the image acquisition module and the signal transmission module are all connected to the controller signal. The sonar uses active sonar echo to detect the general position of the submarine cable. The instrument will send sound waves into the sea, and the time it takes for the echo to be transmitted back to the robot fish can be used to calculate the shape and position of the submarine cable under the sturgeon robot. The active sonar can accurately measure the target distance and can detect fixed targets.

The patent of the present invention is based on the principle of bionics, starting from the practical application of functional bionics in engineering, and is designed by imitating the sturgeon in nature. The appearance of the sturgeon in nature is imitated, and the head is pointed and slightly tilted, which is conducive to the obstacle crossing of the sturgeon-like robot; the upper end of the sturgeon tail is long and the lower end is short, so it can move close to the surface of the cable; the sturgeon-like tendon line 72 is similar to the dragon tendon of the sturgeon, so that the mechanical structure has a stronger connection mechanism. Active sonar is used to sense the position of the submarine cable at a long distance, and the eyes of the sturgeon are imitated by the camera at a close distance. For the image blur caused by uneven light on the seabed and turbid water, the convolutional neural network based on particle swarm optimization is used to realize the fault detection of the submarine cable. The submarine cable detection method adopts the bionic principle, effectively overcomes the shortcomings of the traditional method, and uses active sonar detection to imitate the sturgeon's snout, detect the position of the submarine cable and further use it for the detection of the type of submarine cable fault. Therefore, on the one hand, the active sonar reduces the influence of external noise on it in principle, and on the other hand, the position of the submarine cable can be determined through multiple detections, which reduces the requirements from the functional level of realization.

Embodiment 2: Submarine cable detection mainly relies on the data collected by the camera to carry out corresponding work.

There are many ways to classify submarine cables. Here, they are divided into three categories according to different fault handling methods for subsequent processing: (1) caused by external forces, such as the hanging of the ship anchor, the dragging of the fishing net, the bite of the fish, etc. This type of fault requires reconnection of the line; (2) suspended, with the scouring of the submarine undercurrent, the pipeline will be suspended, and this type of fault is solved by adding a casing; (3) electrical fault, the cable is not standardized when it leaves the factory and there are bulges, deformation, etc. This type of fault requires further inspection to avoid further damage. Therefore, the present invention mainly uses the sturgeon-like robot to determine whether the submarine cable belongs to the normal state and the category of external force, suspended, or electrical fault.

A method for detecting submarine cable faults, based on the above-mentioned sturgeon-like robot, comprises the following steps:

Step 1: Establish a convolutional neural network prediction model. The convolutional neural network includes an input layer, a convolution layer, a pooling layer, a fully connected layer, and an output layer. After learning and training, a convolutional neural network mathematical model with the best performance for submarine cable fault detection is obtained. In step 1, the submarine cable detection prediction model is established using the following steps:

S1: Collect images and perform horizontal flipping, random subtraction, scale transformation, rotation and image expansion of submarine cable data to establish a data set including normal submarine cables and those caused by external forces, suspension and electrical faults.

The convolutional neural network dataset is randomly and evenly divided into training subset, validation subset, and test subset. The training set is used to learn the model parameters of the convolutional neural network, the validation subset selects hyperparameters by evaluating network performance, and the test subset measures network performance through generalization error.

S2: Use the particle swarm optimization algorithm to train the hyperparameters of the convolutional neural network, namely the number of convolutional layers, convolution kernel size, and the number of convolution kernels, and optimize them.

S3: Use the optimized number of convolutional layers, convolution kernel size, and number of convolutions to train the convolutional neural network, and import the data into the input layer, pooling layer, and fully connected layer. The convolutional neural network can extract the trained fault features.

S4: Calculate the loss function after output, update the weights through the gradient descent method, and implement multiple iterations of the convolutional neural network to make the training model converge.

S5: After multiple iterations, the test set is used to judge the classification performance of the convolutional neural network based on particle swarm optimization.

S6: According to the performance evaluation results, adjust the parameters of the number of nodes, number of iterations, and loss function value in the convolutional neural network.

S7: Repeat the above steps to finally obtain the convolutional neural network mathematical model for submarine cable fault detection with the best performance.

Specifically, the image acquisition module includes a camera equipped with a light source. The input layer in step one is a submarine cable image. The input layer expands the submarine cable data image and further performs an enhancement operation on the image.

The convolution layer convolves the feature vector with the convolution kernel and realizes the seabed image feature extraction through the activation function response.

The mathematical expression of the convolutional layer neuron is:

(1)

(1)

为卷积神经网络卷积层

的第

个通道的输出；

为卷积神经网络卷积层

的第

个通道的输出；

为用于计算第

个通道净激活的输入特征图子集；

为卷积运算符号；

为卷积层

输入向量

与神经元

连接的权重矩阵；

为卷积层

第

个特征图的偏差值，

为激活函数。in,

Convolutional layer of convolutional neural network

No.

The output of each channel;

Convolutional layer of convolutional neural network

No.

The output of each channel;

To calculate the

A subset of input feature maps with net activations of channels;

is the convolution operator symbol;

For the convolutional layer

Input Vector

With neurons

The weight matrix of the connection;

For the convolutional layer

No.

The deviation value of the feature map,

is the activation function.

The mathematical expression of the pooling layer is:

(2)

(2)

为卷积神经网络池化层

第

个通道的权重系数；

为池化函数；

为池化层

第

个通道的偏置项。in,

Pooling layer for convolutional neural network

No.

The weight coefficient of each channel;

is the pooling function;

For the pooling layer

No.

The offset term for each channel.

The mathematical expression of the fully connected layer is as follows:

(3)

(3)

为全连接层

的输出；

为全连接层

的网络权重系数；

为全连接层

的输出；

为全连接层

的偏置项。in,

Fully connected layer

Output:

Fully connected layer

The network weight coefficient of

Fully connected layer

Output:

Fully connected layer

The bias term.

The output layer gives specific classification results to determine the normal state of the submarine cable and the specific categories of failures caused by external forces, suspension, and electrical faults.

Step 2: Place the sturgeon-like robot in water and dive into deep water. During the swimming process, the sonar sends a signal to detect the location of the submarine cable. After determining the location of the submarine cable, the sonar sends a signal to the controller, and the sturgeon-like robot reaches the location of the submarine cable.

Step three: The sturgeon-like robot swims along the extension direction of the submarine cable. The image acquisition module collects image information of the submarine cable in real time and sends it to the controller. The convolutional neural network is used to determine the status and fault type of the submarine cable and store the data information.

Step 4: After the sturgeon robot floats up to the surface, it sends data information to the receiving terminal on the ship or on the ground through the signal transmission module, and locates the location of the signal through the base station to determine the location coordinates of the fault point. The fault location of the submarine cable is just below the position where the sturgeon robot floats up. The type and status of the fault can be known through the received data information, and a maintenance plan can be further formulated.

Step 5: Repeat the process from step 1 to step 4 to determine the fault points of the submarine cable and the fault type of each fault point in turn. The staff will formulate a fault handling plan based on the data information.

In the process of using convolutional neural networks to detect submarine cables, compared with traditional optical-based visual inspection, the deep learning algorithm based on convolutional neural networks has good model generalization ability, can adapt to the complex environment of the seabed, has good recognition and classification capabilities, and can effectively identify various faults of submarine cables.

Parts not described in the present invention can be implemented by adopting or drawing on existing technologies.

The embodiments of the present invention are given for the purpose of illustration and description, and are not intended to be exhaustive or to limit the invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present invention and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for specific uses.

Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the protection scope of the present invention.

Claims (7)

1. A method for detecting submarine cable faults, comprising a sturgeon-like robot, the sturgeon-like robot comprising a fish head portion, a trunk unit, a fish tail portion and a control system, wherein the fish head portion comprises a sturgeon-like head shell, two pectoral fins are symmetrically arranged on the left and right sides of the head shell, and a dorsal fin is arranged on the top; The fish tail part is located at the rear side of the fish head part, and is connected to the rear side of the head shell through a plurality of trunk units arranged in sequence. The fish tail part includes a sturgeon-like tail shell, and a tail fin is arranged at the rear end of the tail shell; The trunk unit includes a trunk shell and a steering drive mechanism arranged inside the trunk shell. The trunk units are connected head to tail in sequence. The front end of the first trunk unit is fixedly connected to the rear end of the head shell, and the rear end of the last trunk unit is fixedly connected to the tail shell. Each of the trunk units cooperates with the fish head part to realize the overall swing of the sturgeon-like robot, providing the robot with driving force to move forward; The control system includes a controller, a sonar, an image acquisition module and a signal transmission module. The image acquisition module is arranged at the front end of the head shell. The sonar, the image acquisition module and the signal transmission module are all connected to the controller signal. The method for detecting submarine cable faults comprises the following steps: Step 1: Establish a convolutional neural network prediction model. The convolutional neural network includes an input layer, a convolution layer, a pooling layer, a fully connected layer, and an output layer. After learning and training, a convolutional neural network mathematical model with the best performance for submarine cable fault detection is obtained. Step 2: Place the sturgeon-like robot in water and dive to a deep water area. During the swimming process, the sonar sends a signal to detect the location of the submarine cable. After determining the location of the submarine cable, the sonar sends a signal to the controller, and the sturgeon-like robot reaches the location of the submarine cable. Step 3: The sturgeon-like robot swims along the extension direction of the submarine cable. The image acquisition module collects image information of the submarine cable in real time and sends it to the controller. The state and fault type of the submarine cable are determined by the convolutional neural network, and the data information is stored. Step 4: After the sturgeon-like robot floats up to the surface, it sends the data information to the receiving terminal on the ship or on the ground through the signal transmission module, and locates the position where the signal is sent through the base station to determine the position coordinates of the fault point; Step 5: Repeat the process from step 1 to step 4 to determine the fault points and fault types of the submarine cable in turn. The staff will formulate a fault handling plan based on the data information. In step 1, the establishment of the submarine cable detection prediction model includes the following steps: S1: Collect images and perform horizontal flipping, random subtraction, scale transformation, rotation, and submarine cable data image expansion to establish a data set containing normal submarine cables and those caused by external forces, suspension, and electrical faults; The convolutional neural network data set is randomly divided into training subset, validation subset and test subset. The training subset is used to learn the model parameters of the convolutional neural network. The validation subset selects hyperparameters by evaluating the network performance. The test subset measures the network performance by generalization error. S2: Use the particle swarm optimization algorithm to train the hyperparameters of the convolutional neural network, namely the number of convolutional layers, convolution kernel size, and the number of convolution kernels, and optimize them; S3: Use the optimized number of convolutional layers, convolutional kernel size, and number of convolutional kernels to train the convolutional neural network, and import the data into the input layer, pooling layer, and fully connected layer. The convolutional neural network can extract the trained fault features; S4: Calculate the output loss function, update the weights by gradient descent method, and implement multiple iterations of the convolutional neural network to make the training model converge; S5: After multiple iterations, the test subset is used to judge the classification performance of the convolutional neural network based on particle swarm optimization; S6: According to the performance evaluation results, adjust the parameters of the number of nodes, number of iterations, and loss function value in the convolutional neural network; S7: Repeat the above steps to finally obtain the convolutional neural network mathematical model for submarine cable fault detection with the best performance. 2. The method for detecting submarine cable faults according to claim 1, characterized in that the trunk shell is a section of annular shell, and there is a gap between any two adjacent trunk shells and between the first trunk shell and the head shell; The steering drive mechanism includes a dual-axis servo motor and a steering bracket. The dual-axis servo motor is fixed inside the trunk shell. The steering bracket is located at the front side of the dual-axis servo motor and is fixedly connected to its output shaft. A connecting frame is fixed to the rear side of each dual-axis servo motor. 3. The method for detecting submarine cable faults according to claim 2, characterized in that the front end of the steering bracket in the first trunk unit is fixedly connected to the rear end of the head shell, and the front end of the steering bracket in the remaining trunk units is fixedly connected to the rear end of the connecting frame in the trunk unit adjacent to the front side; The rear end of the connecting frame in the last trunk unit is fixedly connected to the front end of the tail shell. 4. The submarine cable fault detection method according to claim 2 is characterized in that a motor bracket is fixedly provided on the inner side of each of the trunk shells, and a dual-axis servo motor is fixed on the motor bracket, and its two output shafts are arranged vertically; Two float blocks made of buoyancy material are symmetrically arranged on the left and right sides of the dual-axis servo motor, and the two float blocks are fixedly connected to the inner side wall of the trunk shell. 5. The method for detecting submarine cable faults according to claim 4, characterized in that each trunk shell comprises two shell monomers arranged opposite to each other, an articulation block is fixed to the upper end of each shell monomer, and the two articulation blocks of the same trunk shell are staggered and articulated by a rotating shaft; Bolt blocks are fixed to adjacent sides of the bottoms of the two shell monomers, and the two bolt blocks at the bottom of the same trunk shell are opposite to each other and are fixedly connected by bolts. 6. The method for detecting submarine cable faults according to claim 4, characterized in that the motor support comprises two support monomers, the support monomers comprise a T-shaped structure formed by a cross bar and a longitudinal bar fixedly connected, and the two support monomers are arranged in an axisymmetric manner; One end of the two cross bars that is away from each other is fixedly connected to the inner side wall of the trunk shell, and the other end is fixedly plugged with the corresponding bracket single body longitudinal rod to form a closed area for placing the dual-axis servo motor. 7. The method for detecting submarine cable faults according to claim 3, characterized in that each of the connecting frames is provided with connecting terminals at the top and bottom, and each connecting terminal is connected to the rear end of the head shell through a rib wire made of elastic material; A side of each of the pectoral fins close to the head shell is fixedly connected to a transverse axis, the transverse axis is rotationally sealed with the side wall of the head shell, and an end of each transverse axis away from the pectoral fin is connected to a first servo motor; The dorsal fin is arranged vertically, and a vertical shaft is fixedly connected to its bottom. The vertical shaft is rotatably sealed with the top wall of the head shell, and the lower end of the vertical shaft is connected to the output end of the second servo motor fixed in the head shell.

Images