Real-time attitude estimation method of underwater inertial navigation system

Abstract

The invention relates to the technical field of underwater navigation, and discloses a real-time attitude estimation method of an underwater inertial navigation system. The method comprises the steps that the underwater vehicle collects multi-axis angular velocity and linear acceleration data through an inertial measurement unit and uploads the data to a central processing module, and the central processing module conducts kinematics modeling on the data to generate attitude change initial feature vectors; receiving water flow velocity data of a Doppler velocimeter and pressure data of a depth sensor, and performing flow field interference compensation optimization on the attitude change initial feature vector based on the data to generate an anti-interference attitude feature vector; fusing the anti-interference attitude feature vector and a historical attitude estimation result to generate space-time fusion attitude state representation, and then resolving a real-time Euler angle through a nonlinear filter; and matching a corresponding attitude stability control instruction from the navigation strategy library according to the real-time Euler angle, and issuing the attitude stability control instruction to a propeller execution unit. The method can effectively cope with underwater complex environment interference and guarantee stable operation of the underwater vehicle.

Description

Technical Field

[0001] This invention relates to the field of underwater navigation technology, specifically to a real-time attitude estimation method for underwater inertial navigation systems. Background Technology

[0002] In the field of modern ocean exploration and development, underwater vehicles play a crucial role, widely used in many critical tasks such as deep-sea resource exploration, marine environmental monitoring, underwater infrastructure inspection, and military reconnaissance. The efficient and accurate execution of these tasks highly depends on the precise navigation and attitude control capabilities of underwater vehicles. The complex and variable underwater environment, characterized by strong corrosiveness, high pressure, low temperature, severely limited light, and complex current field interference, presents numerous severe challenges to the navigation and attitude estimation of underwater vehicles. Inertial navigation systems (INS) are based on Newton's laws of motion. Using accelerometers and gyroscopes in their inertial measurement units (IMUs), they measure an object's acceleration and angular velocity, thereby calculating its position, velocity, and attitude. In underwater environments, where traditional global navigation satellite system (GNSS) signals cannot effectively penetrate seawater, INS have become one of the primary navigation methods for underwater vehicles due to their strong autonomy, immunity to external electromagnetic interference, and ability to provide continuous real-time navigation information. Strapdown inertial navigation systems directly mount accelerometers and gyroscopes onto the vehicle, using a computer to calculate the attitude matrix in real time. This converts accelerometer information from the vehicle's coordinate system to the navigation coordinate system, thus completing the navigation calculation. However, INS are not without flaws. Their biggest problem is the accumulation of errors over time. This is because factors such as gyroscope drift and accelerometer zero-value bias cause the measurement data deviation to gradually increase over time, ultimately severely affecting navigation accuracy. For example, during long-duration underwater navigation missions, even if the initial attitude and position measurements are highly accurate, as time goes on, due to the accumulation of errors, the actual position of the vehicle may deviate significantly from the position calculated by the inertial navigation system, leading to deviations or even failures in mission execution. To overcome this limitation of inertial navigation systems, multi-sensor fusion methods are commonly employed. For example, inertial navigation systems are combined with acoustic positioning systems and visual navigation systems, aiming to improve navigation accuracy through the complementary advantages of different sensors. However, these methods still face numerous challenges in practical applications. While acoustic positioning systems can provide relatively accurate position information, they are significantly affected by factors such as underwater sound speed variations, obstacles, and noise interference, resulting in a limited positioning range. Furthermore, limited communication bandwidth and slow data transmission speed lead to a low update frequency of navigation information, making it difficult to reflect the real-time position and motion status of underwater vehicles in a timely manner. In underwater environments, visual navigation systems often suffer from severely compromised image quality due to poor lighting conditions and turbid water, leading to reduced accuracy in feature extraction and matching, thus impacting navigation accuracy.

[0003] Data fusion and processing from different sensors is challenging, requiring the development of efficient data fusion algorithms and processing technologies to address issues such as sensor registration, time synchronization, and information weight allocation. Summary of the Invention

[0004] The purpose of this invention is to provide a real-time attitude estimation method for underwater inertial navigation systems to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a real-time attitude estimation method for an underwater inertial navigation system, the method comprising: The underwater vehicle acquires multi-axis angular velocity and linear acceleration data through an inertial measurement unit and uploads the multi-axis angular velocity and linear acceleration data to a central processing module; the central processing module performs kinematic modeling on the multi-axis angular velocity and linear acceleration data to generate an initial feature vector for attitude change; The central processing module receives water flow velocity data from the Doppler velocimeter and pressure data from the depth sensor; and based on the water flow velocity data and pressure data, it performs flow field interference compensation optimization on the initial characteristic vector of the attitude change to generate an anti-interference attitude characteristic vector. The central processing module integrates the anti-interference attitude feature vector with the historical attitude estimation results to generate a spatiotemporal fusion attitude state representation; and calculates the real-time Euler angles based on the spatiotemporal fusion attitude state representation using a nonlinear filter. The central processing module matches the corresponding attitude stabilization control command from the navigation strategy library based on the real-time Euler angle, and sends the attitude stabilization control command to the thruster execution unit.

[0006] Preferably, the multi-axis angular velocity and linear acceleration data include the angular velocity of the three-axis gyroscope, the output value of the three-axis accelerometer, and the sampling timestamp, and the historical attitude estimation results include the historical average pitch angle, the trend of roll angle change, and the cumulative error of yaw angle.

[0007] Preferably, the central processing module performs kinematic modeling on the multi-axis angular velocity and linear acceleration data to generate an initial feature vector for attitude change, including: The quaternion differential equations are iteratively calculated on the angular velocities of each of the three-axis gyroscopes and the output values ​​of the three-axis accelerometers to obtain the quaternion update sequence; the sliding window covariance analysis is performed on the quaternion update sequence to extract the initial feature vector of the attitude change.

[0008] Preferably, the central processing module optimizes the initial attitude change feature vector based on the water flow velocity and pressure data to generate an anti-interference attitude feature vector, including: The water density distribution at the current depth is calculated based on the pressure data; a three-dimensional velocity field tensor model is constructed using the water velocity data; the initial feature vector of the attitude change is input into a torque compensator based on the fluid dynamics equations, and the fluid disturbance torque compensation value is calculated using the three-dimensional velocity field tensor model and the water density distribution; the fluid disturbance torque compensation value and the initial feature vector of the attitude change are fused to generate the anti-interference attitude feature vector.

[0009] Preferably, the central processing module fuses the anti-interference attitude feature vector with the historical attitude estimation results to generate a spatiotemporal fused attitude state representation, including: The historical attitude estimation results are subjected to Kalman gain weighting to generate a historical state confidence matrix; the anti-interference attitude feature vector and the historical state confidence matrix are cross-dimensionally correlated and mapped through a spatiotemporal convolution kernel to generate the spatiotemporal fusion attitude state representation.

[0010] Preferably, the central processing module calculates real-time Euler angles based on the spatiotemporal fusion attitude state representation using a nonlinear filter, including: The spatiotemporal fusion attitude state representation is input into the attitude solver based on the Lie group differential manifold; the current quaternion estimate is calculated using the tangent space projection of the rotation transformation matrix; and Euler angle transformation is performed based on the current quaternion estimate to obtain the real-time Euler angles.

[0011] Preferably, the central processing module matches the corresponding attitude stabilization control command from the navigation strategy library based on the real-time Euler angles, including: Identify the pitch angle deviation, roll angle oscillation frequency, and yaw angle offset in the real-time Euler angles; retrieve the optimal control response curve from the navigation strategy library based on the pitch angle deviation, roll angle oscillation frequency, and yaw angle offset; convert the optimal control response curve into a thruster power distribution command to generate the attitude stabilization control command.

[0012] Preferably, the anti-interference attitude feature vector and the historical state confidence matrix are correlated and mapped across dimensions using a spatiotemporal convolution kernel to generate the spatiotemporal fused attitude state representation, including: The historical state confidence matrix is ​​adaptively aligned using the parallel translation principle on the Riemannian manifold; a spatiotemporal tensor product space is constructed based on the aligned historical state confidence matrix and the anti-interference attitude feature vector; and a full-dimensional homogeneous mapping is performed in the spatiotemporal tensor product space to generate the spatiotemporal fused attitude state representation.

[0013] Preferably, the historical state confidence matrix is ​​subjected to curvature adaptive alignment using the parallel shift principle on the Riemannian manifold, including: Calculate the manifold connection coefficient based on the current motion acceleration; perform a tangent-cluster translation operation on the historical state confidence matrix using the manifold connection coefficient; optimize the alignment parameters based on the geodesic distance between the translated historical state confidence matrix and the anti-interference attitude feature vector to complete curvature adaptive alignment.

[0014] Preferably, the attitude solver based on the Lie group differential manifold further performs a dynamic reliability assessment after outputting real-time Euler angles, including: The covariance matrix of the attitude solution residual is calculated in real time, and the main diagonal elements of the covariance matrix are extracted to construct the residual feature vector; The residual feature vector is compared with a preset reliability threshold range. When the values ​​of at least two dimensions in the residual feature vector continuously exceed the reliability threshold range, an attitude calculation reliability decline alarm signal is generated. Based on the attitude calculation reliability degradation alarm signal, an adaptive adjustment command for the sampling frequency of the inertial measurement unit is triggered, and the adaptive adjustment command is sent to the sensor control unit to increase the acquisition density of multi-axis angular velocity and linear acceleration data.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This real-time attitude estimation method for an underwater inertial navigation system provides fundamental feature support for attitude estimation through the acquisition of multi-axis angular velocity and linear acceleration data and kinematic modeling. Using the data from the inertial measurement unit as the initial input can directly reflect the motion state of the vehicle, laying the data foundation for subsequent processing. By incorporating water velocity data from a Doppler velocimeter and pressure data from a depth sensor, the initial characteristic vector of attitude change is optimized for flow field interference compensation, significantly enhancing the system's adaptability to complex underwater flow fields. Flow field interference is a significant factor contributing to attitude estimation errors. Through the fusion and correction of external sensor data, the impact of water flow impact can be effectively offset, making the generated anti-interference attitude characteristic vector more closely resemble the actual motion state and reducing the error accumulation that occurs when relying solely on inertial data for calculation. By fusing anti-interference attitude feature vectors with historical attitude estimation results to generate a spatiotemporal fusion attitude state representation, the spatiotemporal correlation of data is fully utilized. Historical attitude information contains the vehicle's motion trend, and combining it with real-time anti-interference features enables a smooth transition in the time dimension, avoiding the impact of instantaneous data fluctuations on the estimation results. At the same time, it integrates information from different sources in the spatial dimension, improving the completeness and consistency of the attitude representation. Based on the spatiotemporal fusion attitude state representation, real-time Euler angles are calculated using a nonlinear filter, further improving the accuracy and stability of attitude estimation. The nonlinear filter can handle complex nonlinear motion models, effectively suppressing noise interference. Compared to traditional linear filtering methods, it is more adaptable to the changing motion states of underwater vehicles, ensuring that the calculated Euler angles accurately reflect the vehicle's attitude changes. Based on real-time Euler angles, the corresponding attitude stabilization control commands are matched from the navigation strategy library and issued for execution, realizing closed-loop linkage between attitude estimation and control. Real-time and accurate attitude information provides a reliable basis for the generation of control commands, ensuring that the thruster actuator can make timely adjustments, enabling the vehicle to maintain a stable attitude in complex underwater environments and enhancing the vehicle's mission execution capabilities. This method, through the synergistic effect of multi-source data fusion, flow field interference compensation, spatiotemporal feature integration, and nonlinear solution, forms a complete attitude estimation and control system. It can continuously output high-precision real-time attitude information in complex underwater environments, thereby improving the navigation performance and motion stability of underwater vehicles. Attached Figure Description

[0016] Figure 1 This is a schematic diagram illustrating the working principle of the real-time attitude estimation method for the underwater inertial navigation system described in this invention. Figure 2 A flowchart defining multi-axis data and historical attitude results; Figure 3 Flowchart for flow field disturbance compensation optimization; Figure 4 This is a flowchart for dynamic reliability assessment and adjustment. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1This invention provides a real-time attitude estimation method for an underwater inertial navigation system, the method comprising: The underwater vehicle acquires multi-axis angular velocity and linear acceleration data in real time via its inertial measurement unit (IMU) and transmits the data to the central processing module (CPU). The CPU performs kinematic modeling on the received data, generating an initial characteristic vector for attitude changes. Simultaneously, the CPU receives water flow velocity data from a Doppler velocimeter and pressure data from a depth sensor. Based on this data, it optimizes the initial characteristic vector for attitude changes by compensating for flow field disturbances, generating an anti-interference attitude characteristic vector. Subsequently, the CPU fuses the anti-interference attitude characteristic vector with historical attitude estimation results to generate a spatiotemporal fusion attitude state representation and calculates real-time Euler angles using a nonlinear filter. Finally, based on the real-time Euler angles, the CPU matches the corresponding attitude stabilization control command from the navigation strategy library and sends it to the thruster actuator to complete the attitude adjustment.

[0019] Example 1: See Figure 2 The underwater vehicle's inertial measurement unit (IMU) continuously collects angular velocities from the three-axis gyroscope and output values ​​from the three-axis accelerometer, recording the sampling timestamp for each data point. The three-axis gyroscope measures the vehicle's rotational angular velocities around the X, Y, and Z axes, while the three-axis accelerometer measures linear acceleration along the three axes. This data is transmitted to the central processing module via a high-speed data bus to ensure real-time attitude estimation. The central processing module preprocesses the received multi-axis angular velocity and linear acceleration data, including removing sensor noise, compensating for temperature drift, and aligning timestamps to ensure data consistency and accuracy.

[0020] Historical attitude estimation results include the historical mean pitch angle, the trend of roll angle variation, and the cumulative error of yaw angle. The historical mean pitch angle is calculated using a moving average to reflect the long-term attitude changes of the vehicle in the vertical direction. The trend of roll angle variation is extracted through differential operations to describe the dynamic characteristics of the vehicle in lateral sway. The cumulative error of yaw angle is estimated through integral operations to measure the cumulative deviation of the vehicle in heading control. This historical data is stored in the cache of the central processing module for subsequent attitude fusion calculations.

[0021] The central processing module performs iterative calculations using quaternion differential equations on the angular velocities of the three-axis gyroscopes and the output values ​​of the three-axis accelerometers. The quaternion differential equations update the quaternion states based on the current angular velocity and time step; the quaternions are used to describe the vehicle's three-dimensional rotational attitude. During the iterative calculations, the central processing module uses numerical integration to solve the differential equations, ensuring the continuity of attitude updates. The quaternion update sequence records the attitude changes of the vehicle at different time points, forming a continuous attitude evolution trajectory.

[0022] The central processing module performs sliding window covariance analysis on the quaternion update sequence. The sliding window covers quaternion data within a certain time range, and the window size is adaptively adjusted according to the dynamic characteristics of the aircraft. Covariance analysis calculates the statistical characteristics of quaternion changes within the window, including the amplitude of angular velocity fluctuations and the trend of acceleration changes. By analyzing the eigenvalues ​​of the covariance matrix, the central processing module extracts the dynamic features of attitude changes, forming an initial feature vector of attitude changes. This feature vector contains dynamic change information of the aircraft in the pitch, roll, and yaw dimensions, providing basic data for subsequent flow field disturbance compensation.

[0023] The initial feature vector of attitude change is further fused with historical attitude estimation results. The central processing module uses a weighted averaging method to combine the current attitude change features with historical attitude data to improve the stability of attitude estimation. The fused data is used to generate a spatiotemporal fused attitude state representation, providing input for subsequent nonlinear filter calculation of real-time Euler angles. The entire process adopts a modular design to ensure the independence and scalability of each computational stage, adapting to the application requirements of different underwater environments.

[0024] The central processing module monitors data quality in real time during the calculation process. If abnormal sensor data or calculation errors exceeding a preset threshold are detected, a data resampling or algorithm parameter adjustment mechanism is triggered. This adaptive adjustment mechanism can effectively cope with dynamic changes in the underwater environment and improve the robustness of attitude estimation.

[0025] Example 2: See Figure 3 The central processing module receives pressure data from depth sensors, reflecting the real-time water depth information of the vehicle. Based on the pressure-depth conversion relationship, the central processing module calculates the water density distribution at the current depth layer. Water density exhibits a non-linear variation with depth; the central processing module employs a piecewise linear interpolation method to construct a density distribution model that accurately reflects the gradient changes in water density at different depths. Simultaneously, water velocity data collected by the Doppler velocimeter is transmitted to the central processing module. This velocity data includes water velocity components in different directions within three-dimensional space. The central processing module performs coordinate transformation and normalization on the velocity data to eliminate measurement biases caused by sensor installation positions and constructs a three-dimensional velocity field tensor model. This tensor model comprehensively describes the spatial distribution characteristics of velocity in the current aquatic environment, providing fundamental data for subsequent flow field interference analysis.

[0026] The initial characteristic vector of attitude change is input into a torque compensator based on fluid dynamics equations. The torque compensator employs computational fluid dynamics methods, simplifying the vehicle's shape into an equivalent fluid dynamics model. The central processing module calculates the water pressure distribution at various points on the vehicle's surface based on a three-dimensional velocity field tensor model, and, combined with water density distribution data, derives the dynamic disturbance torque generated by the flow field on the vehicle. The calculation considers the coupling effect between the vehicle's motion and the water flow, including the influence of complex fluid phenomena such as boundary layer separation and eddy generation. The torque compensator outputs a fluid disturbance torque compensation value, which includes torque components in three axes, corresponding to disturbance compensation in the pitch, roll, and yaw directions, respectively.

[0027] The central processing module fuses the fluid disturbance torque compensation value with the initial attitude change feature vector. The fusion process employs a weighted superposition method, dynamically adjusting the compensation weights based on the fluid disturbance intensity under different depth and flow velocity conditions. The compensated data forms an anti-interference attitude feature vector, which effectively suppresses the impact of flow field disturbances on attitude estimation. This anti-interference attitude feature vector is further fused with historical attitude estimation results. The historical data undergoes Kalman gain weighting to generate a historical state confidence matrix. The Kalman gain is adaptively adjusted based on sensor measurement noise and system process noise to ensure the accuracy of historical data fusion.

[0028] The anti-interference attitude feature vector and the historical state confidence matrix are mapped across dimensions using a spatiotemporal convolution kernel. The spatiotemporal convolution kernel employs a three-dimensional convolution structure, establishing feature associations in both time and space dimensions. During convolution operations, the central processing module dynamically adjusts the convolution kernel parameters based on the vehicle's motion state to adapt to attitude change characteristics under different operating conditions. The convolution result generates a spatiotemporally fused attitude state representation, which integrates the current anti-interference features and historical attitude evolution trends to form a complete description of the vehicle's motion state. The entire computation process employs a parallel processing architecture to ensure that complex data fusion operations are completed within a limited time window.

[0029] The central processing module monitors changes in the flow field environment in real time. When a significant change in water density or flow velocity is detected, it automatically triggers a model parameter update mechanism. This mechanism adjusts the computational accuracy and sampling frequency of the fluid dynamics model to ensure accurate disturbance compensation under different water conditions. Simultaneously, the system dynamically verifies the output of the torque compensator. When the compensation value exceeds a reasonable range, an anomaly handling process is initiated to prevent erroneous compensation data from affecting the attitude estimation results. The resulting anti-interference attitude feature vector and spatiotemporally fused attitude state representation provide high-precision input data for subsequent attitude calculation and control command generation.

[0030] A layered computing architecture is adopted, with the bottom layer processing raw sensor data, the middle layer performing flow field modeling and disturbance compensation, and the top layer completing spatiotemporal fusion and state estimation. The layers communicate through standardized data interfaces, ensuring the system's modularity and scalability. The central processing module incorporates a multi-level caching mechanism to optimize data access efficiency and meet the time constraints of real-time computing. Key intermediate variables are continuously recorded during system operation for subsequent algorithm optimization and performance analysis, without affecting the real-time attitude estimation calculation process. This design enables the vehicle to maintain stable and reliable attitude estimation performance in complex and variable underwater environments.

[0031] Example 3: The spatiotemporal fused attitude state representation is fed into an attitude solver based on a Lie group differential manifold for processing. This solver uses the mathematical representation of the special orthogonal group SO(3) to describe the three-dimensional rotational motion of the vehicle. During the solution process, the attitude solver first maps the spatiotemporal fused attitude state representation to the tangent space, and establishes the differential equation of the rotational motion through the transformation relationship between Lie algebras and Lie groups. The rotation matrix... The differential relationship that evolves over time can be expressed as: In the formula, Denotes the rotation matrix at time t. Let be the angular velocity vector at the current moment. This represents the operator that converts a three-dimensional vector into a skew-symmetric matrix. The differential equation describes how the rotation matrix changes over time and is the core mathematical foundation for attitude calculation.

[0032] The attitude solver uses numerical integration to solve the differential equation, ensuring computational accuracy through the fourth-order Runge-Kutta method. In each integration step, the system predicts the quaternion representation for the next moment based on the current angular velocity measurement and the attitude state from the previous moment. During the quaternion update process, the solver performs normalization to prevent quaternion modulus drift caused by numerical calculations. After the quaternion update is complete, the system calculates the vehicle's current pitch angle θ, roll angle φ, and yaw angle ψ using a standard quaternion-Euler angle conversion relationship.

[0033] The central processing module analyzes and processes the calculated real-time Euler angles. The system first establishes an Euler angle deviation model to identify the difference between the current attitude and the desired attitude. Pitch angle deviation is obtained by comparing the current pitch angle with the target pitch angle, reflecting the degree of attitude deviation of the vehicle in the longitudinal direction. The roll angle oscillation frequency is extracted through fast Fourier transform analysis, characterizing the vehicle's swaying characteristics in the lateral direction. Yaw angle offset is calculated through integral operations, measuring the cumulative error of the vehicle in heading control.

[0034] The navigation strategy library stores a large number of pre-generated control response curves, each corresponding to specific aircraft dynamic characteristics. The central processing module uses a nearest neighbor search algorithm to find the control response curve in the strategy library that best matches the current attitude characteristics. During the search, the system comprehensively considers the similarity of three dimensions: pitch angle deviation, roll angle oscillation frequency, and yaw angle offset, ensuring that the selected control strategy can comprehensively cope with the current attitude anomalies. After matching, the system converts the optimal control response curve into a thruster power allocation scheme.

[0035] The generation of thruster power allocation commands employs a distributed optimization method. The central processing module establishes a thrust allocation matrix based on the vehicle's dynamic model, converting the desired control torque into specific thrust commands for each thruster. Thrust allocation considers factors such as thruster placement, maximum thrust limits, and energy consumption optimization, using quadratic programming to solve for the optimal thrust combination. The generated thrust commands include the target rotational speed and duration of action for each thruster, ensuring the vehicle can smoothly and efficiently adjust to the desired attitude.

[0036] The entire attitude calculation and control command generation process adopts a hierarchical architecture. The bottom layer is responsible for sensor data acquisition and preprocessing, the middle layer completes attitude calculation and state analysis, and the top layer implements control strategy matching and execution. Each layer communicates through standardized data interfaces to ensure the system's modularity and scalability. The central processing module has a built-in real-time monitoring mechanism to continuously track the convergence of attitude calculation and the execution effect of control commands. When abnormal calculation results or poor control performance are detected, the system automatically triggers a recalculation mechanism to adjust calculation parameters or rematch the control strategy.

[0037] During system operation, the central processing module records key intermediate variables and status information. This data includes, but is not limited to: raw sensor readings, fused attitude state, quaternion sequences generated during the calculation process, Euler angle variation curves, matched control strategy parameters, and thrust command execution status. This data is stored in a circular buffer, providing a basis for system debugging and performance analysis without affecting the processing efficiency of real-time control. The system employs an adaptive sampling strategy, dynamically adjusting the data recording frequency based on the vehicle's motion state, optimizing storage space utilization while ensuring information integrity.

[0038] The attitude solver is implemented with full consideration of computational efficiency and numerical stability. The algorithm employs optimization techniques such as fixed-point arithmetic and table lookup to improve execution speed while ensuring computational accuracy. For critical computational steps, such as quaternion updates and Euler angle transformations, the system incorporates numerical validity checks to prevent physically meaningless results. Furthermore, the solver includes multiple built-in exception handling routines to effectively handle anomalies such as sensor data loss and numerical overflow, ensuring stable system operation under various conditions.

[0039] The issuance and execution of control commands employ a time-triggered mechanism. The central processing module ensures that control commands are delivered to each thruster execution unit on time and in an orderly manner through precise clock synchronization. Each thruster is equipped with an independent controller responsible for receiving central commands and converting them into specific motor drive signals. The thruster controller incorporates a closed-loop control algorithm to adjust the motor speed in real time to achieve the target thrust requirements. A bidirectional communication link is established between the central processing module and the thruster controller, ensuring reliable transmission of control commands and enabling real-time acquisition of thruster execution status information.

[0040] The system employs a multi-level fault-tolerant design to enhance reliability. At the hardware level, key sensors and actuators are redundantly configured; at the software level, the algorithm implementation includes mechanisms for data verification, timeout detection, and fault recovery. When a sensor or thruster malfunctions, the system can automatically switch to backup equipment or adjust the control strategy, ensuring the vehicle maintains basic attitude control capabilities even with partial equipment failure. This fault-tolerant design significantly improves the system's survivability and mission completion rate in complex underwater environments.

[0041] Example 4: During the underwater vehicle's dive mission, the fusion of the historical state confidence matrix and the anti-interference attitude feature vector needs to fully consider the nonlinear characteristics of the motion state. Taking the actual operation of a certain type of underwater vehicle performing an observation mission at a depth of 300 meters as an example, the central processing module recorded the motion state parameters as shown in Table 1: Table 1: Examples of motion state parameters for underwater vehicles.

[0042]

[0043] The central processing module first calculates the manifold communication coefficient based on the current acceleration data. When the vehicle makes a depth adjustment with an acceleration of 0.2 m / s², the system detects a significant change in its motion state. The calculation of the communication coefficient comprehensively considers the vehicle's instantaneous acceleration, rate of change of angular velocity, and curvature characteristics of its historical trajectory. During the calculation, the system employs an adaptive sliding window technique, dynamically adjusting the window size according to the intensity of the motion. A larger window is used during stable motion to ensure stability, while a smaller window is switched to improve response speed during maneuvers.

[0044] The tangent bundle translation operation of the historical state confidence matrix is ​​specifically manifested as a dynamic adjustment of attitude parameters. Taking the data at 1024.5 seconds in the aforementioned observation task as an example, the system detected a reversal in the roll angle change trend and immediately initiated a translation compensation mechanism. The translation amount is calculated based on the difference between the current motion acceleration and the historical motion pattern, and the compensated roll angle data maintains better geometric consistency with the current manifold structure. During the translation process, the system monitors the constraint relationships between various attitude parameters in real time to prevent data distortions that violate kinematic laws.

[0045] Geodesic distance optimization employs an iterative approximation method. During system initialization, default alignment parameters are set. In actual operation, these parameters are progressively optimized by continuously observing the matching degree between historical data and real-time features. Taking yaw angle data as an example, when geomagnetic anomalies are detected affecting the magnetic compass, the system automatically increases the weighting coefficient of historical data, utilizing long-term heading stability to suppress short-term measurement noise. The optimization process uses damped least squares to ensure convergence speed while avoiding parameter oscillations.

[0046] After curvature adaptive alignment, the system constructs a spatiotemporal tensor product space. The dimensional design of this space comprehensively considers the number of degrees of freedom and motion characteristics of the underwater vehicle. For a typical six-DOF underwater vehicle, the spatiotemporal tensor product space contains a combination of position, attitude, and their derivatives of various orders. The eigenvectors are arranged according to their physical meaning, with position-related parameters first, attitude-related parameters second, and motion derivatives arranged in ascending order. This structural design ensures that subsequent full-dimensional homogeneous mappings maintain a clear physical meaning.

[0047] The execution process of the full-dimensional homogeneous mapping adopts a hierarchical processing strategy. The primary mapping handles linear combinations of independent motion dimensions, the intermediate mapping establishes coupling relationships between different degrees of freedom, and the high-level mapping establishes dynamic correlations across time scales. Mapping parameters are initialized based on the vehicle's dynamic characteristics and continuously updated during operation through an online learning mechanism. The system maintains a dynamic parameter library, storing the optimal mapping parameter combinations for different operating conditions, and automatically switches to the most suitable parameter set when changes in environmental conditions are detected.

[0048] Taking a seabed topography tracking mission as an example, when the vehicle entered a complex topography area from a flat seabed area, the system detected a significant increase in the frequency of attitude changes. The central processing module immediately initiated a dynamic parameter adjustment process: first, it narrowed the sliding window for curvature calculation to improve sensitivity to rapid attitude changes; then, it adjusted the compensation coefficient for tangential translation to enhance adaptability to sudden movements; finally, it optimized the weight allocation of the full-dimensional mapping to strengthen the correlation between roll and pitch dimensions. These adjustments enabled the system to complete parameter optimization within 3 seconds, adapting promptly to changes in motion state caused by topographic variations.

[0049] The system employs multiple technologies to ensure computational efficiency. Historical state data is stored using a circular buffer structure, satisfying both the need for large-capacity data storage and maintaining high-speed data access. Core algorithms such as curvature calculation and tangent bundle translation utilize a parallel computing architecture, fully leveraging the computing power of multi-core processors. The construction of the spatiotemporal tensor product space employs an incremental update strategy, recalculating only the changed data dimensions. These optimizations enable the entire processing flow to be completed within 10 milliseconds, meeting the timing requirements of real-time control.

[0050] Anomaly handling mechanisms are crucial for ensuring the reliable operation of the system. When sensor data anomalies are detected or numerical instability occurs during calculations, the system immediately initiates emergency procedures. For brief data anomalies, a predictive compensation algorithm generates alternative data; for persistent system-level failures, the system switches to a degraded operation mode, using simplified algorithms to maintain basic functionality. The system has a built-in self-checking program that periodically verifies the operational status of core algorithms and automatically triggers recovery procedures when problems are detected. In the recent three months of sea trials, this mechanism successfully handled 17 sensor data anomaly events, ensuring the vehicle's continuous operational capability.

[0051] The system generates abundant intermediate data during operation, which is categorized and stored for analysis. Raw sensor data is stored in a high-precision cache, intermediate variables from the calculation process are stored in a performance optimization area, and the final output is written to a high-reliability storage area. Data management employs a tag-based indexing mechanism, with each data block accompanied by a complete timestamp and operating condition tag, supporting multi-dimensional data retrieval and analysis. The system automatically performs data integrity checks weekly to correct potential storage errors and maintain the health of the database.

[0052] The maintenance interface provides convenient management functions for the system. Through standard communication protocols, operators can monitor the system's operating status in real time, adjust key parameter thresholds, and download historical operating data. The interface design employs hierarchical permission management, with different levels of operators possessing corresponding functional permissions. All parameter modifications and system operations are recorded in detail, forming a complete operation log. In the most recent software upgrade, the maintenance interface added an operating status visualization function, which can intuitively display key information such as curvature changes and alignment effects.

[0053] Example 5: See Figure 4After outputting real-time Euler angles, the attitude solver initiates a dynamic reliability assessment process. This process continuously monitors the residual characteristics during the solution process and constructs a multi-dimensional quality evaluation system. The central processing module first extracts the residual sequence from the intermediate calculation results of the attitude solution. These residuals reflect the differences between the actual sensor measurements and the model predictions. After standardization, the eigenvalue distribution of the covariance matrix of the residual sequence is calculated. The magnitude and proportional relationship of the eigenvalues ​​reveal the distribution of solution errors in different degrees of freedom; larger eigenvalues ​​usually correspond to motion dimensions with significant errors.

[0054] Principal component analysis (PCA) was used to construct the residual eigenvectors. The central processing module selected the diagonal elements of the covariance matrix as the basic features, and added statistical measures such as kurtosis and skewness of the residuals to form an extended feature set. The dimensionality of the eigenvectors was designed to consider the physical characteristics of attitude calculation, with pitch, roll, and yaw each corresponding to an independent feature subspace. The eigenvector elements were normalized to eliminate scale differences caused by different physical dimensions, ensuring consistency in subsequent threshold comparisons.

[0055] The reliability threshold range is set based on empirical data accumulated over long-term system operation. The central processing module maintains a dynamically updated threshold database, storing normal error ranges under different operating conditions. The threshold range is divided into two levels: warning and alarm. The warning level indicates that the error has increased but is still within a controllable range, while the alarm level means that the error has approached the system's tolerance limit. When environmental conditions or navigation status change, the system automatically loads the corresponding threshold parameter set to ensure that the evaluation criteria match the current operating conditions.

[0056] The generation of reliability degradation alarm signals employs a multi-condition triggering mechanism. The system not only detects exceedances in individual feature dimensions but also focuses on coordinated anomalies across multiple related dimensions. For example, when pitch and roll errors simultaneously exceed warning thresholds, and yaw error shows a continuous increasing trend, the system determines that the attitude calculation reliability has decreased. The alarm signal includes detailed anomaly descriptions, indicating the main anomaly dimensions, error change trends, and possible influencing factors. The alarm signal is transmitted to the system monitoring center via a dedicated communication channel to avoid being overwhelmed by regular data streams.

[0057] The generation of sampling frequency adjustment commands follows a gradual adjustment strategy. The initial adjustment range is determined based on the degree of error exceeding limits, typically using a step size of 10% of the base sampling rate. The central processing module monitors the effect after adjustment; if the error trend is not effectively contained, the adjustment intensity is further increased, but not exceeding 50% of the base sampling rate. Frequency adjustment takes into account the actual limitations of the sensor hardware, avoiding exceeding its maximum operating frequency. The adjustment command includes a smooth transition requirement to prevent data discontinuity issues caused by sudden changes in the sampling rate.

[0058] After receiving adaptive adjustment commands, the sensor control unit performs multi-level parameter adjustments. At the hardware level, the clock frequency of the analog-to-digital converter is reconfigured; at the software level, the cutoff frequency of the digital filter and the size of the data buffer are adjusted. With the sampling rate increased, the system automatically compensates for the increased communication load, prioritizing the transmission bandwidth of critical attitude data. The sensor control unit feeds back the actual sampling parameters to the central processing module, forming a closed-loop adjustment circuit.

[0059] The dynamic reliability assessment system adopts a modular architecture. The core assessment algorithm is decoupled from specific sensor types, acquiring the required data through a standardized interface. This design allows the system to flexibly adapt to different types of inertial measurement units (IMUs), requiring only the interface module to be replaced without modifying the core algorithm. Assessment parameters are stored in a configurable rule base, supporting online updates and expansion. During system operation, the basis and results of assessment decisions are automatically recorded, forming a complete audit log.

[0060] Anomaly recovery mechanisms are a crucial component of reliability assessment systems. When error characteristics return to the normal range, the system does not immediately revert to the original sampling frequency but maintains the increased sampling rate for a period of observation. Once the error is confirmed to be stable within a safe range, the sampling frequency is gradually reduced in stages. This delayed recovery strategy effectively prevents frequent oscillations in the system under critical conditions. For anomalies lasting for extended periods, the system automatically initiates in-depth diagnostic procedures to analyze the root cause and generate maintenance recommendations.

[0061] The central processing module implements various auxiliary functions to enhance the practicality of the evaluation system. A historical error trend visualization tool helps operators intuitively understand system status changes. A multi-level classification storage mechanism for evaluation results optimizes data retrieval efficiency. Adaptive learning capabilities enable the system to progressively optimize threshold parameters, better adapting to the actual operating characteristics of specific aircraft. These auxiliary functions are provided through a standard human-machine interface, supporting remote access and operation.

[0062] The system employs an intelligent scheduling strategy for resource management during runtime. Dynamic reliability assessment tasks are given high priority to ensure their computational resource requirements are met promptly. When the processor load is high, the system automatically simplifies non-critical computational steps to guarantee the real-time performance of core assessment functions. Memory usage utilizes a flexible allocation mechanism, dynamically adjusting the cache size based on current task demands. This resource management approach enables the system to maintain stable performance output under limited hardware conditions.

[0063] Collaborative operation with other subsystems is key to achieving overall effectiveness. The dynamic reliability assessment system shares anomaly characteristic data with the fault diagnosis system, jointly refining the assessment of system status. When interacting with the control system, the assessment results are converted into appropriate control parameter adjustment suggestions. In conjunction with the data logging system, critical assessment data is marked as high-priority storage content. This deep cross-system integration ensures that the reliability assessment results produce practical effects.

[0064] The system design fully considers the specific needs of different application scenarios. During long-duration cruise missions, the evaluation strategy prioritizes stability, avoiding unnecessary sampling rate adjustments. When performing high-maneuverability missions, the evaluation criteria automatically increase sensitivity to quickly respond to attitude changes. In cases of sensor performance degradation, the evaluation system compensates for errors caused by hardware limitations. This scenario-adaptive capability enables the system to meet diverse mission requirements.

[0065] The maintenance interface provides comprehensive system management functions. Operators can query the current evaluation status, modify threshold parameters, and view historical records. The interface design employs hierarchical access control, with different levels of users possessing corresponding operational permissions. All configuration changes and system operations generate detailed logs, supporting post-event traceability and analysis. The maintenance interface supports remote access, facilitating status monitoring and parameter adjustment of deployed aircraft systems.

[0066] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0067] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for real-time attitude estimation of an underwater inertial navigation system, characterized in that, include: The underwater vehicle acquires multi-axis angular velocity and linear acceleration data through an inertial measurement unit and uploads the multi-axis angular velocity and linear acceleration data to a central processing module; the central processing module performs kinematic modeling on the multi-axis angular velocity and linear acceleration data to generate an initial feature vector for attitude change; The central processing module receives water flow velocity data from the Doppler velocimeter and pressure data from the depth sensor; and based on the water flow velocity data and pressure data, it performs flow field interference compensation optimization on the initial characteristic vector of the attitude change to generate an anti-interference attitude characteristic vector. The central processing module integrates the anti-interference attitude feature vector with the historical attitude estimation results to generate a spatiotemporal fusion attitude state representation; and calculates the real-time Euler angles based on the spatiotemporal fusion attitude state representation using a nonlinear filter. The central processing module matches the corresponding attitude stabilization control command from the navigation strategy library based on the real-time Euler angle, and sends the attitude stabilization control command to the thruster execution unit.

2. The real-time attitude estimation method for an underwater inertial navigation system according to claim 1, characterized in that, The multi-axis angular velocity and linear acceleration data include the angular velocity of the three-axis gyroscope, the output value of the three-axis accelerometer, and the sampling timestamp, and the historical attitude estimation results include the historical average pitch angle, the trend of roll angle change, and the cumulative error of yaw angle.

3. The real-time attitude estimation method for an underwater inertial navigation system according to claim 2, characterized in that, The central processing module performs kinematic modeling on the multi-axis angular velocity and linear acceleration data to generate an initial feature vector for attitude change, including: performing quaternion differential equation iterative calculation on the angular velocity of each of the three-axis gyroscopes and the output values ​​of the three-axis accelerometers to obtain a quaternion update sequence; and performing sliding window covariance analysis on the quaternion update sequence to extract the initial feature vector for attitude change.

4. The real-time attitude estimation method for an underwater inertial navigation system according to claim 3, characterized in that, The central processing module optimizes the initial characteristic vector of attitude change based on the water velocity and pressure data to generate an anti-interference attitude characteristic vector by performing flow field disturbance compensation optimization. This includes: calculating the water density distribution at the current depth based on the pressure data; constructing a three-dimensional velocity field tensor model using the water velocity data; inputting the initial characteristic vector of attitude change into a torque compensator based on the fluid dynamics equations, and calculating the fluid disturbance torque compensation value using the three-dimensional velocity field tensor model and the water density distribution; and fusing the fluid disturbance torque compensation value with the initial characteristic vector of attitude change to generate the anti-interference attitude characteristic vector.

5. The real-time attitude estimation method for an underwater inertial navigation system according to claim 4, characterized in that, The central processing module integrates the anti-interference attitude feature vector with the historical attitude estimation results to generate a spatiotemporal fusion attitude state representation, including: performing Kalman gain weighting on the historical attitude estimation results to generate a historical state confidence matrix; and performing cross-dimensional correlation mapping between the anti-interference attitude feature vector and the historical state confidence matrix through a spatiotemporal convolution kernel to generate the spatiotemporal fusion attitude state representation.

6. The real-time attitude estimation method for an underwater inertial navigation system according to claim 5, characterized in that, The central processing module calculates real-time Euler angles based on the spatiotemporal fusion attitude state representation using a nonlinear filter, including: inputting the spatiotemporal fusion attitude state representation into an attitude solver based on a Lie group differential manifold; calculating the current quaternion estimate using the tangent space projection of the rotation transformation matrix; and performing Euler angle transformation based on the current quaternion estimate to obtain the real-time Euler angles.

7. The real-time attitude estimation method for an underwater inertial navigation system according to claim 6, characterized in that, The central processing module matches the corresponding attitude stabilization control command from the navigation strategy library based on the real-time Euler angles, including: identifying the pitch angle deviation, roll angle oscillation frequency, and yaw angle offset in the real-time Euler angles; retrieving the optimal control response curve from the navigation strategy library based on the pitch angle deviation, roll angle oscillation frequency, and yaw angle offset; and converting the optimal control response curve into a thruster power distribution command to generate the attitude stabilization control command.

8. The real-time attitude estimation method for an underwater inertial navigation system according to claim 7, characterized in that, The process of generating the spatiotemporal fused attitude state representation by performing cross-dimensional correlation mapping between the anti-interference attitude feature vector and the historical state confidence matrix through a spatiotemporal convolution kernel includes: performing curvature adaptive alignment of the historical state confidence matrix using the parallel translation principle on the Riemannian manifold; constructing a spatiotemporal tensor product space based on the aligned historical state confidence matrix and the anti-interference attitude feature vector; and performing a full-dimensional homogeneous mapping in the spatiotemporal tensor product space to generate the spatiotemporal fused attitude state representation.

9. The real-time attitude estimation method for an underwater inertial navigation system according to claim 8, characterized in that, The curvature adaptive alignment of the historical state confidence matrix is ​​performed using the parallel translation principle on the Riemannian manifold, including: calculating the manifold connection coefficient based on the current motion acceleration; performing a tangent-cluster translation operation on the historical state confidence matrix using the manifold connection coefficient; and optimizing the alignment parameters based on the geodesic distance between the translated historical state confidence matrix and the anti-interference attitude feature vector to complete the curvature adaptive alignment.

10. The real-time attitude estimation method for an underwater inertial navigation system according to claim 6, characterized in that, The attitude solver based on the Lie group differential manifold, after outputting real-time Euler angles, further performs a dynamic reliability assessment, including: The covariance matrix of the attitude solution residual is calculated in real time, and the main diagonal elements of the covariance matrix are extracted to construct the residual feature vector; The residual feature vector is compared with a preset reliability threshold range. When the values ​​of at least two dimensions in the residual feature vector continuously exceed the reliability threshold range, an attitude calculation reliability decline alarm signal is generated. Based on the attitude calculation reliability degradation alarm signal, an adaptive adjustment command for the sampling frequency of the inertial measurement unit is triggered, and the adaptive adjustment command is sent to the sensor control unit to increase the acquisition density of multi-axis angular velocity and linear acceleration data.

Images