Underwater robot depth and course control system

Abstract

The invention discloses an underwater robot depth and course control system, which relates to the technical field of underwater robots, and is characterized in that an equivalent strain energy anomaly index and a local maximum principal stress anomaly index are extracted through a structure-heat-fluid coupled three-dimensional finite element simulation model, and a failure risk prediction model is constructed by using a machine learning model; taking the comprehensive feature vector as input, and outputting a sealing structure failure prediction coefficient as a basis for dynamically adjusting a control strategy; the system limits or optimizes the propeller power, buoyancy adjusting frequency and course angle change rate in real time according to the failure prediction coefficient of the sealing structure so as to reduce the structural load and delay the fatigue process, a prediction model is continuously optimized through an error sample driven incremental learning mode, closed-loop self-learning and self-adaptive evolution of the control system are achieved, and the system reliability is improved. Therefore, the operation safety and the task execution stability of the underwater robot in the complex deep sea environment are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of underwater robots, specifically to a depth and heading control system for underwater robots. BACKGROUND

[0002] Depth and heading control of underwater robots refers to adjusting the vertical position (depth) and horizontal direction (heading) of the robot underwater through a control system to achieve stable operation and precise navigation. Depth control usually relies on sensors, buoyancy adjustment or thrusters to maintain the target depth, while heading control adjusts the direction of thrusters or rudders to make the robot move according to the predetermined path. These two control systems work together to ensure that underwater robots complete tasks such as exploration, search and rescue or data collection in complex environments.

[0003] The prior art has the following deficiencies:

[0004] In deep-sea high-pressure environments (such as water depths exceeding 3000 meters), the actuators of underwater vehicles (such as thrusters or servo rudders) face the technical risk of seal structure failure, mainly due to long-term pressure fatigue of seal materials, low-temperature embrittlement or mismatch of thermal expansion and contraction coefficients with the shell structure, resulting in micro-cracks or micro-permeation channels at the seal interface. In this case, seawater can penetrate into the internal drive mechanism, causing motor circuit short circuit, control abnormality, and even actuator failure, which in turn causes abnormal propulsion or attitude loss of control, and in severe cases, the entire machine cannot float, the task is interrupted, and the equipment is permanently lost. SUMMARY

[0005] The purpose of the present application is to provide a depth and heading control system for underwater robots to solve the deficiencies in the background art.

[0006] To achieve the above purpose, the present application provides the following technical solution: a depth and heading control system for underwater robots, comprising a data acquisition module, a calculation module, an adjustment control module and a feedback update module;

[0007] Data acquisition module: acquire multi-source environmental and operating state data from the robot body;

[0008] Calculation module: build a seal structure fatigue evolution model and a failure risk prediction model, combine finite element simulation and machine learning model, and calculate the seal structure failure prediction coefficient of the actuator;

[0009] Adjustment control module: dynamically adjust the control strategy according to the seal structure failure prediction coefficient, including limiting or optimizing the thruster power, buoyancy adjustment frequency, and heading angle change rate;

[0010] Feedback updating module: the optimized actual operation result is fed back to the model for analysis, so as to update and optimize the failure prediction mechanism, and realize closed-loop self-learning of the control strategy.

[0011] Preferably, the multi-source environment and operating state data include but are not limited to depth, temperature, water pressure, thruster load, motor current, motor cavity electrical insulation value, material strain and micro water seepage signal.

[0012] Preferably, the calculation of the sealing structure failure prediction coefficient comprises:

[0013] A three-dimensional finite element model containing structure-thermal-fluid coupling effects is established to model the sealing area of the underwater actuator, and the combined load effects of high-pressure water environment, temperature change and mechanical vibration are simulated;

[0014] In the simulation model, the crack initiation point and microstructure parameters are set, and the fatigue cycle loading method is used to simulate the stress distribution and crack propagation path of the sealing material under different time periods;

[0015] According to the simulation results, the key performance degradation characteristic values including the equivalent strain energy anomaly index and the local maximum principal stress anomaly index are extracted and input into the failure risk prediction model as training labels to improve the fitting accuracy of the machine learning model to the physical behavior.

[0016] Preferably, the equivalent strain energy anomaly index and the local maximum principal stress anomaly index are converted into a comprehensive feature vector, the comprehensive feature vector is taken as the input of the machine learning model, the machine learning model takes each group of comprehensive feature vectors as the prediction target to predict the sealing structure failure prediction coefficient label of the actuator, minimizes the sum of prediction errors of the sealing structure failure prediction coefficient labels of all actuators as the training target, trains the machine learning model until the sum of prediction errors converges, and stops the model training, determines the sealing structure failure prediction coefficient of the actuator according to the model output result, wherein the machine learning model is a long short-term memory network or an extreme gradient boosting algorithm.

[0017] Preferably, the method for obtaining the equivalent strain energy anomaly index EWEI is:

[0018] In the process of structural finite element simulation, the equivalent strain energy density distribution value W of the entire sealing structure under multi-condition loading cycle is obtained i (x,y,z,t);

[0019] The simulation obtained time series data is standardized, the average strain energy and the standard deviation σ W (x,y,z) of the target point in the time period T are calculated.

[0020] The equivalent strain energy abnormality index EWEI is calculated, and the expression is as follows: Wherein, W max represents the maximum strain energy in the period T.

[0021] Preferably, the method for obtaining the local maximum principal stress abnormality index is as follows:

[0022] For each grid unit i, the maximum principal stress time series vector is extracted: A sample set S is constructed The N points are regarded as the distribution in the S-dimensional space;

[0023] The number of neighbors k is set, and the k nearest neighbor set N k (i) of each point i is calculated; k (i,j) = max{k-dist(j), dist(i,j)}; wherein, k-dist(j) is the distance from point j to its kth nearest neighbor; the local reachable density of point i is defined as: The local outlier factor LOF k (i) of each point i is calculated, and the expression is as follows: It is defined as the local maximum principal stress abnormality index of point i.

[0024] Preferably, the obtained sealing structure failure prediction coefficient is compared with a gradient threshold value, the gradient threshold value includes a first threshold value and a second threshold value, and the first threshold value is smaller than the second threshold value; the sealing structure failure prediction coefficient is compared with the first threshold value and the second threshold value respectively;

[0025] If the sealing structure failure prediction coefficient is greater than the second threshold value, the active diving instruction is stopped, and the change rate of the heading angle is limited to be not greater than 5 degrees per second;

[0026] If the sealing structure failure prediction coefficient is greater than or equal to the first threshold value and less than or equal to the second threshold value, the output power of the propeller is reduced by not less than 20% to reduce the structural load;

[0027] If the sealing structure failure prediction coefficient is less than the first threshold value, the normal state is reached, and no adjustment is needed.

[0028] Preferably, the adjustment control module adjusts the key control parameters in real time according to the grade interval of the current sealing structure failure prediction coefficient FPC, and the adjustment adopts a linear limiting function as follows:

[0029] The propeller power limiting function is as follows: P adj = P nom ·(1-α·FPC); wherein, Padj represents the adjusted propeller power; P nom represents the original set propeller rated power; a represents the control sensitivity coefficient;

[0030] the buoyancy adjustment frequency control function, expressed as: f adj = f nom ·(1-β·FPC); in the formula, f adj represents the adjusted buoyancy adjustment frequency; f nom represents the buoyancy adjustment frequency of the system in the normal state; β represents the sensitivity coefficient of the buoyancy system to FPC;

[0031] the course angle change rate limiting function, expressed as: ω adj = ω max ·(1-γ·FPC); in the formula, ω max represents the upper limit of the real-time change rate of the course angle; ω adj is the adjusted course angle change rate, and γ represents the course adjustment sensitivity coefficient.

[0032] In the above technical solution, the technical effects and advantages provided by the present application are as follows:

[0033] 1. The present application significantly improves the stability and task execution efficiency of the underwater robot by combining the sealing structure failure prediction coefficient of the underwater robot actuator in the deep-sea high-pressure environment with the dynamic adjustment of the control strategy. The finite element simulation and machine learning model are combined to accurately calculate the sealing structure failure prediction coefficient of the actuator, to monitor and analyze the fatigue, temperature change and mechanical load of the sealing element in real time, and to continuously optimize the failure prediction mechanism through the feedback update module, so as to ensure that the robot can adaptively adjust the control strategy in extreme environments, effectively reduce the risk of structural damage, and prolong the service life of the equipment.

[0034] 2. The closed-loop self-learning mechanism of the present application can dynamically adjust the key control parameters such as propeller power, buoyancy adjustment frequency and course angle change rate according to the actual operation results, and when the risk of sealing structure failure increases, it can avoid fault expansion by limiting power output and course rate, etc., to ensure the smooth progress of the task. Through this accurate dynamic control method and real-time feedback optimization, the present application not only enhances the safety of the underwater robot in complex sea areas, but also improves the reliability and autonomy of task execution. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments or prior art of the present application, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art according to these drawings.

[0036] Figure 1 The system module mind map of the present application. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0038] Embodiment, please refer to Figure 1 As shown in the embodiment, the underwater robot depth and heading control system comprises a data acquisition module, a calculation module, an adjustment control module and a feedback updating module.

[0039] The data acquisition module acquires multi-source environment and running state data from the robot body.

[0040] The calculation module constructs a sealed structure fatigue evolution model and a failure risk prediction model, and calculates a sealing structure failure prediction coefficient of the actuator by combining finite element simulation and machine learning models.

[0041] The adjustment control module dynamically adjusts the control strategy according to the sealing structure failure prediction coefficient, including limiting or optimizing the propeller power, the buoyancy adjustment frequency, and the heading angle change rate.

[0042] The feedback updating module feeds back the optimized actual running result to the model for analysis, so as to update and optimize the failure prediction mechanism, and realize closed-loop self-learning of the control strategy.

[0043] In the embodiment of the present application, the underwater robot body integrates multiple types of sensors for real-time acquisition of multi-source environment and structure running state data, specifically including the following contents:

[0044] The underwater robot is provided with a high-precision pressure sensor (such as a MEMS type piezoresistive depth gauge), which is installed outside the sealed cavity and connected to the communication interface of the control system. The water pressure value output by the pressure sensor is converted into underwater depth data by a standard depth conversion formula, and the measurement accuracy is better than ±0.1m.

[0045] A distributed NTC thermistor array is arranged on the outer surface of the sealed cabin and the inside of the propeller cabin. This module is used to monitor the surrounding seawater temperature and the internal heating of the equipment. The temperature data sampling frequency is not less than 1Hz, which is used to judge the local hot spots, stress unevenness and material thermal fatigue risk.

[0046] Current, voltage and power detection circuits are integrated in the drive control module of each thruster to obtain the current consumption of the motor and the load variation trend in real time. The control system preliminarily judges the signs of mechanical overload or seal deterioration through load abnormalities and drive waveform distortion.

[0047] By setting an electrically insulating sensor (such as a dielectric strength probe or a leakage detection module) inside the sealed motor compartment, the conductivity change in the cavity is monitored in real time. Once micro-seepage or insulation degradation occurs in the motor compartment, the module will output an insulation drop signal and transmit it to the central control unit as an alarm signal.

[0048] MEMS strain gauges or fiber Bragg grating (FBG) sensors are arranged at key areas of the sealing structure (such as the sealing ring flange interface and the thruster bearing) to measure the material deformation caused by water pressure fluctuations, mechanical loads or temperature changes. This strain data will be an important input for the sealing fatigue evolution model.

[0049] A miniature conductivity sensor or a moisture-sensitive sensor is provided inside the robot to detect the change in electrical conductivity after a small amount of water molecules enters the sealed cavity. When the electrical conductivity exceeds the set threshold, the system can identify it as a potential sealing micro-leakage event.

[0050] All the above-mentioned sensors are connected to the control system through an embedded multi-channel acquisition module (supporting I 2 CAN and RS485 communication protocols), and the data is uniformly time-stamped and written into the state data buffer area at a frequency of 1-5 Hz, providing basic data support for subsequent sealing structure failure prediction and control optimization.

[0051] In this embodiment, a sealing structure failure prediction method for actuators based on the fusion of finite element simulation and machine learning models is provided, which is suitable for state health assessment of key components such as thrusters, servo steering machines and buoyancy adjustment devices of underwater robots in deep sea environment. This method realizes sealing structure fatigue evolution modeling, key feature extraction and risk prediction through the construction of a calculation module.

[0052] Firstly, a three-dimensional finite element simulation model including structure-thermal-fluid coupling effects is constructed, and the sealing area of the underwater actuator is modeled in detail; the load conditions include:

[0053] Static high pressure (simulating deep sea water pressure);

[0054] Periodic temperature fluctuations (simulating deep sea temperature difference or thermal load);

[0055] Mechanical excitation (simulating the vibration caused by repeated start-stop of thrusters).

[0056] The model material parameters are adopted with multi-level constitutive relations, including elastic modulus, Poisson's ratio, thermal expansion coefficient, S-N curve and crack propagation rate.

[0057] In the above simulation model, the initial crack position, micro-defect parameters and other initial states are set, the multi-cycle fatigue loading process is simulated, and the key physical quantities, including stress, strain energy density and their time-varying trends, are recorded at multiple grid nodes.

[0058] Two key performance degradation indicators are extracted from the simulation data, including equivalent strain energy anomaly index and local maximum principal stress anomaly index.

[0059] The equivalent strain energy anomaly index EWEI is obtained by:

[0060] The equivalent strain energy density distribution value W i (x,y,z,t) of the entire sealing structure in the multi-condition loading cycle is obtained in the structure finite element simulation process.

[0061] The time series data obtained by simulation are standardized, and the average strain energy and its standard deviation σ W (x,y,z) of the target point in the time period T are calculated.

[0062] The equivalent strain energy anomaly index EWEI is calculated, and the expression is: Where: W max represents the maximum strain energy in the cycle T. EWEI reflects the abnormal degree of energy concentration in the local area, and the larger the value is, the more dangerous the potential fatigue crack area is.

[0063] The method for obtaining the local maximum principal stress anomaly index MSPDI is:

[0064] For each grid element i, its maximum principal stress time series vector is extracted: A sample set is formed.

[0065] Set the number of neighbors k (for example, k = 10), and calculate the k nearest neighbor set N k (i) for each point i; for any two points i, j, define the reachability distance as: reach-dist k (i,j) = max{k-dist(j), dist(i,j)}; where k-dist(j) is the distance from point j to its kth nearest neighbor; define the local reachability density (LRD) of point i as: For each point i, calculate its local outlier factor LOF k (i), the expression is: Define the LOF value as the local maximum principal stress anomaly index of the point. The local maximum principal stress anomaly index reflects the degree of dispersion of the sealing structure region and the overall stress state, and an abnormally high value indicates a stress concentration risk point.

[0066] Convert the equivalent strain energy anomaly index and the local maximum principal stress anomaly index into a comprehensive feature vector, use the comprehensive feature vector as the input of a machine learning model, and use the machine learning model to predict the sealing structure failure prediction coefficient label of the actuator as the prediction target, minimize the sum of the prediction errors of the sealing structure failure prediction coefficient labels of all actuators as the training target, train the machine learning model until the sum of the prediction errors converges, and stop the model training. According to the model output result, determine the sealing structure failure prediction coefficient of the actuator, wherein the machine learning model is a long short-term memory network or an extreme gradient boosting algorithm.

[0067] Compare the obtained sealing structure failure prediction coefficient with a gradient threshold, the gradient threshold includes a first threshold and a second threshold, and the first threshold is less than the second threshold. Compare the sealing structure failure prediction coefficient with the first threshold and the second threshold respectively;

[0068] If the sealing structure failure prediction coefficient is greater than the second threshold (for example, 0.8), stop the active diving command, and limit the change rate of the heading angle to be not greater than 5 degrees per second;

[0069] If the sealing structure failure prediction coefficient is greater than or equal to the first threshold and less than or equal to the second threshold, reduce the propeller output power by not less than 20% to reduce the structural load;

[0070] If the sealing structure failure prediction coefficient is less than the first threshold (for example, 0.6), it is in a normal state at this time, and no adjustment is needed.

[0071] The adjustment control module adjusts the key control parameters in real time according to the grade interval of the current sealing structure failure prediction coefficient FPC. The adjustment adopts a linear limiting function as follows:

[0072] The propeller power limiting function is expressed as: P adj = P nom ·(1-α·FPC); in the formula, P adj represents the adjusted propeller power; P nom represents the original set propeller rated power; and a represents a control sensitivity coefficient (recommended value range: 0.5-0.8). When FPC approaches 1, the power is limited to 20-50% of the initial power, so as to reduce the propeller shaft load and structural fatigue.

[0073] The buoyancy adjustment frequency control function is expressed as: f adj = f nom ·(1-β·FPC); in the formula, f adj represents the adjusted buoyancy adjustment frequency (unit: Hz); f nom represents the buoyancy adjustment frequency of the system in the normal state; β represents the sensitivity coefficient of the buoyancy system to FPC (value: 0.3-0.6). The more frequent the buoyancy adjustment action is, the greater the pressure fluctuation of the cavity is, so the buoyancy change frequency should be actively reduced when FPC rises to reduce the sealing burden.

[0074] The heading angle change rate limiting function is expressed as: ω adj = ω max ·(1-γ·FPC); in the formula, ω max represents the upper limit of the real-time change rate of the heading angle (unit: ° / s); ω adj is the adjusted heading angle change rate, and γ represents the heading adjustment sensitivity coefficient (recommended value: 0.4-0.7). Rapid turning may cause propeller unbalanced load or structural lateral stress concentration, and limiting the heading change rate when FPC rises helps to maintain the sealing stability.

[0075] In the embodiment, in order to further improve the accuracy and robustness of the failure prediction of the sealing structure, and at the same time enhance the control strategy adaptation ability of the underwater robot in long-term complex tasks, a feedback updating module is introduced into the system to realize a closed-loop self-learning mechanism of the control strategy. The module is mainly used to dynamically correct the prediction model according to the actual operation results, thereby improving the intelligent level of the overall control system. The specific implementation steps of the module include the following contents:

[0076] After the system executes the control strategy based on the failure prediction coefficient, the system collects the actual operation state data and task environment data of the actuator in real time through the sensor module and the state recording unit, including but not limited to:

[0077] Propeller response curve (such as the actual deviation between speed, current and command signal);

[0078] Sealing structure strain change value (collected by strain gauge or fiber Bragg grating);

[0079] Micro water seepage sensor output signal (used to identify early leakage events of the seal);

[0080] FPC value change trajectory during control execution;

[0081] Final state label (such as whether the system enters the safety mode, whether a failure occurs or normal operation is maintained).

[0082] All collected data are time-stamped and stored synchronously for subsequent model feedback training.

[0083] The system compares the actual collected state data with the failure prediction coefficient (FPC) output by the original prediction model, calculates the difference, and defines it as error:

[0084] The error sample set includes error values, error trends over time, and control strategy parameters (thruster power, buoyancy frequency, heading angle change rate).

[0085] The sample set is stored in a structured manner and can be used as an incremental learning data source.

[0086] By extracting error characteristics, the system identifies the source of prediction bias and supports model adaptive correction.

[0087] The system inputs the error sample set to the machine learning training module and optimizes the failure prediction model using any of the following methods or combinations:

[0088] Incremental learning: fine-tune neural network weights based on the current model structure without overall retraining, suitable for online updates;

[0089] Transfer fine-tuning: preserve the historical model structure and master weights, and perform lightweight retraining using current task data;

[0090] Decision threshold adjustment: automatically optimize FPC warning threshold settings based on error trends to improve the robustness of control triggering.

[0091] The updated model is deployed back to the running control system for real-time FPC prediction and control strategy generation in the next cycle of tasks, forming a closed-loop optimization process of prediction → control → feedback → learning.

[0092] The present application provides an underwater robot control system based on the failure prediction coefficient of the sealing structure, which combines finite element simulation and machine learning algorithm, calculates and analyzes the failure risk of the sealing structure in real time, dynamically adjusts the control parameters such as thruster power, buoyancy adjustment frequency and heading angle change rate, to prolong the service life of the actuator and improve the system reliability. In addition, the system also introduces a feedback update module, which collects real-time running data and environmental state data, calculates the error and continuously optimizes the prediction model, forms a closed-loop self-learning mechanism, so that the control strategy can continuously adapt and optimize in long-time complex tasks, thereby realizing intelligent and precise control of underwater robots.

[0093] The above formulas are dimensionless values, and the formulas are obtained by software simulation of a large amount of data to obtain the most recent real situation, and the preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0094] The above-described embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented by software, the above-described embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center through wired (for example, infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0095] The above description is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. A depth and heading control system for an underwater robot, characterized in that: It includes a data acquisition module, a calculation module, an adjustment and control module, and a feedback and update module; Data acquisition module: Acquires multi-source environmental and operational status data from the robot itself; Calculation module: Constructs a fatigue evolution model and a failure risk prediction model for the sealing structure, and combines finite element simulation and machine learning models to calculate the failure prediction coefficients of the sealing structure of the actuator; Adjustment control module: Dynamically adjust the control strategy based on the failure prediction coefficient of the sealing structure, including limiting or optimizing the thruster power, buoyancy adjustment frequency, and heading angle change rate; Feedback Update Module: Feeds back the optimized actual operating results to the model for analysis, so as to update and optimize the failure prediction mechanism and realize closed-loop self-learning of the control strategy.

2. The underwater robot depth and heading control system according to claim 1, characterized in that: The multi-source environmental and operational status data includes, but is not limited to, depth, temperature, water pressure, thruster load, motor current, motor cavity electrical insulation value, material strain, and micro-seepage signals.

3. The underwater robot depth and heading control system according to claim 1, characterized in that: The calculation of the failure prediction coefficients for the sealing structure includes: A three-dimensional finite element model incorporating structural-thermal-fluid coupling effects was established to model the sealing region of the underwater actuator, simulating the combined load effects of high-pressure water environment, temperature changes, and mechanical vibration. In the simulation model, crack initiation point and microstructure parameters are set, and the fatigue cyclic loading method is used to simulate the stress distribution and crack propagation path of the sealing material under different time periods. Key performance degradation features, including the equivalent strain energy anomaly index and the local maximum principal stress anomaly index, are extracted from the simulation results and used as training labels to be input into the failure risk prediction model in order to improve the fitting accuracy of the machine learning model to the physical behavior.

4. The underwater robot depth and heading control system according to claim 3, characterized in that: The equivalent energy anomaly index and the local maximum principal stress anomaly index are converted into a comprehensive feature vector. The comprehensive feature vector is used as the input of the machine learning model. The machine learning model uses the prediction of the sealing structure failure prediction coefficient label of the actuator for each set of comprehensive feature vectors as the prediction objective, and minimizes the sum of prediction errors of the sealing structure failure prediction coefficient labels of all actuators as the training objective. The machine learning model is trained until the sum of prediction errors converges and the model training stops. The sealing structure failure prediction coefficient of the actuator is determined based on the model output. The machine learning model is a long short-term memory network or an extreme gradient boosting algorithm.

5. The underwater robot depth and heading control system according to claim 4, characterized in that: The method for obtaining the equivalent variable energy anomaly index is as follows: During the structural finite element simulation, the equivalent strain energy density distribution value W of the entire sealed structure under multiple loading conditions was obtained. i (x,y,z,t); The time-series data obtained from the simulation are standardized, and the average strain energy of the target point within time period T is calculated. and its standard deviation σ W (x,y,z); The equivalent variable energy anomaly index (EWEI) is calculated using the following expression: Among them: W max This represents the maximum strain energy within a period T.

6. The underwater robot depth and heading control system according to claim 5, characterized in that: The method for obtaining the local maximum principal stress anomaly index is as follows: For each mesh element i, extract its maximum principal stress time vector: Construct a sample set Consider it as the distribution of N points in an S-dimensional space; Given the number of nearest neighbors k, calculate the set N of its k nearest neighbors for each point i. k (i); For any two points i and j, the reachable distance is defined as: reach-dist k (i,j)=max{k-dist(j),dist(i,j)}; where k-dist(j) is the distance from point j to its k-th nearest neighbor; the local reachability density of point i is defined as: For each point i, calculate its Local Outlier Factor (LOF). k (i), the expression is: It is defined as the local maximum principal stress anomaly index at point i.

7. The underwater robot depth and heading control system according to claim 6, characterized in that: The obtained failure prediction coefficients of the sealing structure are compared with the gradient thresholds, which include a first threshold and a second threshold, and the first threshold is less than the second threshold. The failure prediction coefficients of the sealing structure are compared with the first threshold and the second threshold respectively. If the failure prediction coefficient of the sealing structure is greater than the second threshold, the active diving command is stopped and the rate of change of the heading angle is limited to no more than 5 degrees per second. If the failure prediction coefficient of the sealed structure is greater than or equal to the first threshold and less than or equal to the second threshold, the thruster output power shall be reduced by no less than 20% to reduce the structural load. If the failure prediction coefficient of the sealing structure is less than the first threshold, it is in a normal state and no adjustment is needed.

8. The underwater robot depth and heading control system according to claim 7, characterized in that: The control module adjusts key control parameters in real time based on the current failure prediction coefficient (FPC) level range of the sealing structure. The adjustment uses the following linear constraint function: The thruster power limiting function is expressed as: P adj =P nom ·(1-α·FPC); where P adj P represents the adjusted thruster power; nom This represents the original rated power of the thruster; α represents the control sensitivity coefficient. The buoyancy adjustment frequency control function is expressed as: f adj =f nom ·(1-β·FPC); where f adj Indicates the adjusted buoyancy adjustment frequency; f nom β represents the buoyancy adjustment frequency of the system under normal conditions; β represents the sensitivity coefficient of the buoyancy system to the FPC. The rate-of-change function for the heading angle is expressed as: ω adj =ω max ·(1-γ·FPC); where ω max This indicates the upper limit of the real-time rate of change of the heading angle; ω adj The adjusted rate of change of heading angle is represented by γ, which is the heading adjustment sensitivity coefficient.

Images