A method and system for rapid determination of attitude quaternions for deformation monitoring of hydraulic structures
By constructing a rapid attitude quaternion determination system, the problems of thermal inertia time scale mismatch and manifold destruction in attitude monitoring of large hydraulic structures were solved, realizing the physical rationality of attitude calculation and real-time early warning response.
Patent Information
- Application Number
- CN202611262076.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-19
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies for attitude monitoring of large hydraulic structures suffer from problems such as thermal inertia timescale mismatch, Euclidean space linear operations disrupting manifold geometry and topology, and excessive computational overhead at edge nodes, leading to false deformation reports and computational delays.
A rapid attitude quaternion determination system is constructed by employing a thermal hysteresis compensation model based on exponential decay time memory kernels, a Riemannian manifold smoothing mechanism based on Lie algebra logarithmic and exponential mapping, and a lightweight anomaly identification method based on quaternion geodesic inner product. The system includes attitude observation, thermal hysteresis processing, weighted fusion, manifold smoothing, and anomaly identification modules.
It achieves matching of thermal compensation timescale with dam thermal inertia, suppresses false deformation false alarms, maintains manifold geometric constraints, reduces computational overhead, and meets the real-time early warning requirements of low-power edge nodes.
Smart Images

Figure CN123111178A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of surveying and mapping data processing technology, and relates to a method and system for rapid determination of attitude quaternions. Background Technology
[0002] Safety monitoring of large hydraulic structures (such as ultra-high arch dams, gravity dams, and large-scale slope protection projects) relies on high-precision automated deformation and attitude monitoring systems. Dual-axis inclinometers and temperature sensors are key loads for acquiring local three-dimensional attitude changes and environmental data. To avoid the gimbal lock singularity inherent in Euler angle representation and reduce the computational load on edge computing nodes, high-precision attitude monitoring algorithms commonly use quaternions to parameterize the three-dimensional attitude. However, quaternions only solve the mathematical form of attitude parameterization and do not eliminate the physical errors introduced by sensors in complex field environments. During long-term operation, the observation data is subject to the coupling effects of complex physical fields such as solar radiation, water temperature gradient, wind load, and flood discharge vibration, introducing system biases and dynamic disturbances such as quasi-static thermal distortion, baseline drift, and high-frequency random noise. Therefore, there is an urgent need to establish effective thermal distortion compensation and high-frequency smoothing filtering mechanisms. Existing technologies mainly employ sensor compensation methods based on temperature mapping and linear filtering methods based on Euclidean space, but both have significant limitations.
[0003] The key shortcomings of existing technologies can be summarized in the following three aspects: First, the thermal inertia time scale is mismatched: existing temperature compensation methods use the absolute value of the temperature at the current moment and the instantaneous rate of change as input, and the effective sensing window is only on the order of seconds to minutes. However, the thickness of large hydraulic structures dams is tens of meters, and the structural thermoelastic deformation caused by heat conduction often lags by several hours. The compensation amount cannot reflect the true state of heat accumulation inside the structure. Under the condition of drastic alternation of day and night temperature differences, it is easy to produce overcompensation or undercompensation, inducing false deformation false alarms; Second, linear operations in Euclidean space destroy the manifold geometric topology: three-dimensional pose State data essentially exists on a unit hyperspherical manifold. Existing methods directly perform linear averaging or weighted fusion on attitude sequences in flat Euclidean space, which destroys the manifold topology and unit modulus constraints of the rotation group. Under high-frequency flutter conditions, this causes non-physical distortion in the smoothed attitude trajectory. Third, the computational overhead of edge nodes is too high: Existing methods use Euler angle representation and matrix iterative filtering, which involve a large number of inverse trigonometric functions and matrix operations. This causes severe computational overload and early warning response delay on low-power IoT edge nodes with limited computing power, restricting the timeliness of monitoring structural anomalies. Summary of the Invention
[0004] To address the issues of thermal inertia timescale mismatch, Euclidean space linear operations disrupting manifold geometry and topology, and excessive computational overhead at edge nodes in attitude quaternion determination, as described in the background art, this invention provides a method and system for rapid attitude quaternion determination for deformation monitoring of hydraulic structures. It introduces a thermal hysteresis compensation model based on exponentially decaying time memory kernels, a Riemannian manifold smoothing mechanism based on Lie algebra-logarithmic-exponential mapping, and a lightweight anomaly identification method based on quaternion geodesic inner product. These methods ensure the physical rationality of attitude calculation and meet the requirements for real-time early warning response from low-power edge computing nodes.
[0005] In a first aspect, the present invention provides a method for rapid determination of attitude quaternions for deformation monitoring of hydraulic structures, comprising: S1. Obtain the roll and pitch angles measured at the current moment by the dual-axis inclinometers deployed at key nodes of the hydraulic structure, as well as the ambient temperature collected by the temperature sensor at the same location; set the yaw angle to zero, and convert the roll and pitch angles into observation quaternions that satisfy the unit modulus constraint based on the analytical mapping relationship from Euler angles to quaternions. S2. Extract the temperature observation sequence within a predetermined historical time window prior to the current moment, and perform a weighted integral on the temperature observation sequence using an exponentially decaying time memory kernel function to obtain an equivalent hysteresis thermal distortion scalar characterizing the internal thermal accumulation state of the dam body. The decay coefficient of the exponentially decaying time memory kernel function is determined through offline calibration experiments based on the thermal conductivity of the dam concrete material. Construct a thermistor quaternion perturbation matrix characterizing the thermal expansion and contraction of the sensor base based on the equivalent hysteresis thermal distortion scalar. S3. Based on the observation quaternion obtained in S1 and the thermal quaternion perturbation matrix obtained in S2, and considering the measurement characteristics of the dual-axis inclinometer which is sensitive to the horizontal tilt direction but not sensitive to the yaw direction, an anisotropic spatial covariance weighted matrix is constructed. Based on this, a generalized weighted least squares objective function containing the observation quaternion residuals and thermal perturbation parameters is established and solved iteratively. The thermal hysteresis error component is removed from the observation quaternion, and the rough deformed quaternion is output. S4. The rough deformed quaternion is mapped to the Lie algebra tangent space using a logarithmic operator to obtain the tangent vector; using the tangent vector at adjacent discrete time nodes and its first derivative estimate as control conditions, a Hermitian cubic spline curve is constructed to smooth the tangent vector sequence to obtain a smoothed tangent vector; then, the smoothed tangent vector is inversely mapped back to the unit hyperspherical manifold using an exponential operator to output the true deformed quaternion that satisfies the unit modulus constraint. S5. Perform geodesic inner product operation on the actual deformation quaternion and the pre-stored safety reference attitude quaternion to obtain the attitude deviation scalar d; compare the attitude deviation scalar d with the preset anomaly judgment threshold δ: when d > δ, it is determined that the structure has undergone abnormal deformation and an early warning signal is triggered; when d ≤ δ, it is determined that the structure is in a normal state and return to step S1 to continue collecting the next epoch observation data.
[0006] Furthermore, in step S1, the analytic mapping relationship from Euler angles to quaternions is as follows: , in, This is the roll angle. The pitch angle; To observe a quaternion, its four components are a scalar component and three vector components, respectively.
[0007] Furthermore, in step S2, the temperature observation sequence within a historical time window of length M prior to the current time k is extracted. Calculate the equivalent hysteresis thermal distortion scalar using the following formula. : , in, Lagging behind the current time Historical temperature observations of the step It is an exponentially decaying time-memory kernel function; The decay coefficient of the exponentially decaying time-memory kernel function. Related to the thermal conductivity of the dam concrete material and determined through offline calibration experiments, the memory decay rate of historical temperatures is controlled; the weighted integral makes the contribution of recent temperature to the equivalent hysteresis thermal distortion scalar greater than that of long-term temperature, so as to quantitatively characterize the thermal accumulation state inside the concrete dam body in terms of time phase. The attenuation coefficient The initial values were determined through offline calibration experiments based on a one-dimensional heat conduction model of the dam body. Specifically, the dam body thickness L and the thermal diffusivity ξ of the dam concrete material at the sensor placement locations were obtained, and the values were calculated according to τ = L. 2 / ξ Calculate the dam body heat conduction lag time constant τ, and take the attenuation coefficient The ratio of the sampling interval Δt of the temperature observation sequence to the heat conduction lag time constant τ is λ = Δt / τ; the length M of the historical time window is taken as the smallest integer not less than c·τ / Δt, where c is a preset window coverage coefficient not less than 1, so that the duration of the historical time window covers the dam body heat conduction lag time constant τ.
[0008] Furthermore, in step S2, the thermal quaternion perturbation matrix Build as: , In the formula, , For the constant mechanical bias of the sensor, , The temperature-attitude coupling drift coefficients in the X and Y axes are respectively, and are predetermined through offline calibration experiments; the yaw component of the thermal quaternion perturbation matrix is set to zero, consistent with the setting of zero yaw angle in step S1.
[0009] Furthermore, in step S3, the anisotropic space covariance weighting matrix W = diag(w x w y w z ), where w x w y Take the higher weight, w z Based on the unobservable characteristic of the yaw direction, take a value much smaller than w. x w y The weights are adjusted to suppress the interference of yaw direction noise on the fusion results; The weights of the anisotropic spatial covariance weighted matrix are dynamically adjusted based on the high-frequency dynamic load conditions. Specifically, the vibration intensity index ρ(k) is calculated based on the sliding variance of the geodesic deviation between adjacent epochs in the quaternion sequence observed before the current time. When ρ(k) exceeds a preset vibration threshold, it is determined that the load is under flood discharge excitation or strong wind load conditions, and the weights are adjusted accordingly. z Press w z / (1+γ·ρ(k)) is adjusted downwards, where γ is a preset adjustment coefficient, and w x w y Keep the baseline weights unchanged; when ρ(k) is lower than the preset vibration threshold, w z Restore the baseline weights; The generalized weighted least squares objective function is: , In the formula, Indicates the current moment and the previously preset length is Summing multiple epochs within a historical window. For epoch index; To use the constant mechanical bias of the sensor , and temperature-attitude coupling drift coefficient , As parameters, in epochs The equivalent hysteresis thermal distortion scalar The input thermal error prediction quaternion model, whose expression is related to the thermal quaternion perturbation matrix. The construction form remains consistent; the iteration terminates when the residual change between two adjacent iterations is lower than a preset convergence threshold, and the optimal estimation result is output, which is the rough deformed quaternion. ; After the iterative solution terminates, calculate the fusion residual for the current epoch. Starting with the initial value of the attenuation coefficient λ, and following the formula λ(k+1)=λ(k)-μ·Δe 2 (k) / Δλ is used to perform an online recursive correction on the attenuation coefficient λ, where Δe 2 (k) / Δλ is the numerical partial derivative of the square of the fusion residual with respect to the attenuation coefficient λ with respect to the preset differential step size, and μ is the preset adaptive step size factor; the corrected attenuation coefficient λ is used to calculate the equivalent hysteresis thermal distortion scalar in the next epoch, so that the memory decay rate of the exponential decay time memory kernel function is adaptively adjusted online with the actual thermal response state of the dam.
[0010] Furthermore, in step S4, the rough-deformed quaternion is... Through logarithmic operators Expanding to a flat Lie algebra space, we introduce a first-order Taylor expansion to construct a piecewise function: , , In the formula, q w For the rough deformation quaternion scalar component, q v For the rough deformation quaternion The vector components, ε is a preset minimum positive threshold; when When it is approximated by a first-order Taylor expansion, that is... After logarithmic mapping Transformed into eigenvectors in a flat Lie algebra space .
[0011] Furthermore, in step S4, within the Lie algebra tangent space, the tangent vector at adjacent discrete time nodes is used... The first derivative estimate is used as the control condition. Hermitian cubic spline curves are constructed, and smooth tangent vectors are generated. ; Smooth the tangent vector using the exponential operator Inverse mapping restores the model to a unit hyperspherical manifold, outputting the true deformed quaternion. ;set up Given the magnitude of the tangent vector, the mapping relationship is restored as follows: , After restoration by the exponential operator, the true transformed quaternion The unit modulus length constraint is satisfied.
[0012] Furthermore, in step S5, during the initialization phase of the monitoring system, inclinometer observation data is collected when the structure is in a known safe state. This data is then processed through steps S1 to S4 to obtain the safe reference attitude quaternion. ; During the real-time monitoring phase, the solution obtained at the current moment will be... and Perform the four-dimensional vector geodesic inner product operation and define the attitude deviation scalar. : , In the formula, This indicates taking the absolute value, used to eliminate symbolic ambiguity caused by the double-covering property of quaternions; This indicates that the current attitude is completely consistent with the reference attitude. A value close to 1 indicates the largest deviation; The anomaly judgment threshold δ is dynamically adjusted based on the environmental disturbance state. Specifically, the temperature change rate index r(k) is calculated based on the temperature observation sequence before the current time, and the vibration intensity index ρ(k) is calculated based on the sliding variance of the geodesic deviation between adjacent epochs in the observation quaternion sequence. The anomaly judgment threshold for the current epoch is determined according to δ(k)=δ0·(1+η1·|r(k)|+η2·ρ(k)), where δ0 is the benchmark threshold calibrated under normal structural and stable environmental conditions, and η1 and η2 are preset sensitivity coefficients. When the alternating day and night temperature difference leads to an increase in the temperature change rate index, or when the vibration intensity index increases due to flood discharge vibration or strong wind load conditions, the anomaly judgment threshold is adjusted upward accordingly to suppress false structural deformation false alarms caused by thermal compensation residuals and high-frequency residual noise.
[0013] Secondly, the present invention provides a rapid attitude quaternion determination system for deformation monitoring of hydraulic structures, used to execute the rapid attitude quaternion determination method for deformation monitoring of hydraulic structures as described above, including: The attitude observation module is configured to acquire the roll and pitch angles measured at the current moment by the dual-axis inclinometers deployed at key nodes of the hydraulic structure, as well as the ambient temperature collected by the temperature sensor at the same location; the yaw angle is set to zero, and the roll and pitch angles are converted into observation quaternions that satisfy the unit modulus constraint based on the analytical mapping relationship from Euler angles to quaternions. The thermal hysteresis processing module is configured to extract the temperature observation sequence within a predetermined historical time window prior to the current moment, and perform a weighted integral on the temperature observation sequence using an exponentially decaying time memory kernel function to obtain an equivalent hysteresis thermal distortion scalar characterizing the internal thermal accumulation state of the dam body. The decay coefficient of the exponentially decaying time memory kernel function is determined through offline calibration experiments based on the thermal conductivity of the dam concrete material. Based on the equivalent hysteresis thermal distortion scalar, a thermistor quaternion perturbation matrix characterizing the thermal expansion and contraction of the sensor base is constructed. The weighted fusion module is configured to construct an anisotropic spatial covariance weighted matrix based on the observed quaternion and the thermal quaternion perturbation matrix, taking into account the measurement characteristics of the dual-axis inclinometer which is sensitive to the horizontal tilt direction but not sensitive to the yaw direction. Based on this, a generalized weighted least squares objective function containing the observed quaternion residuals and thermal perturbation parameters is established and solved iteratively. The thermal hysteresis error component is extracted from the observed quaternion, and a rough deformed quaternion is output. The manifold smoothing module is configured to map the rough deformed quaternion to the Lie algebra tangent space using a logarithmic operator to obtain a tangent vector; using the tangent vector at adjacent discrete time nodes and its first derivative estimate as control conditions, a Hermitian cubic spline curve is constructed to smooth the tangent vector sequence to obtain a smoothed tangent vector; then, the smoothed tangent vector is inversely mapped back to the unit hyperspherical manifold using an exponential operator, and the output is a true deformed quaternion that satisfies the unit modulus constraint; The anomaly identification module is configured to perform a geodesic inner product operation on the actual deformation quaternion and the pre-stored safety reference attitude quaternion to obtain an attitude deviation scalar d; and compare the attitude deviation scalar d with a preset anomaly judgment threshold δ: when d > δ, it is determined that the structure has undergone abnormal deformation and an early warning signal is triggered; when d ≤ δ, it is determined that the structure is in a normal state and the attitude observation module is controlled to continue to collect the observation data of the next epoch. The attitude observation module, thermal hysteresis processing module, weighted fusion module, manifold smoothing module, and anomaly identification module are deployed on the same IoT edge computing node. The observation quaternion output by the attitude observation module and the thermal quaternion perturbation matrix output by the thermal hysteresis processing module are jointly input to the weighted fusion module. The weighted fusion module, manifold smoothing module, and anomaly identification module are sequentially connected, with the output of the previous module serving as the input of the next module.
[0014] Thirdly, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for rapid determination of attitude quaternions for deformation monitoring of hydraulic structures.
[0015] Compared with the prior art, the present invention has the following advantages: (1) The thermal compensation time scale matches the thermal inertia of the dam body and suppresses false deformation false alarms: The present invention uses an exponential decay time memory kernel function to perform weighted integration on the temperature observation sequence within a predetermined length of historical window before the current moment, to obtain an equivalent hysteresis thermal distortion scalar that characterizes the internal thermal accumulation state of the dam body, and the decay coefficient is determined offline based on the thermal conductivity of the dam concrete material; Compared with the existing technology, which only uses the absolute value of the current temperature and the instantaneous rate of change as input and the effective sensing window is only on the order of seconds to minutes, the mechanism of the present invention can quantitatively characterize the phase characteristics of the thermal conduction of large-volume concrete dam body that lags by several hours, so that the compensation amount reflects the true thermal accumulation state inside the structure, effectively suppressing the over-compensation or under-compensation phenomenon under the condition of drastic alternation of day and night temperature difference, and reducing false structural deformation false alarms; (2) Improved accuracy of thermal hysteresis error removal and more accurate attitude calculation results: This invention addresses the measurement characteristics of dual-axis inclinometers, which are sensitive to the horizontal tilt direction but not to the yaw direction. It constructs an anisotropic spatial covariance weighted matrix and uses it to establish a generalized weighted least squares objective function that includes observation quaternion residuals and thermal disturbance parameters for iterative solution. This weighting mechanism allows observation information from different directions to participate in the calculation according to its credibility. It can accurately remove the thermal hysteresis error component from the observation quaternion, avoid the confusion between thermal disturbance and actual structural deformation, and ensure that the output rough deformation quaternion accurately reflects the actual attitude change of the structure. (3) Maintaining manifold geometric constraints throughout the process and avoiding non-physical distortion of attitude trajectory: Three-dimensional attitude data essentially exists on a unit hyperspherical manifold. Existing technologies directly perform linear mean or weighted fusion on attitude sequences in flat Euclidean space, which will destroy the manifold topology and unit modulus constraints of the rotation group. This invention maps the rough deformed quaternion to the Lie algebra tangent space through the logarithmic operator, smooths the tangent vector sequence with Hermite cubic spline curves in the tangent space, and then restores it to the unit hyperspherical manifold through the inverse mapping of the exponential operator. The smoothing process satisfies the unit modulus constraint and the rigid body rotation continuity condition, which can effectively suppress the flutter noise introduced by high-frequency dynamic disturbances such as flood discharge excitation and wind load, and avoid non-physical distortion of the smoothed attitude trajectory. (4) Lightweight anomaly identification calculation meets the real-time early warning requirements of edge nodes: In the anomaly identification stage, the present invention directly performs geodesic inner product operation on the real deformed quaternion and the pre-stored safety reference attitude quaternion to obtain the attitude deviation scalar and compare it with the threshold to complete the judgment. This avoids the large number of inverse trigonometric functions and matrix operations involved in the Euler angle representation and matrix iterative filtering of the existing technology. At the same time, the quaternion parameterization representation itself avoids the gimbal deadlock singularity inherent in the Euler angle representation method. The above attitude observation, thermal hysteresis modeling, weighted fusion, manifold smoothing and anomaly identification modules can be deployed as a whole on the same IoT edge computing node. The computational overhead is low, which effectively alleviates the computing power overload problem of low power edge nodes and meets the real-time early warning response requirements of large hydraulic engineering structure anomaly monitoring. (5) The three mechanisms are connected in sequence to form a complete process and ensure the physical rationality of the solution: thermal hysteresis compensation, manifold smoothing and lightweight anomaly identification are connected in sequence, and the output of the previous step is used as the input of the next step: thermal hysteresis compensation provides clean attitude input after stripping thermal error for manifold smoothing, and manifold smoothing provides real attitude trajectory with manifold constraint preserved for anomaly identification. The three together constitute a complete attitude quaternion fast determination process applicable to the edge computing nodes of hydraulic structures. The entire link from error modeling, trajectory smoothing to anomaly identification ensures the physical rationality and engineering reliability of attitude solution.
[0016] In summary, this invention acquires roll angle, pitch angle, and ambient temperature at the same location measured by a dual-axis inclinometer and converts them into observed quaternions. It then uses exponential decay time memory to verify the weighted integral of historical temperature sequences, constructing a thermally sensitive quaternion perturbation matrix. Next, it constructs an anisotropic space covariance weighted matrix and establishes a generalized weighted least squares objective function for iterative solution, eliminating thermal hysteresis errors. The resulting quaternions are then mapped to the Lie algebra tangent space using a logarithmic operator and smoothed using Hermitian cubic splines, followed by inverse mapping using an exponential operator to restore them to a unit hyperspherical manifold. Finally, it identifies structural anomalies and triggers early warnings through geodesic inner product operations. This invention matches the thermal compensation timescale with the dam's thermal inertia, effectively suppressing false deformation alarms, maintaining manifold geometric constraints throughout the smoothing process, avoiding non-physical distortions of attitude trajectories, and achieving lightweight computation, meeting the real-time early warning response requirements of low-power edge nodes. Attached Figure Description
[0017] Figure 1 This is a schematic diagram of the overall architecture and deployment of the hydraulic structure deformation monitoring system of the present invention.
[0018] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation
[0019] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.
[0020] Example 1
[0021] The overall architecture and deployment diagram of the hydraulic structure deformation monitoring system are shown below. Figure 1 As shown, in terms of overall architecture and deployment, this embodiment provides a method for rapid determination of attitude quaternions for deformation monitoring of hydraulic structures. The flowchart is as follows. Figure 2 As shown, the specific steps are as follows.
[0022] S1. Obtain the roll and pitch angles measured at the current moment by the dual-axis inclinometers deployed at key nodes of the hydraulic structure, as well as the ambient temperature collected by the temperature sensor at the same location; set the yaw angle to zero, and convert the roll and pitch angles into observation quaternions that satisfy the unit modulus constraint based on the analytical mapping relationship from Euler angles to quaternions.
[0023] In practice, the hydraulic structure deformation monitoring system first acquires the roll angle measured at the current moment by biaxial inclinometers deployed at key nodes of the hydraulic structure. (Rotation around the X-axis), pitch angle (Rotation around the Y-axis), and the absolute temperature collected by the ambient temperature sensor at the same location. This serves as the input for this step.
[0024] After obtaining the original angle data, considering the physical deformation constraints of the hydraulic structure, its yaw twist around the vertical direction is minimal, and the yaw angle is approximately set. This reduces the 3D attitude calculation problem to an analytical mapping problem that relies solely on two-axis tilt angle observations, thus avoiding the introduction of unobservable third-axis errors.
[0025] Based on the above, and according to the analytical mapping relationship from Euler angles to quaternions, the roll angle is... With pitch angle Convert to unit observation quaternion The mapping relationship is analyzed as follows: , in, This is the roll angle. The pitch angle; To observe a quaternion, its four components are a scalar component and three vector components, respectively.
[0026] This mapping avoids the Euler angle gimbal deadlock singularity while transforming two-dimensional tilt observations into those satisfying unit modulus constraints. The quaternion representation provides a unified basis for attitude parameterization in subsequent steps. However, the observation quaternions generated at this point... The error includes static baseline drift due to the thermal inertia of large-volume concrete, and this thermal hysteresis error needs to be addressed subsequently.
[0027] S2. Extract the temperature observation sequence within a predetermined historical time window prior to the current moment, and perform a weighted integral on the temperature observation sequence using an exponentially decaying time memory kernel function to obtain an equivalent hysteresis thermal distortion scalar characterizing the internal thermal accumulation state of the dam body. The decay coefficient of the exponentially decaying time memory kernel function is determined through offline calibration experiments based on the thermal conductivity of the dam concrete material. Construct a thermistor quaternion perturbation matrix characterizing the thermal expansion and contraction of the sensor base based on the equivalent hysteresis thermal distortion scalar.
[0028] In view of the physical characteristics of the heat conduction process of large-volume concrete dams having hourly time phase lag, this invention does not adopt a static compensation scheme with the absolute value of the current temperature as input, but instead establishes an equivalent hysteresis thermal distortion scalar with time memory effect.
[0029] First, extract the temperature observation sequence within a historical time window of length M prior to the current time k. Calculate the equivalent hysteresis thermal distortion scalar using the following formula. : , in, Lagging behind the current time Historical temperature observations of the step It is an exponentially decaying time-memory kernel function; The decay coefficient of the exponentially decaying time-memory kernel function. Related to the thermal conductivity of the dam concrete material and determined through offline calibration experiments, the memory decay rate of historical temperatures is controlled; the weighted integral makes the contribution of recent temperature to the equivalent hysteresis thermal distortion scalar greater than that of long-term temperature, so as to quantitatively characterize the thermal accumulation state inside the concrete dam body in the time phase, rather than just reflecting the instantaneous temperature change on the surface.
[0030] In practical operation, attenuation coefficient The initial values were determined through offline calibration experiments based on a one-dimensional heat conduction model of the dam body. Specifically, the dam body thickness L and the thermal diffusivity ξ of the dam concrete material at the sensor placement locations were obtained, and the values were calculated according to τ = L. 2 / ξ Calculate the dam body heat conduction lag time constant τ, and take the attenuation coefficient The ratio of the sampling interval Δt of the temperature observation sequence to the heat conduction lag time constant τ is λ = Δt / τ; the length M of the historical time window is taken as the smallest integer not less than c·τ / Δt, where c is a preset window coverage coefficient not less than 1, so that the duration of the historical time window covers the dam body heat conduction lag time constant τ.
[0031] Taking a biaxial inclinometer installed on an observation pier on the dam crest as an example: the equivalent thermal inertia thickness at the sensor installation location is L = 0.3 m, and the thermal diffusivity of the dam concrete is ξ = 1.0 × 10⁻⁶ m. -6 m 2 / s, according to τ=L 2 The dam's thermal conduction lag time constant τ is calculated to be 9.0 × 10⁻⁶. 4 The sampling interval for the temperature observation sequence is Δt = 600 s, therefore the initial value of the attenuation coefficient λ = Δt / τ ≈ 6.7 × 10⁻⁶. -3 If the window coverage factor c = 3, then the historical time window length M should be no less than 3 × 9.0 × 10⁻⁶. 4 The smallest integer of / 600 = 450, i.e., M = 450, corresponds to a window duration of = historical time window length M × sampling interval Δt = 75 h, covering the heat conduction hysteresis time constant τ, and the residual weight e of the historical temperature at the end of the window. -3 ≈5%, which is negligible.
[0032] Then, the equivalent hysteresis thermal distortion scalar is introduced into the quaternion space to construct a thermistor quaternion perturbation matrix characterizing the thermal expansion and contraction of the sensor base. The expression is: , In the formula, , For the constant mechanical bias of the sensor, , The temperature-attitude coupling drift coefficients in the X and Y axes are respectively, and are predetermined through offline calibration experiments; the yaw component of the thermal quaternion perturbation matrix is set to zero, consistent with the setting of zero yaw angle in step S1.
[0033] At this point, the observed quaternions output in step S1 are... The thermal quaternion perturbation matrix constructed in step S2 Together they serve as the anisotropic weighted fusion input for step S3, used to extract from... The thermal hysteresis error component is stripped from the sample to extract the true structural deformation signal.
[0034] S3. Based on the observation quaternion obtained in S1 and the thermal quaternion perturbation matrix obtained in S2, an anisotropic spatial covariance weighted matrix is constructed to address the measurement characteristics of the dual-axis inclinometer, which is sensitive to the horizontal tilt direction but insensitive to the yaw direction. Based on this, a generalized weighted least squares objective function containing the observation quaternion residuals and thermal perturbation parameters is established and solved iteratively. The thermal hysteresis error component is removed from the observation quaternion, and a rough deformed quaternion is output.
[0035] To address the physical measurement characteristics of dual-axis inclinometers, which are sensitive to horizontal tilt (X and Y axes) but insensitive to vertical yaw (Z axis), an anisotropic spatial covariance weighting matrix W = diag(w x w y w z ), where w x w y Take the higher weight, w z Based on the unobservable characteristic of the yaw direction, take a value much smaller than w. x w y The weights, such as w z Compared to w x w y It should be at least an order of magnitude smaller to suppress the interference of yaw direction noise on the fusion results.
[0036] The weights of the anisotropic spatial covariance weighted matrix are dynamically adjusted based on the high-frequency dynamic load conditions. Specifically, the vibration intensity index ρ(k) is calculated based on the sliding variance of the geodesic deviation between adjacent epochs in the quaternion sequence observed before the current time. When ρ(k) exceeds the preset vibration threshold, it is determined that the load is under flood discharge excitation or strong wind load conditions, and the weights are adjusted accordingly. z Press w z / (1+γ·ρ(k)) is adjusted downwards, where γ is a preset adjustment coefficient, and w x w y Keep the baseline weights unchanged; when ρ(k) is lower than the preset vibration threshold, w z Restore the baseline weights.
[0037] With a measurement angle noise σ x ≈σ y Taking a high-precision dual-axis inclinometer with a yaw rate of approximately 0.001° as an example: Since there is no direct observation in the yaw direction, we take w... x =w y =1, w z =0.01, that is, the benchmark weighted matrix W = diag(1, 1, 0.01), w z Compared to w x w yTwo orders of magnitude smaller; the variance of the geodesic deviation between adjacent epochs is calculated using a sliding window of 60 epochs as the vibration intensity index ρ(k), which is measured in square degrees. A preset vibration threshold ρ is used. th =1×10 -4 (Corresponding to an RMS jitter of approximately 0.01°), ρ th Measured in square degrees, the preset adjustment coefficient γ = 1 × 10 4 γ is in units of the reciprocal of square degrees. The measured ρ(k) under flood discharge vibration conditions is approximately 1 × 10⁻⁶. -2 w z According to 0.01 / (1+10) 4 ×10 -2 Reduced to approximately 9.9 × 10 -5 ; after ρ(k) falls below the threshold, w z Restored to 0.01. In actual operation, σ x σ y ρ th The specific values of γ depend on the actual sensor accuracy level and the field conditions, and should be confirmed based on the datasheet specifications of the inclinometer used and the field trial operation data.
[0038] Based on the above, a generalized weighted least squares objective function is established that includes the observation quaternion residuals and thermal perturbation parameters: , In the formula, Indicates the current moment and the previously preset length is Summing multiple epochs within a historical window. For epoch index; To use the constant mechanical bias of the sensor , and temperature-attitude coupling drift coefficient , As parameters, in epochs The equivalent hysteresis thermal distortion scalar The input thermal error prediction quaternion model, whose expression is related to the thermal quaternion perturbation matrix. The construction form remains consistent; the iteration terminates when the residual change between two adjacent iterations is lower than a preset convergence threshold, and the optimal estimation result is output, which is the rough deformed quaternion. .
[0039] In addition, after the iterative solution terminates, the fusion residual for the current epoch is calculated. Starting with the initial value of the attenuation coefficient λ, and following the formula λ(k+1)=λ(k)-μ·Δe 2(k) / Δλ is used to perform an online recursive correction on the attenuation coefficient λ, where Δe 2 (k) / Δλ is the numerical partial derivative of the square of the fusion residual with respect to the attenuation coefficient λ with respect to the preset differential step size, and μ is the preset adaptive step size factor; the corrected attenuation coefficient λ is used to calculate the equivalent hysteresis thermal distortion scalar in the next epoch, so that the memory decay rate of the exponential decay time memory kernel function is adaptively adjusted online with the actual thermal response state of the dam.
[0040] With an initial value of λ of 6.7 × 10 -3 Difference step size h = 5 × 10 -4 Adaptive step size factor μ = 5 × 10 2 For example: After the iteration terminates at a certain epoch, the fused residual e(k) ≈ 1.28 × 10 -5 After perturbing λ by h, the squared residual increment is recalculated, and Δe is obtained by forward difference. 2 (k) / Δλ≈2.4×10 -7 Then λ(k+1)=6.7×10 -3 -5×10 2 ×2.4×10 -7 ≈6.58×10 -3 The single-step correction amplitude is approximately 1.8%; the corrected λ is used for the calculation of the equivalent hysteresis thermal distortion scalar in the next epoch. The online correction range of λ is limited to [1.3 × 10⁻⁶]. -3 2×10 -2 In practice, the values of μ and h are strongly dependent on the residual magnitude and sensor configuration. The values above are self-consistent examples, and the final values need to be confirmed based on the field calibration data.
[0041] Although Low-frequency temperature drift interference has been eliminated, but under high-frequency dynamic load conditions such as flood discharge vibration and strong wind load, its attitude sequence still contains high-frequency flutter noise components, which directly affect... Performing linear smoothing on the sequence in Euclidean space will destroy the manifold constraints of the unit hypersphere where the quaternion resides (i.e., the geometric topology of the SO(3) manifold). Therefore, it is necessary to proceed to step S4 to perform topology-preserving manifold smoothing in the Lie algebra tangent space.
[0042] S4. Map the rough deformed quaternion to the Lie algebra tangent space using a logarithmic operator to obtain the tangent vector; construct a Hermitian cubic spline curve to smooth the tangent vector sequence using the tangent vector and its first derivative estimate at adjacent discrete time nodes as control conditions to obtain the smoothed tangent vector; then inversely map the smoothed tangent vector back to the unit hyperspherical manifold using an exponential operator to output the true deformed quaternion that satisfies the unit modulus constraint.
[0043] This step uses the rough deformed quaternion output from step S3. For the input, this step performs topological smoothing on the high-frequency flutter noise components while maintaining the quaternion unit hyperspherical manifold constraint. If directly applied to Euclidean space... The sequence performs a linear mean operation, and the smoothing result will deviate from the unit hypersphere, violating the unit modulus constraint of rigid body rotation, resulting in non-physical distortion.
[0044] Therefore, this embodiment executes the following sub-steps in sequence: Logarithmic dimensionality reduction mapping: transforming coarsely deformed quaternions Through logarithmic operators Expand to a flat Lie algebra space and extract its eigenvectors. We introduce a first-order Taylor expansion to construct a piecewise function: , , In the formula, q w For the rough deformation quaternion scalar component, q v For the rough deformation quaternion The vector components, ε is a preset minimum positive threshold; when When the first-order Taylor expansion is used as an approximation, i.e., v raw ≈2q v After logarithmic mapping Transformed into eigenvectors in a flat Lie algebra space Regular numerical operations can be performed within this space without violating manifold constraints.
[0045] Taking double-precision floating-point implementation as an example, let ε = 1 × 10 -6 :when When the deformation angle does not exceed approximately 0.16°, the first-order Taylor expansion approximation v is used. raw ≈2q v The approximate relative error at the boundary is approximately 3 × 10⁻⁶. -7 It is far lower than the sensor's noise floor; when At that time, according to Precise calculations are required. If deployed on single-precision floating-point edge nodes, ε should ideally be 1 × 10⁻⁶. -4 Up to 1×10 -6 A larger value is selected to avoid cancellation errors. In practice, the final value of ε needs to be confirmed in conjunction with the floating-point implementation of the edge computing nodes.
[0046] Tangent-space Hermitian cubic spline smoothing: In the Lie algebraic tangent space, the tangent vector at adjacent discrete time nodes is used as the basis for smoothing. The first derivative estimate is used as the control condition. Hermitian cubic spline curves are constructed, and smooth tangent vectors are generated. Compared to ordinary cubic splines, Hermite cubic splines simultaneously constrain function values and derivative values at nodes, ensuring the first-order continuity of the smooth trajectory. This effectively filters out flutter noise components introduced by high-frequency dynamic loads such as flood discharge vibration and strong wind loads, while retaining low-frequency signals that characterize the actual structural deformation.
[0047] Exponential dimensionality reduction: smoothing the tangent vector using the exponential operator Inverse mapping restores the model to a unit hyperspherical manifold, outputting the true deformed quaternion. ;set up Given the magnitude of the tangent vector, the mapping relationship is restored as follows: , After restoration by the exponential operator, the true transformed quaternion The unit modulus length constraint is satisfied.
[0048] Right now Satisfying the unit modulus constraint Mathematically, it is guaranteed to be a legal rotational orientation on the corresponding SO(3) manifold. As the input for the geodesic inner product operation in step S5, it has simultaneously eliminated low-frequency thermal hysteresis drift error and high-frequency flutter noise, and can be used for terminal abnormality identification.
[0049] S5. Perform geodesic inner product operation on the actual deformation quaternion and the pre-stored safety reference attitude quaternion to obtain the attitude deviation scalar d; compare the attitude deviation scalar d with the preset anomaly judgment threshold δ: when d > δ, it is determined that the structure has undergone abnormal deformation and an early warning signal is triggered; when d ≤ δ, it is determined that the structure is in a normal state and return to step S1 to continue collecting the next epoch observation data.
[0050] The true deformed quaternion output in step S4 As input, this step performs lightweight anomaly identification on low-power IoT edge computing nodes. Instead of using Euler angle inverse trigonometric function operations, it directly uses the quaternion geodesic inner product as a scalar distance metric to avoid gimbal deadlock risk and reduce computational overhead.
[0051] Specifically, during the initialization phase of the hydraulic structure deformation monitoring system, inclinometer observation data is collected when the structure is in a known safe state. The data is then processed through steps S1 to S4 to obtain the safe reference attitude quaternion. And it is stored locally on the edge computing node.
[0052] During the real-time monitoring phase, the solution obtained at the current moment will be... and Perform the four-dimensional vector geodesic inner product operation and define the attitude deviation scalar. : , In the formula, This indicates taking the absolute value, used to eliminate symbolic ambiguity caused by the double-covering property of quaternions; This indicates that the current attitude is completely consistent with the reference attitude. A value close to 1 indicates the maximum deviation. This operation consists of only 4 multiplications and 3 additions, totaling 7 basic arithmetic operations. The computational cost for each warning judgment is fixed and does not change with the number of monitoring nodes or the scale of historical data.
[0053] scalar of attitude deviation Compared with the preset anomaly detection threshold The comparison is performed, and the following branch condition is executed: when When an abnormal deformation is detected in the structure, the edge computing node triggers a hardware interrupt warning signal, and simultaneously... And the corresponding timestamp is uploaded to the cloud data platform; when If the structure is determined to be in a normal state, return to step S1 to continue collecting observation data for the next epoch.
[0054] In addition, the anomaly judgment threshold δ is dynamically adjusted according to the environmental disturbance state. Specifically, the temperature change rate index r(k) is calculated based on the temperature observation sequence before the current time, and the vibration intensity index ρ(k) is calculated based on the sliding variance of the geodesic deviation between adjacent epochs in the observation quaternion sequence. The anomaly judgment threshold for the current epoch is determined according to δ(k)=δ0·(1+η1·|r(k)|+η2·ρ(k)), where δ0 is the benchmark threshold calibrated under normal structural and stable environmental conditions, and η1 and η2 are preset sensitivity coefficients. When the alternation of day and night temperature difference leads to an increase in the temperature change rate index, or when the vibration intensity index increases due to flood discharge vibration or strong wind load conditions, the anomaly judgment threshold is adjusted accordingly to suppress false structural deformation false alarms caused by thermal compensation residuals and high-frequency residual noise.
[0055] Taking a monitoring point of a dam as an example: the benchmark threshold δ0 is determined by taking 30 days of historical data during normal operation. -7 (Corresponding to a geodesic deviation angle of approximately 0.11°), the temperature change rate index r(k) is measured in °C / h, and the vibration intensity index ρ(k) is measured in square degrees (°). 2 The measurement is performed using the same sliding window as step S3; η1 = 0.25 h / °C and η2 = 100 °C are used. -2 Under steady operating conditions (|r|≈0.2 °C / h, ρ≈2×10), -6 ° 2 ), δ(k)≈5.26×10-7 Under conditions of severe diurnal temperature variation and flood discharge vibration (|r|≈2 °C / h, ρ≈1×10), -2 ° 2 δ(k) was adjusted upwards to approximately 1.25 × 10⁻⁶. -6 , which is 2.5 times the baseline; δ(k) is limited to [5×10 -7 2.5×10 -6 This is to prevent the threshold from being excessively raised, leading to missed reports. In practice, the final values of δ0, η1, and η2 depend on the statistical distribution of the normal operating data and the required early warning sensitivity.
[0056] Thus, this embodiment fully realizes the data processing flow from original tilt angle observation to terminal anomaly identification, and successively solves three defects: thermal inertia time scale mismatch, Euclidean space linear operation violating unit hyperspherical manifold constraints, and excessive computational overhead of edge nodes. It outputs high-quality attitude quaternions that meet physical rationality constraints and real-time early warning judgment results.
[0057] Example 2
[0058] A rapid attitude quaternion determination system for deformation monitoring of hydraulic structures is proposed, which consists of an attitude observation module, a thermal hysteresis processing module, a weighted fusion module, a manifold smoothing module, and an anomaly identification module.
[0059] The attitude observation module is configured to acquire the roll and pitch angles measured at the current moment by the dual-axis inclinometers deployed at key nodes of the hydraulic structure, as well as the ambient temperature collected by the temperature sensor at the same location; the yaw angle is set to zero, and the roll and pitch angles are converted into observation quaternions that satisfy the unit modulus constraint based on the analytical mapping relationship from Euler angles to quaternions.
[0060] The thermal hysteresis processing module is configured to extract the temperature observation sequence within a predetermined historical time window prior to the current moment, and perform a weighted integral on the temperature observation sequence using an exponentially decaying time memory kernel function to obtain an equivalent hysteresis thermal distortion scalar characterizing the internal thermal accumulation state of the dam body. The decay coefficient of the exponentially decaying time memory kernel function is determined through offline calibration experiments based on the thermal conductivity of the dam concrete material. Based on the equivalent hysteresis thermal distortion scalar, a thermistor quaternion perturbation matrix characterizing the thermal expansion and contraction of the sensor base is constructed.
[0061] The weighted fusion module is configured to construct an anisotropic spatial covariance weighted matrix based on the observed quaternions and the thermal quaternion perturbation matrix, taking into account the measurement characteristics of the dual-axis inclinometer which is sensitive to the horizontal tilt direction but insensitive to the yaw direction. Based on this matrix, a generalized weighted least squares objective function containing the observed quaternion residuals and thermal perturbation parameters is established and solved iteratively. The thermal hysteresis error component is extracted from the observed quaternions, and a rough deformed quaternion is output.
[0062] The manifold smoothing module is configured to map the rough deformed quaternion to the Lie algebra tangent space using a logarithmic operator to obtain a tangent vector; using the tangent vector at adjacent discrete time nodes and its first derivative estimate as control conditions, a Hermitian cubic spline curve is constructed to smooth the tangent vector sequence to obtain a smoothed tangent vector; then, the smoothed tangent vector is inversely mapped back to the unit hyperspherical manifold using an exponential operator, and the output is a true deformed quaternion that satisfies the unit modulus constraint.
[0063] The anomaly identification module is configured to perform a geodesic inner product operation on the actual deformation quaternion and the pre-stored safety reference attitude quaternion to obtain an attitude deviation scalar d; and compare the attitude deviation scalar d with a preset anomaly judgment threshold δ: when d > δ, it is determined that the structure has undergone abnormal deformation and an early warning signal is triggered; when d ≤ δ, it is determined that the structure is in a normal state and the attitude observation module is controlled to continue to collect observation data for the next epoch.
[0064] The attitude observation module, thermal hysteresis processing module, weighted fusion module, manifold smoothing module, and anomaly identification module are deployed on the same IoT edge computing node. The observation quaternion output by the attitude observation module and the thermal quaternion perturbation matrix output by the thermal hysteresis processing module are jointly input to the weighted fusion module. The weighted fusion module, manifold smoothing module, and anomaly identification module are sequentially connected, with the output of the previous module serving as the input of the next module.
[0065] The specific implementation methods of each module in this system are the same as those described in Example 1, and will not be repeated here.
[0066] Example 3
[0067] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for rapid determination of attitude quaternions for deformation monitoring of hydraulic structures as described in Embodiment 1 above, and the system for rapid determination of attitude quaternions for deformation monitoring of hydraulic structures as described in Embodiment 2.
[0068] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented in various computer languages, such as object-oriented programming languages like Java, C++, Python, and interpreted scripting languages like JavaScript.
[0069] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, electronic devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing electronic device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing electronic device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0070] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing electronic device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0071] These computer program instructions can also be loaded onto a computer or other programmable data processing electronic device to cause a series of operational steps to be performed on the computer or other programmable electronic device to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable electronic device for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0072] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.
[0073] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A method for rapid determination of attitude quaternions for deformation monitoring of hydraulic structures, characterized in that, include: S1. Obtain the roll angle and pitch angle measured at the current moment by the dual-axis inclinometers deployed at key nodes of the hydraulic structure, as well as the ambient temperature collected by the temperature sensor at the same location. Set the yaw angle to zero, and based on the analytical mapping relationship between Euler angles and quaternions, convert the roll angle and pitch angle into observation quaternions that satisfy the unit module constraint; S2. Extract the temperature observation sequence within a predetermined historical time window prior to the current moment, and perform a weighted integral on the temperature observation sequence using an exponentially decaying time memory kernel function to obtain an equivalent hysteresis thermal distortion scalar characterizing the internal thermal accumulation state of the dam body. The decay coefficient of the exponentially decaying time memory kernel function is determined through offline calibration experiments based on the thermal conductivity of the dam concrete material. Construct a thermistor quaternion perturbation matrix characterizing the thermal expansion and contraction of the sensor base based on the equivalent hysteresis thermal distortion scalar. S3. Based on the observation quaternion obtained in S1 and the thermal quaternion perturbation matrix obtained in S2, and considering the measurement characteristics of the dual-axis inclinometer which is sensitive to the horizontal tilt direction but not sensitive to the yaw direction, an anisotropic spatial covariance weighted matrix is constructed. Based on this, a generalized weighted least squares objective function containing the observation quaternion residuals and thermal perturbation parameters is established and solved iteratively. The thermal hysteresis error component is removed from the observation quaternion, and the rough deformed quaternion is output. S4. The rough deformed quaternion is mapped to the Lie algebra tangent space using a logarithmic operator to obtain the tangent vector; using the tangent vector at adjacent discrete time nodes and its first derivative estimate as control conditions, a Hermitian cubic spline curve is constructed to smooth the tangent vector sequence to obtain a smoothed tangent vector; then, the smoothed tangent vector is inversely mapped back to the unit hyperspherical manifold using an exponential operator to output the true deformed quaternion that satisfies the unit modulus constraint. S5. Perform geodesic dot product operation between the actual deformed quaternion and the pre-stored safety reference attitude quaternion to obtain the attitude deviation scalar d. The attitude deviation scalar d is compared with the preset anomaly judgment threshold δ: when d > δ, it is determined that the structure has undergone abnormal deformation and an early warning signal is triggered; when d ≤ δ, it is determined that the structure is in a normal state and the process returns to step S1 to continue collecting the next epoch observation data.
2. The method for rapid determination of attitude quaternions according to claim 1, characterized in that: In step S1, the analytic mapping relationship from Euler angles to quaternions is as follows: , in, This is the roll angle. The pitch angle; To observe a quaternion, its four components are a scalar component and three vector components, respectively.
3. The method for rapid determination of attitude quaternions according to claim 1, characterized in that: In step S2, the temperature observation sequence within a historical time window of length M prior to the current time k is extracted. Calculate the equivalent hysteresis thermal distortion scalar using the following formula. : , in, Lagging behind the current time Historical temperature observations of the step For exponentially decaying time memory kernel functions; The decay coefficient of the exponentially decaying time-memory kernel function. Related to the thermal conductivity of the dam concrete material and determined through offline calibration experiments, the memory decay rate of historical temperatures is controlled; the weighted integral makes the contribution of recent temperature to the equivalent hysteresis thermal distortion scalar greater than that of long-term temperature, so as to quantitatively characterize the thermal accumulation state inside the concrete dam body in terms of time phase. The attenuation coefficient The initial values were determined through offline calibration experiments based on a one-dimensional heat conduction model of the dam body. Specifically, the dam body thickness L and the thermal diffusivity ξ of the dam concrete material at the sensor placement locations were obtained, and the values were calculated according to τ = L. 2 / ξ Calculate the dam body heat conduction lag time constant τ, and take the attenuation coefficient The ratio of the sampling interval Δt of the temperature observation sequence to the heat conduction lag time constant τ is λ = Δt / τ; the length M of the historical time window is taken as the smallest integer not less than c·τ / Δt, where c is a preset window coverage coefficient not less than 1, so that the duration of the historical time window covers the dam body heat conduction lag time constant τ.
4. The method for rapid determination of attitude quaternions according to claim 3, characterized in that: In step S2, the thermal quaternion perturbation matrix Build as: , In the formula, , For the constant mechanical bias of the sensor, , The temperature-attitude coupling drift coefficients in the X and Y axes are respectively, and are predetermined through offline calibration experiments; the yaw component of the thermal quaternion perturbation matrix is set to zero, consistent with the setting of zero yaw angle in step S1.
5. The method for rapid determination of attitude quaternions according to claim 4, characterized in that: The step S3, the anisotropic space covariance weighted matrix W = diag(w x , w y , w z ), wherein w x , w y Take high weight, w z According to the weight of the unobservable characteristics of the yaw direction, which is much smaller than w x , w y , to suppress the interference of the yaw direction noise on the fusion result; The weight of the anisotropic spatial covariance weighting matrix is dynamically adjusted according to the high-frequency dynamic load working condition, specifically: the vibration intensity index ρ(k) is calculated according to the sliding variance of the geodetic line deviation between adjacent ephemeris in the observation quaternion sequence before the current moment, when ρ(k) exceeds the preset vibration threshold, it is determined that it is in the flood discharge excitation or strong wind load working condition, w z is adjusted downward according to w z / (1+γ·ρ(k)), γ is a preset adjustment coefficient, w x , w y The reference weight remains unchanged; When ρ(k) is lower than the preset vibration threshold, w z Restore the baseline weights; The generalized weighted least squares objective function is: , In the formula, Indicates the current moment and the previously preset length is Summing multiple epochs within a historical window. For epoch index; To use the constant mechanical bias of the sensor , and temperature-attitude coupling drift coefficient , As parameters, in epochs The equivalent hysteresis thermal distortion scalar The input thermal error prediction quaternion model, whose expression is related to the thermal quaternion perturbation matrix. The construction form remains consistent; the iteration terminates when the residual change between two adjacent iterations is lower than a preset convergence threshold, and the optimal estimation result is output, which is the rough deformed quaternion. ; After the iterative solution terminates, calculate the fusion residual for the current epoch. Starting with the initial value of the attenuation coefficient λ, and following the formula λ(k+1)=λ(k)-μ·Δe 2 (k) / Δλ is used to perform an online recursive correction on the attenuation coefficient λ, where Δe 2 (k) / Δλ is the numerical partial derivative of the square of the fusion residual with respect to the attenuation coefficient λ with respect to the preset difference step size, and μ is the preset adaptive step size factor; The corrected attenuation coefficient λ is used to calculate the equivalent hysteresis thermal distortion scalar in the next epoch, so that the memory decay rate of the exponential decay time memory kernel function is adaptively adjusted online according to the actual thermal response state of the dam.
6. The method for rapid determination of attitude quaternions according to claim 1, characterized in that: In step S4, the rough deformation quaternion Through logarithmic operators Expanding to a flat Lie algebra space, we introduce a first-order Taylor expansion to construct a piecewise function: , , In the formula, q w For the rough deformation quaternion scalar component, q v For the rough deformation quaternion The vector components, ε is a preset minimum positive threshold; when When it is approximated by a first-order Taylor expansion, that is... ; After logarithmic mapping Transformed into eigenvectors in a flat Lie algebra space .
7. The method for rapid determination of attitude quaternions according to claim 6, characterized in that: In step S4, within the Lie algebra tangent space, the tangent vectors at adjacent discrete time nodes are used... The first derivative estimate is used as the control condition. Hermite cubic spline curves are constructed, and smooth tangent vectors are generated. ; Smooth the tangent vector using the exponential operator Inverse mapping restores the model to a unit hyperspherical manifold, outputting the true deformed quaternion. ;set up Given the magnitude of the tangent vector, the mapping relationship is restored as follows: , After restoration by the exponential operator, the true transformed quaternion The unit modulus length constraint is satisfied.
8. The method for rapid determination of attitude quaternions according to claim 1, characterized in that: In step S5, during the initialization phase of the monitoring system, tiltmeter observation data is collected when the structure is in a known safe state. The data is then processed through steps S1 to S4 to obtain the safe reference attitude quaternion. ; During the real-time monitoring phase, the solution obtained at the current moment will be... and Perform the four-dimensional vector geodesic inner product operation and define the attitude deviation scalar. : , In the formula, This indicates taking the absolute value, used to eliminate symbolic ambiguity caused by the double-covering property of quaternions; This indicates that the current attitude is completely consistent with the reference attitude. A value close to 1 indicates the largest deviation; The anomaly judgment threshold δ is dynamically adjusted based on the environmental disturbance state. Specifically, the temperature change rate index r(k) is calculated based on the temperature observation sequence before the current time, and the vibration intensity index ρ(k) is calculated based on the sliding variance of the geodesic deviation between adjacent epochs in the observation quaternion sequence. The anomaly judgment threshold for the current epoch is determined according to δ(k)=δ0·(1+η1·|r(k)|+η2·ρ(k)), where δ0 is the benchmark threshold calibrated under normal structural and stable environmental conditions, and η1 and η2 are preset sensitivity coefficients. When the alternating day and night temperature difference leads to an increase in the temperature change rate index, or when the vibration intensity index increases due to flood discharge vibration or strong wind load conditions, the anomaly judgment threshold is adjusted upward accordingly to suppress false structural deformation false alarms caused by thermal compensation residuals and high-frequency residual noise.
9. A system for rapid determination of attitude quaternions for deformation monitoring of hydraulic structures, used to execute the method for rapid determination of attitude quaternions for deformation monitoring of hydraulic structures as described in any one of claims 1 to 8, characterized in that, include: The attitude observation module is configured to acquire the roll and pitch angles measured at the current moment by the dual-axis inclinometers deployed at key nodes of the hydraulic structure, as well as the ambient temperature collected by the temperature sensor at the same location; the yaw angle is set to zero, and the roll and pitch angles are converted into observation quaternions that satisfy the unit modulus constraint based on the analytical mapping relationship from Euler angles to quaternions. The thermal hysteresis processing module is configured to extract the temperature observation sequence within a predetermined historical time window prior to the current moment, and perform a weighted integral on the temperature observation sequence using an exponentially decaying time memory kernel function to obtain an equivalent hysteresis thermal distortion scalar characterizing the internal thermal accumulation state of the dam body. The decay coefficient of the exponentially decaying time memory kernel function is determined through offline calibration experiments based on the thermal conductivity of the dam concrete material. Based on the equivalent hysteresis thermal distortion scalar, a thermistor quaternion perturbation matrix characterizing the thermal expansion and contraction of the sensor base is constructed. The weighted fusion module is configured to construct an anisotropic spatial covariance weighted matrix based on the observed quaternion and the thermal quaternion perturbation matrix, taking into account the measurement characteristics of the dual-axis inclinometer which is sensitive to the horizontal tilt direction but not sensitive to the yaw direction. Based on this, a generalized weighted least squares objective function containing the observed quaternion residuals and thermal perturbation parameters is established and solved iteratively. The thermal hysteresis error component is extracted from the observed quaternion, and a rough deformed quaternion is output. The manifold smoothing module is configured to map the rough deformed quaternion to the Lie algebra tangent space using a logarithmic operator to obtain a tangent vector; using the tangent vector at adjacent discrete time nodes and its first derivative estimate as control conditions, a Hermitian cubic spline curve is constructed to smooth the tangent vector sequence to obtain a smoothed tangent vector; then, the smoothed tangent vector is inversely mapped back to the unit hyperspherical manifold using an exponential operator, and the output is a true deformed quaternion that satisfies the unit modulus constraint; The anomaly identification module is configured to perform geodesic inner product operation on the actual deformed quaternion and the pre-stored safety reference attitude quaternion to obtain the attitude deviation scalar d. The attitude deviation scalar d is compared with a preset anomaly judgment threshold δ: when d > δ, it is determined that the structure has undergone abnormal deformation and an early warning signal is triggered; when d ≤ δ, it is determined that the structure is in a normal state and the attitude observation module is controlled to continue to collect the observation data of the next epoch. The attitude observation module, thermal hysteresis processing module, weighted fusion module, manifold smoothing module, and anomaly identification module are deployed on the same IoT edge computing node. The observation quaternion output by the attitude observation module and the thermal quaternion perturbation matrix output by the thermal hysteresis processing module are jointly input to the weighted fusion module. The weighted fusion module, manifold smoothing module, and anomaly identification module are sequentially connected, with the output of the previous module serving as the input of the next module.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, it implements the method for rapid determination of attitude quaternions for deformation monitoring of hydraulic structures as described in any one of claims 1 to 8.
