A base station antenna reliability diagnosis method

Abstract

The application provides a base station antenna reliability diagnosis method, relates to the technical field of antenna diagnosis, and comprises the following steps: performing initial detection on a to-be-tested antenna, determining the reference performance state of the to-be-tested antenna before environmental stress is applied, wherein the reference performance state comprises appearance, mechanical performance and electrical performance; placing the to-be-tested antenna under preset environmental test conditions and applying environmental stress for a preset time; performing final detection on the to-be-tested antenna after the environmental stress is applied, and obtaining the final performance state of the to-be-tested antenna; comparing the final performance state with the reference performance state, and obtaining the reliability information of the antenna according to a preset criterion. In this way, the method combines environmental stress testing, vibration analysis and automatic diagnosis of high-precision damage identification, improves the reliability evaluation efficiency and accuracy of the base station antenna, reduces maintenance costs, and ensures long-term stable operation of the communication network.

Description

Technical Field

[0001] This application relates to the field of antenna diagnostic technology, and more specifically, to a method for diagnosing the reliability of base station antennas. Background Technology

[0002] With the rapid development of 5G and future communication technologies, base station antennas, as a core component of wireless communication networks, directly affect the stability and coverage quality of the network. Base station antennas are typically exposed to complex outdoor environments, facing various environmental stresses such as random vibration, temperature changes, and wind loads. These stresses can cause structural damage to the antenna, such as fatigue cracks, loose connections, or material degradation, leading to performance degradation or failure. Currently, existing base station antenna reliability diagnostic methods mainly rely on manual inspection, static electrical performance testing, or simple mechanical checks. These methods have several limitations: first, low diagnostic accuracy, unable to accurately quantify the location and extent of structural damage; second, poor real-time performance, unable to dynamically monitor changes during the application of environmental stress; and third, reliance on a large number of sensors, resulting in high costs and complex deployment. Furthermore, traditional methods often ignore the influence of environmental factors on vibration modal parameters, leading to diagnostic results being affected by noise interference and a high misjudgment rate.

[0003] Therefore, there is an urgent need for an automated diagnostic method that integrates environmental stress testing, vibration analysis, and high-precision damage identification to improve the efficiency and accuracy of base station antenna reliability assessment, reduce maintenance costs, and ensure the long-term stable operation of communication networks. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this application provides a method and system for diagnosing the reliability of base station antennas.

[0005] In a first aspect, this application provides a method for diagnosing the reliability of a base station antenna, including:

[0006] Initial testing is performed on the antenna under test to determine its baseline performance state before environmental stress is applied. The baseline performance state includes appearance, mechanical properties, and electrical properties.

[0007] The antenna under test is placed under preset environmental test conditions, and environmental stress is applied for a preset duration.

[0008] After the environmental stress is applied, the antenna under test is subjected to a final test to obtain the final performance status of the antenna under test.

[0009] The final performance state is compared with the baseline performance state, and the reliability information of the antenna is obtained according to a preset criterion.

[0010] Optionally, the environmental stress is random vibration stress, and the environmental stress applied for a preset duration includes:

[0011] Random vibrations following a predetermined acceleration spectral density curve are sequentially applied to the antenna under test along multiple preset axial directions.

[0012] Optional, also includes:

[0013] A finite element model of the antenna structure under test is obtained, and modal analysis is performed on the finite element model to generate multiple basic mode shape functions. A mode shape dictionary is constructed based on the multiple basic mode shape functions. A probabilistic observation model is constructed in advance based on the Gaussian process regression method. The probabilistic observation model characterizes the response relationship between the vibration modal parameters of the antenna structure under test and the corresponding ambient temperature and structural damage state. Vibration response data is collected using a sparse sensor array arranged on the surface of the antenna structure under test, and corresponding ambient temperature data is collected simultaneously.

[0014] Based on the mode shape dictionary, a sparse reconstruction algorithm is used to process the vibration response data collected by the sparse sensor array to generate full-field mode shape data of the antenna structure under test; based on the full-field mode shape data, a modal identification operation is performed to extract the current measured modal parameters of the antenna structure under test.

[0015] Multiple particles are initialized, and the state of each particle corresponds to a damage information hypothesis. Based on the particle filtering algorithm, the current measured modal parameters, the ambient temperature data, and the damage information hypothesis of each particle are input into the probabilistic observation model to calculate the posterior probability value of each particle, and the weight of each particle is updated based on the posterior probability value.

[0016] The damage information of the antenna structure under test is determined based on the updated particle weights; the damage information includes damage location information and damage degree information.

[0017] Optionally, determining the damage information of the antenna structure under test based on the updated particle weights includes:

[0018] The effective sample size of the plurality of particles is calculated based on the updated particle weights; the effective sample size is determined based on the weight values ​​of the plurality of particles.

[0019] The effective sample size is compared with a preset particle set degradation threshold to determine whether the particle set has degraded.

[0020] In response to the effective sample size being less than the particle set degradation threshold, a regularized resampling operation is performed on the plurality of particles;

[0021] The damage information of the antenna structure under test is determined based on the particle state distribution of the descendant particle set after regularization resampling.

[0022] Optionally, the regularized resampling includes:

[0023] Based on the updated particle weights, a descendant particle set is generated by extracting particles with replacement from the plurality of particles.

[0024] Calculate the particle state covariance of the descendant particle set, and adaptively determine the bandwidth of the Gaussian kernel function based on the particle state covariance;

[0025] The descendant particle set is smoothed based on the Gaussian kernel function after the bandwidth is determined, so as to apply a small perturbation to the particle state in the descendant particle set and obtain a descendant particle set with enhanced diversity.

[0026] Optionally, smoothing the set of descendant particles based on the Gaussian kernel function after the bandwidth is determined includes:

[0027] For each descendant particle in the descendant particle set, the corresponding damage location is extracted from the damage information of the descendant particle.

[0028] Based on the damage location, local physical property parameters of the region corresponding to the damage location are extracted from the finite element model; the local physical property parameters specifically include the local stiffness matrix or modal strain energy density of the region.

[0029] An anisotropic Gaussian perturbation kernel function is constructed for each descendant particle based on the local physical property parameters; specifically, the covariance matrix of the anisotropic Gaussian perturbation kernel function is calculated and generated based on the local physical property parameters.

[0030] The state of the descendant particles is subjected to a small perturbation using the anisotropic Gaussian perturbation kernel function to generate a diverse set of descendant particles that matches the physical characteristics of the antenna structure under test.

[0031] Optionally, generating the full-field mode shape data of the antenna structure under test includes:

[0032] Based on the ambient temperature data and the damage information determined by particle filtering at the previous moment, the ambient temperature data and the damage information are input into the probabilistic observation model to generate a prior weight vector; wherein, each element of the prior weight vector represents the activation probability of the corresponding basic mode function in the mode dictionary under the current working condition.

[0033] The prior weight vector is used to weight each basic mode function in the mode shape dictionary to form a weighted mode shape dictionary;

[0034] Based on the weighted mode shape dictionary, the weighted L1 norm minimization algorithm is used to solve the vibration response data collected by the sparse sensor array to determine the mode shape coefficients used to reconstruct the full-field mode shape data;

[0035] Based on the mode shape coefficients and the weighted mode shape dictionary, the full-field mode shape data of the antenna structure under test are reconstructed.

[0036] Optionally, the acquisition of vibration response data using a sparse sensor array arranged on the surface of the antenna structure under test includes:

[0037] Based on the finite element model, multiple potential sensor location points are defined, and an initial measurement matrix is ​​constructed using the response values ​​of the basic mode functions in the mode dictionary at all potential sensor location points.

[0038] Based on the initial measurement matrix and the mode shape dictionary, a cross-coherence coefficient is calculated to characterize the sparse reconstruction accuracy; the cross-coherence coefficient is used to measure the degree of linear correlation between the responses of each potential sensor location point in the initial measurement matrix;

[0039] An iterative optimization algorithm is used to screen potential sensor locations with the goal of minimizing the mutual coherence coefficient, in order to determine an optimal combination of sensor locations that makes the mutual coherence coefficient reach a preset convergence condition.

[0040] Multiple sensors are arranged on the surface of the antenna structure under test according to the optimal sensor position combination to form the sparse sensor array.

[0041] Optionally, the pre-construction of the probabilistic observation model based on the Gaussian process regression method includes:

[0042] The response relationship is decomposed into a global smoothing trend term characterizing the influence of ambient temperature and a local nonlinear effect term characterizing the influence of structural damage.

[0043] For the global smoothing trend term, a first Gaussian kernel function is constructed, and the influence weight of the ambient temperature input dimension on the trend term is determined using a first automatic association and discrimination mechanism;

[0044] For the local nonlinear effect term, a second Gaussian kernel function is constructed, and a second automatic association discrimination mechanism is used to determine the influence weight of the structural damage state input dimension on the effect term; wherein, the first Gaussian kernel function and the second Gaussian kernel function are of different types;

[0045] The first Gaussian kernel function and the second Gaussian kernel function are summed to form a composite Gaussian kernel function that can decouple different physical influences, and the probabilistic observation model is constructed based on the composite Gaussian kernel function.

[0046] Secondly, this application provides a base station antenna reliability diagnostic system, comprising:

[0047] The acquisition module is used to acquire the finite element model of the antenna structure under test, perform modal analysis on the finite element model, generate multiple basic mode shape functions, and construct a mode shape dictionary based on the multiple basic mode shape functions; a probabilistic observation model is pre-constructed based on the Gaussian process regression method, which characterizes the response relationship between the vibration modal parameters of the antenna structure under test and the corresponding ambient temperature and structural damage state; vibration response data is acquired using a sparse sensor array arranged on the surface of the antenna structure under test, and corresponding ambient temperature data is acquired simultaneously;

[0048] The first processing module is used to process the vibration response data collected by the sparse sensor array based on the mode shape dictionary using a sparse reconstruction algorithm to generate full-field mode shape data of the antenna structure under test; and to perform mode recognition operation based on the full-field mode shape data to extract the current measured mode parameters of the antenna structure under test.

[0049] The second processing module is used to initialize multiple particles, each particle's state corresponding to a damage information hypothesis; based on the particle filtering algorithm, the current measured modal parameters, the ambient temperature data, and the damage information hypothesis of each particle are input into the probabilistic observation model to calculate the posterior probability value of each particle, and the weight of each particle is updated based on the posterior probability value.

[0050] The diagnostic module is used to determine the damage information of the antenna structure under test based on the updated particle weights; the damage information includes damage location information and damage degree information.

[0051] Compared with existing technologies, this application effectively solves the problem of the difficulty in directly obtaining full-field vibration data in engineering practice by combining a limited number of sparse sensor arrays and a mode dictionary. It achieves high-precision sparse reconstruction of full-field modal information, significantly reducing the number of sensors used and the cost of data acquisition. In addition, it decouples the coupling effect of ambient temperature and structural damage on vibration modal parameters by using a probabilistic observation model, which significantly improves the accuracy of damage diagnosis and environmental robustness. At the same time, it introduces a particle filtering algorithm, which realizes real-time tracking and accurate identification of structural damage state under the framework of multi-state assumption. This effectively overcomes the uncertainty caused by measurement noise and temperature disturbance, further enhancing the reliability and adaptability of damage diagnosis results, thereby better meeting the needs of practical engineering applications. Attached Figure Description

[0052] Figure 1 A flowchart of a base station antenna reliability diagnosis method provided in this application;

[0053] Figure 2 A flowchart of a method for determining damage information of the antenna structure under test based on updated particle weights, provided in this application;

[0054] Figure 3 A flowchart of a regularized resampling method provided in this application;

[0055] Figure 4 This is a schematic diagram of a base station antenna reliability diagnostic system provided in this application;

[0056] Figure 5 A flowchart of a reliability diagnosis method provided in this application. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0058] This invention provides a base station antenna reliability diagnosis method, aiming to achieve accurate reliability assessment of base station antennas through systematic environmental stress application and performance comparison. This method is applicable to various types of base station antennas, such as multiple-input multiple-output (MIMO) antennas, directional antennas, or omnidirectional antennas, and is particularly suitable for 5G and above communication antennas exposed to complex outdoor environments. The core of this method lies in simulating the stress effects under actual working conditions and quantifying antenna reliability information through before-and-after performance comparisons, thereby guiding antenna design optimization, maintenance, and lifespan prediction.

[0059] See Figure 5 The diagram shown is a flowchart of a reliability diagnosis method provided in this application, including steps S401 to S404, wherein:

[0060] S401: Perform initial testing on the antenna under test to determine its baseline performance state before environmental stress is applied. The baseline performance state includes appearance, mechanical properties and electrical properties.

[0061] In this step, the initial inspection aims to establish a baseline performance for the antenna under test for subsequent comparisons. Specifically, the antenna is first visually inspected, including using an optical microscope or high-definition camera to examine the antenna housing, radiating elements, feed network, and mounting bracket for defects such as scratches, deformation, corrosion, or loosening. For example, a 3D scanner can be used to digitally model the overall antenna structure and quantify dimensional deviations to ensure that the inspection accuracy meets practical requirements. Simultaneously, the location and extent of any visible damage are recorded.

[0062] Secondly, mechanical performance testing is performed, including static strength testing and modal analysis. Specifically, the antenna under test can be fixed on a vibration table or tensile testing machine, and a preset static load, such as the uniform pressure simulating wind load, can be applied to measure deformation, stress, and strain. Data is collected using strain gauge arrays, for example, resistance strain gauges arranged at key nodes. Finite element method (FEM) simulations are then used to verify that the mechanical performance meets design specifications, such as ensuring the stiffness modulus is within a reasonable range.

[0063] Finally, electrical performance testing is performed, including measurements of parameters such as VSWR, gain, radiation pattern, and isolation. In practice, the antenna under test is placed in a microwave anechoic chamber, and a vector network analyzer is used to scan within the operating frequency band to obtain the VSWR curve, gain value, and beamwidth of the radiation pattern. Simultaneously, port isolation and polarization purity are tested. All data is recorded digitally to form a baseline performance status report, including numerical values, graphs, and photographs, ensuring traceability.

[0064] The initial testing described above establishes a comprehensive baseline performance status, preventing deviations from being introduced during subsequent diagnostics. This step can be performed using an automated testing platform, with the time required adjusted according to actual conditions; it is suitable for batch antenna diagnostics.

[0065] S402: Place the antenna under test under a preset environmental test condition and apply environmental stress for a preset duration;

[0066] In this step, the antenna under test is placed in an environmental test chamber or vibration table to simulate stress conditions under actual deployment environments. Preset environmental test conditions include appropriate temperature ranges, humidity, and atmospheric pressure to cover typical weather scenarios. The purpose of applying environmental stress is to accelerate the manifestation of potential damage and assess the antenna's durability.

[0067] Optionally, the environmental stress is random vibration stress, and the environmental stress applied for a preset duration includes: sequentially applying random vibrations to the antenna under test along multiple preset axial directions, following a predetermined acceleration spectral density curve. Specifically, the random vibration stress simulates random loads such as vehicle transportation, wind vibration, or earthquakes. The preset axial directions typically include the X-axis (horizontal transverse), Y-axis (horizontal longitudinal), and Z-axis (vertical direction), which can be appropriately expanded according to the antenna installation orientation.

[0068] The predetermined acceleration spectral density (PSD) curve is designed based on relevant international standards, such as exhibiting a Gaussian distribution within an appropriate frequency band. The application process utilizes an electromagnetic vibration table, with a controller generating random signals to ensure the vibration power spectral density matches the preset curve. The duration is preset according to the antenna type, for example, applied along multiple axes with appropriate intervals between axes to avoid heat accumulation.

[0069] During application, the antenna response is monitored in real time, such as by using accelerometers (placed on the antenna base and radiating end) to collect vibration data, ensuring that it does not exceed safe thresholds. If an anomaly is detected, such as a significant shift in the resonant frequency, application can be interrupted to prevent destructive damage. This step ensures the controllability and repeatability of environmental stress, facilitating subsequent analysis.

[0070] S403: After the environmental stress is applied, the antenna under test is subjected to a final test to obtain the final performance status of the antenna under test;

[0071] In this step, after the environmental stress is applied and the antenna has cooled to a suitable temperature, the antenna under test is subjected to final inspection to capture stress-induced changes. The final inspection method is the same as the initial inspection, including a comprehensive evaluation of appearance, mechanical, and electrical performance. Specifically, the appearance inspection focuses on newly appearing cracks or deformations, and the changes are quantified using the same equipment; the mechanical performance inspection repeats the static load test and compares the strain data; the electrical performance inspection rescans parameters such as VSWR and gain to identify potential degradation.

[0072] To ensure accuracy, non-destructive testing techniques can be introduced, such as ultrasonic flaw detectors to detect internal cracks or thermal imagers to observe abnormal heat distribution. All data records form a final performance status report, which corresponds to the baseline state for easy comparison item by item.

[0073] S404: Compare the final performance state with the baseline performance state, and obtain the reliability information of the antenna based on a preset criterion.

[0074] In this step, reliability is quantified through data comparison. The comparisons include numerical differences, curve deviations, and image matching. Specifically, for appearance, pixel-level differences are calculated using image processing software; for mechanical properties, modal frequency shifts are calculated; and for electrical properties, VSWR changes are calculated.

[0075] The preset criteria are based on industry standards and empirical thresholds. For example, if all parameter variations are within acceptable ranges, the reliability can be judged as "excellent"; if the variations exceed a certain level, it is judged as "unacceptable," and the probability of damage is output, such as using a Bayesian model for estimation. Reliability information includes an overall score, potential failure points, and lifetime prediction, such as using a Weibull distribution model to predict remaining lifetime. The output report can be visualized through the software interface, facilitating user decision-making.

[0076] This embodiment achieves a comprehensive diagnosis of base station antenna reliability through the above steps. In practical applications, parameters can be adjusted according to the specific antenna type, such as increasing salt spray stress to simulate a coastal environment. This method is highly scalable and can be integrated into automated production lines. Ensuring full disclosure, those skilled in the art can implement the invention based on this method.

[0077] Optional, see Figure 1 The diagram shown is a flowchart of a base station antenna reliability diagnosis method provided in this application, which also includes steps S101 to S104, wherein:

[0078] S101: Obtain the finite element model of the antenna structure under test, perform modal analysis on the finite element model to generate multiple basic mode shape functions, and construct a mode shape dictionary based on the multiple basic mode shape functions; pre-construct a probabilistic observation model based on the Gaussian process regression method, the probabilistic observation model characterizing the response relationship between the vibration modal parameters of the antenna structure under test and the corresponding ambient temperature and structural damage state; collect vibration response data using a sparse sensor array arranged on the surface of the antenna structure under test, and simultaneously collect the corresponding ambient temperature data;

[0079] S102: Based on the mode shape dictionary, the vibration response data collected by the sparse sensor array is processed using a sparse reconstruction algorithm to generate full-field mode shape data of the antenna structure under test; based on the full-field mode shape data, a mode recognition operation is performed to extract the current measured mode parameters of the antenna structure under test.

[0080] S103: Initialize multiple particles, each particle's state corresponding to a damage information hypothesis; based on the particle filtering algorithm, input the current measured modal parameters, the ambient temperature data, and the damage information hypothesis of each particle into the probabilistic observation model, calculate the posterior probability value of each particle, and update the weight of each particle based on the posterior probability value;

[0081] S104: Determine the damage information of the antenna structure under test based on the updated particle weights; the damage information includes damage location information and damage degree information.

[0082] Regarding the above S101:

[0083] In practice, the first step is to establish a finite element model of the antenna structure under test. For example, a simulation model can be built using finite element software such as ANSYS or Abaqus based on the actual size, material properties and boundary constraints of the antenna under test, and then the simulation model can be meshed.

[0084] Next, modal analysis is performed on the established finite element model. For example, analysis conditions can be set, eigenvalue calculations can be performed, and modal parameters of several orders of the antenna structure under test can be obtained, including the basic mode shape function, that is, the natural mode shapes of the structure, and the corresponding natural frequencies.

[0085] Subsequently, a mode dictionary is constructed using the obtained basic mode functions. In practice, the mode dictionary is composed of column vectors of basic mode functions obtained from modal analysis, with each column vector representing the spatial vibration mode of the corresponding order, thus forming the data foundation for the subsequent sparse reconstruction algorithm.

[0086] Furthermore, a probabilistic observation model is constructed based on the Gaussian process regression method. In practice, the modal parameters of the antenna structure under test are used as output, while environmental temperature data and structural damage status data are used as input. A large amount of modal parameter data under different environmental temperature conditions and different damage degrees and locations is collected as a training dataset. The Gaussian process regression method is then used to train this dataset to learn and fit the response relationship between modal parameters and temperature and damage. This probabilistic observation model can describe the combined influence of temperature and damage effects on the modal parameters of the antenna structure under test, and can be used to predict the response characteristics of the antenna structure under test during subsequent diagnostic processes.

[0087] The training dataset can be generated by introducing virtual damage of different types, locations and degrees into the finite element model and performing batch simulations by applying different temperature loads to the model under various damage scenarios.

[0088] Finally, vibration response data were acquired using a sparse sensor array optimized and arranged on the surface of the antenna under test. This sensor array has a relatively small number of sensors, and their positions were optimized to ensure the representativeness of the measurement signal and the accuracy of sparse reconstruction.

[0089] For example, the acquisition process can simultaneously acquire the vibration acceleration signal or vibration displacement signal of the antenna under test under actual working conditions, and record the corresponding ambient temperature data to ensure the consistency and time synchronization between the vibration response data and the temperature data, so as to provide accurate data input for subsequent mode shape reconstruction and damage diagnosis.

[0090] Regarding S102 above:

[0091] In the specific implementation, the mode dictionary constructed in step S101 is first used as the basis for sparse reconstruction.

[0092] Specifically, each column vector of the mode shape dictionary represents the vibration response characteristics of the antenna structure under test at all possible measurement points for a certain order of the basic mode shape.

[0093] Next, the vibration response data obtained from the sparse sensor array optimized and arranged on the surface of the antenna under test in step S101 is processed using a sparse reconstruction algorithm. In specific implementations, sparse reconstruction methods such as orthogonal matched pursuit (OMP), basis pursuit (BP), or weighted L1 norm minimization can be used. The collected small amount of sensor data is used as observations, and the mode shape dictionary is used as a sparse basis function set. By solving an optimization problem, the optimal combination coefficients of each basic mode shape function in the mode shape dictionary are determined, thereby accurately reconstructing the full-field mode shape data of the antenna structure under test at all spatial locations.

[0094] For example, if a weighted L1 norm minimization algorithm is used, the activation probability of each basic mode shape function under the current working condition can be predicted using a probabilistic observation model based on the damage information determined at the previous moment and the synchronously collected ambient temperature data. This probability is used to form a priori weight vector, which is then used to weight each basic mode shape function in the mode shape dictionary, prioritizing those with higher activation probabilities for reconstruction calculation. Then, using the weighted mode shape dictionary as the basis function set, a sparse optimization problem is solved using a convex optimization toolbox (e.g., CVX or SPGL1) to obtain the corresponding coefficient vector. This coefficient vector is then combined with the original mode shape dictionary to generate accurate full-field mode shape data.

[0095] After obtaining the full-field vibration mode data, modal identification is further performed. In specific implementations, modal identification may include analyzing the full-field vibration mode data using algorithms such as frequency domain peak picking, random subspace identification (SSI), or characteristic system realization (ERA) to extract the measured modal parameters of the antenna structure under test under the current operating conditions, including modal frequencies, damping ratios, and corresponding mode shapes.

[0096] In this way, full-field modal information can be accurately obtained from data collected from a limited number of measurement points, providing a reliable data foundation for the next step of damage diagnosis.

[0097] Regarding the above S103:

[0098] In practice, the particles are first initialized. In this embodiment, each particle represents a possible damage state hypothesis of the antenna structure under test, where the damage state specifically includes information such as damage location, damage type, and damage degree.

[0099] For example, the damage location can be defined as a preset discrete grid point on the antenna structure under test. The damage type can include stiffness reduction or local material property degradation, and the damage degree can be characterized by the percentage of stiffness reduction or the magnitude of modal strain energy change. During initialization, multiple particles can be generated in the damage information state space using uniform distribution or random sampling methods to cover potential structural damage states and ensure the effectiveness of subsequent particle filtering.

[0100] Subsequently, based on the particle filtering algorithm, the weight of each particle is calculated. Specifically, the measured modal parameters extracted in step S102, such as the measured modal frequencies and mode shapes, the synchronously measured ambient temperature data, and the damage state assumption corresponding to each particle are input into the pre-constructed probabilistic observation model in step S101. The probabilistic observation model calculates the predicted modal parameters under the damage state assumption for each particle, and compares the predicted modal parameters with the current measured modal parameters to obtain the corresponding likelihood probability, i.e., the posterior probability value of each particle.

[0101] The specific likelihood probability calculation can be achieved by quantifying the difference between the predicted value output by the probability observation model and the measured modal parameters. For example, the error between the predicted modal frequency and the measured modal frequency, or the difference index between the predicted mode shape and the measured mode shape, such as the modal guarantee criterion MAC value, can be used as the evaluation basis, and the posterior probability value of each particle can be calculated by using the Gaussian likelihood function.

[0102] After obtaining the posterior probability values ​​of each particle, the particle weights are updated based on a particle filtering algorithm. Specifically, the weight of each particle is proportional to its calculated posterior probability value. Particles with higher weights indicate that their corresponding damage hypothesis matches the current measured data more closely, while particles with lower weights indicate that their hypothesis matches the current observed data less closely. The set of particles with updated weights serves as the basis for determining damage information in the next step.

[0103] In this way, the current vibration mode measurement results and ambient temperature data can be efficiently integrated into the particle filtering framework, and the weight of each particle can be accurately calculated, thus laying a data foundation for determining the damage information of the antenna structure under test.

[0104] Regarding S104 above:

[0105] In practice, the damage information of the antenna structure under test is first determined using the updated particle weights from step S103. Specifically, after updating the weights of each particle, the effective sample size of the particle set can be calculated to evaluate the diversity and degradation degree of the particle set. In practical applications, the inverse variance of the particle weights or the entropy value of the particle weights can be used to calculate the effective sample size.

[0106] Subsequently, if the calculated effective sample size is less than the pre-set particle set degradation threshold, it indicates that the particle set has degraded, meaning that most of the weights are concentrated on a few particles, thus affecting the accuracy of damage diagnosis. Therefore, a regularized resampling operation can be further performed to improve the diversity of the particle set.

[0107] Specific regularized resampling operations may include: first, randomly sampling the particle set with replacement according to the particle weights to obtain a new descendant particle set; then, calculating the particle state covariance matrix of the newly generated descendant particle set, and adaptively determining the bandwidth parameter of the Gaussian kernel function used for particle state perturbation based on the covariance matrix; finally, applying a small perturbation to the state of each particle in the descendant particle set using the determined Gaussian kernel function to increase the diversity of the particle set and avoid the degradation phenomenon caused by the particle set being overly concentrated on a single state assumption.

[0108] Furthermore, to better characterize the physical properties of the antenna structure under test, the Gaussian kernel function used for perturbation can be anisotropic. Specifically, for each descendant particle, based on its corresponding damage location information, local physical property parameters, such as the local stiffness matrix or modal strain energy density, are extracted from the finite element model in step S101. Then, based on these local physical property parameters, an anisotropic Gaussian kernel function is constructed for each particle. Specifically, the covariance matrix of the Gaussian kernel function is calculated and determined using the extracted local physical property parameters. Subsequently, the state of the descendant particles is perturbed using this anisotropic Gaussian kernel function, thereby obtaining a more diverse set of descendant particles that better matches the actual physical properties of the antenna structure under test.

[0109] After obtaining the particle set processed by the above resampling, density clustering algorithms can be further used to perform cluster analysis on these particles in the damage information state space to identify one or more particle clusters with high probability density. For example, DBSCAN or other density clustering methods can be used to automatically identify and distinguish different damage information hypothesis clusters by setting appropriate distance and density thresholds.

[0110] For each identified high-probability-density particle cluster, the state statistical center of each particle cluster is further calculated, which can be the mean or median of the particle states of each cluster, forming candidate damage information states. Then, these candidate damage information states, together with the current ambient temperature data, are input again into the probabilistic observation model constructed in step S101, and the prediction confidence of each candidate damage information state is calculated. This prediction confidence is used to quantitatively characterize the degree of matching and reliability of each candidate damage information state with the actual measurement data under the current temperature conditions.

[0111] Finally, the state with the highest prediction confidence is selected from the above candidate damage information states as the final damage information of the antenna structure under test, thereby clearly determining the damage location information and damage degree information, and realizing accurate damage diagnosis of the antenna structure under test.

[0112] In this way, damage information of the antenna structure under test can be determined efficiently and accurately under complex temperature conditions and noise interference, effectively meeting the practical engineering needs of structural damage diagnosis.

[0113] As an optional implementation method, see [link to implementation details]. Figure 2 The flowchart provided in this application illustrates a method for determining damage information of the antenna structure under test based on updated particle weights, including steps S201 to S204, wherein:

[0114] S201: Calculate the effective sample size of the plurality of particles based on the updated particle weights; the effective sample size is determined based on the weight values ​​of the plurality of particles;

[0115] S202: Compare the effective sample size with a preset particle set degradation threshold to determine whether the particle set has degraded;

[0116] S203: In response to the effective sample size being less than the particle set degradation threshold, a regularized resampling operation is performed on the plurality of particles;

[0117] S204: Determine the damage information of the antenna structure under test based on the particle state distribution of the descendant particle set after regularization resampling.

[0118] In practical implementation, when determining the damage information of the antenna structure under test based on the updated particle weights in step S103, the effective sample size (ESS) of the particle set is first calculated based on the updated particle weight values ​​in the particle filtering algorithm. Specifically, the effective sample size is used to quantitatively evaluate the concentration and diversity of the particle set's weights. For example, the effective sample size can be calculated using the reciprocal of the sum of squares of the particle weights or the entropy value of the particle weights.

[0119] For example, the effective sample size can be calculated as follows: First, calculate the sum of squares of all particle weights, then take the reciprocal of the sum of squares; the resulting value is the effective sample size. When the particle weights are highly concentrated in a few particles, the effective sample size is small; conversely, when the particle weights are relatively evenly distributed, the effective sample size is large.

[0120] Subsequently, the calculated effective sample size is compared with a preset particle set degradation threshold to determine whether the current particle set has degraded. Specifically, a particle set degradation threshold can be preset based on engineering experience or simulation analysis. When the effective sample size is lower than this particle set degradation threshold, it indicates that the weights of the particle set have been concentrated on a small number of particles, indicating that the particle set has degraded. At this time, a regularized resampling operation needs to be performed to restore the diversity of the particle set.

[0121] The specific regularized resampling operation can be implemented as follows: First, random sampling with replacement is performed on the particle set according to the updated particle weights to generate a new descendant particle set; second, based on the obtained descendant particle set, its particle state covariance matrix is ​​calculated, and then the bandwidth parameter of the Gaussian kernel function used for particle state perturbation is adaptively determined according to the covariance matrix; then, the determined Gaussian kernel function is used to apply small perturbations to each particle state of the descendant particle set to increase the state diversity of the descendant particle set and avoid the particle set from being overly concentrated in a single or a few states.

[0122] After completing the above regularization resampling operation, a descendant particle set with enhanced state diversity is obtained. Finally, based on the particle state distribution of this descendant particle set, the damage information of the antenna structure under test is determined. Specifically, a weighted statistical method for the particle set can be used to calculate the statistical characteristics of the particle state distribution, such as the weighted mean or mode of the particle states, to determine the damage location and degree information, thereby obtaining accurate structural damage diagnosis results.

[0123] This effectively overcomes the degradation phenomenon of particle assemblies, improves the accuracy and reliability of damage diagnosis results, and thus meets the needs of damage diagnosis of antenna structures under test in practical engineering.

[0124] As an optional implementation method, see [link to implementation details]. Figure 3 The flowchart of a regularized resampling method provided in this application includes steps S301 to S303, wherein:

[0125] S301: Based on the updated particle weights, extract particles with replacement from the plurality of particles to generate a descendant particle set.

[0126] S302: Calculate the particle state covariance of the descendant particle set, and adaptively determine the bandwidth of the Gaussian kernel function based on the particle state covariance;

[0127] S303: Based on the Gaussian kernel function after the bandwidth is determined, the descendant particle set is smoothed to apply a small perturbation to the particle state in the descendant particle set, thereby obtaining a descendant particle set with enhanced diversity.

[0128] In practice, the regularized resampling operation includes the following steps:

[0129] First, based on the updated particle weights in step S103, random sampling with replacement is performed on the current particle set. Specifically, importance sampling is used, meaning that the probability of each particle being selected is directly proportional to its weight value. This results in particles with higher weights having a greater probability of being selected, while particles with lower weights are eliminated or selected less frequently. Through this sampling process, a new set of descendant particles is obtained, allowing the particle state distribution to initially reflect the probability of damage under the current working condition.

[0130] Subsequently, based on the generated set of descendant particles, the particle state covariance matrix is ​​calculated. In specific implementation, firstly, the weighted mean of the state vectors of all particles in the descendant particle set is calculated. Then, based on the weighted mean and the state vector of each particle, the difference between each particle's state vector and the weighted mean is calculated. Furthermore, each difference vector is combined with the particle weights to calculate the weighted covariance matrix of the particle set's state.

[0131] Then, based on the particle state covariance matrix, the bandwidth parameter of the Gaussian kernel function used for smoothing is adaptively determined. In specific implementations, the Silverman rule or other adaptive bandwidth determination methods can be used to calculate the optimal bandwidth parameter suitable for the current particle state spatial distribution based on the particle state covariance matrix. This ensures that the determined Gaussian kernel function accurately reflects the true distribution of particle states, avoiding both excessive diffusion and excessive concentration of particle states.

[0132] After determining the bandwidth parameter of the Gaussian kernel function, a small perturbation is applied to the state of each particle in the descendant particle set based on the determined Gaussian kernel function to achieve smoothing.

[0133] In practice, the following operations can be performed on a particle-by-particle basis: For each particle in the descendant particle set, a random perturbation vector conforming to a Gaussian distribution is generated using the determined Gaussian kernel function, and this perturbation vector is superimposed on the particle's state, thereby achieving smooth processing of the particle state. This process ensures a significant enhancement of the diversity of particle states, while also effectively preventing the particle set from becoming overly concentrated in a single state and thus degenerating.

[0134] This allows the particle filtering algorithm to maintain high particle state diversity throughout the damage diagnosis process of the antenna structure under test, thereby improving the accuracy and robustness of damage diagnosis and effectively enhancing the ability to identify structural damage under complex working conditions.

[0135] As an optional implementation, smoothing the descendant particle set based on the Gaussian kernel function after the bandwidth is determined includes:

[0136] For each descendant particle in the descendant particle set, the corresponding damage location is extracted from the damage information of the descendant particle.

[0137] Based on the damage location, local physical property parameters of the region corresponding to the damage location are extracted from the finite element model; the local physical property parameters specifically include the local stiffness matrix or modal strain energy density of the region.

[0138] An anisotropic Gaussian perturbation kernel function is constructed for each descendant particle based on the local physical property parameters; specifically, the covariance matrix of the anisotropic Gaussian perturbation kernel function is calculated and generated based on the local physical property parameters.

[0139] The state of the descendant particles is subjected to a small perturbation using the anisotropic Gaussian perturbation kernel function to generate a diverse set of descendant particles that matches the physical characteristics of the antenna structure under test.

[0140] In specific implementation, the particle smoothing process in regularized resampling is further optimized by introducing anisotropic Gaussian kernel functions constrained by structural physical properties to perturb the particle state. The specific implementation method includes the following steps:

[0141] First, for each descendant particle in the descendant particle set obtained in step S303, the damage location information in its corresponding damage state hypothesis is extracted. Specifically, the damage state of a descendant particle is usually represented by the particle's state vector, which explicitly records or encodes the location of the damage region, such as its position index or coordinate point on the finite element model mesh.

[0142] For example, for each particle, one or more grid nodes or cells can be identified as damage location information for subsequent operations.

[0143] Next, using the finite element model of the antenna structure under test established in step S101, and based on the extracted damage location information, local physical characteristic parameters of the corresponding damage region are extracted. These local physical characteristic parameters specifically include, but are not limited to, the local stiffness matrix or modal strain energy density corresponding to the damage location region. These parameters can be pre-stored by preprocessing the finite element model or by offline calculation. For example, the local stiffness matrix can be represented by the element stiffness sub-matrix or the relevant nodal stiffness sub-matrix of the finite element model in the damage location region; the modal strain energy density can be calculated from the mode response of the finite element model in the damage region.

[0144] Subsequently, based on the aforementioned local physical property parameters, a corresponding anisotropic Gaussian perturbation kernel function is constructed for each descendant particle. In specific implementation, the core of the anisotropic Gaussian perturbation kernel function lies in the construction of its covariance matrix. Specifically, the covariance matrix can be directly calculated based on the local stiffness matrix or modal strain energy density of the damaged region. For example, the local stiffness matrix can be appropriately normalized or positive definite to obtain the covariance matrix used to construct the Gaussian kernel function. Furthermore, the calculation process of the covariance matrix can be scaled in conjunction with the bandwidth parameter determined in step S302 to ensure that the constructed anisotropic Gaussian kernel function not only reflects the structural physical properties but also adapts to the current distribution of the particle state.

[0145] After determining and constructing the anisotropic Gaussian perturbation kernel function for each particle, the kernel function is used to apply a small perturbation to the state of each particle in the descendant particle set. Specifically, a random perturbation vector is generated for each particle using its specific anisotropic Gaussian kernel function, and this random perturbation vector is superimposed on the particle's original state vector to obtain the perturbated particle state.

[0146] The resulting particle ensemble obtained through the anisotropic perturbation method exhibits higher physical realism and state diversity. In particular, this method considers the differences in physical characteristics of local regions of the antenna structure under test, enabling the particle perturbation process to accurately reflect the actual dynamic behavior of the structure. This avoids unreasonable state perturbations caused by blindly using isotropic kernel functions, effectively improving the rationality of the particle ensemble's distribution in physical space and the reliability and accuracy of damage diagnosis.

[0147] Therefore, the anisotropic particle perturbation method implemented by combining the local physical characteristic parameters of the structure described above significantly enhances the ability of the damage diagnosis algorithm for the antenna structure under test to capture the real damage state, and can more effectively realize the damage identification and diagnosis of the antenna structure under test under complex engineering conditions.

[0148] As an optional implementation, generating the full-field vibration mode data of the antenna structure under test includes:

[0149] Based on the ambient temperature data and the damage information determined by particle filtering at the previous moment, the ambient temperature data and the damage information are input into the probabilistic observation model to generate a prior weight vector; wherein, each element of the prior weight vector represents the activation probability of the corresponding basic mode function in the mode dictionary under the current working condition.

[0150] The prior weight vector is used to weight each basic mode function in the mode shape dictionary to form a weighted mode shape dictionary;

[0151] Based on the weighted mode shape dictionary, the weighted L1 norm minimization algorithm is used to solve the vibration response data collected by the sparse sensor array to determine the mode shape coefficients used to reconstruct the full-field mode shape data;

[0152] Based on the mode shape coefficients and the weighted mode shape dictionary, the full-field mode shape data of the antenna structure under test are reconstructed.

[0153] In specific implementation, to further improve the accuracy of sparse reconstruction of vibration response data and the precision of modal parameter extraction, the generation process of the full-field mode shape data includes the following specific implementation steps:

[0154] First, based on the synchronously acquired ambient temperature data and the damage information determined by the particle filter algorithm at the previous moment, the data is input into the probabilistic observation model constructed in step S101. Specifically, the ambient temperature data can be obtained by temperature sensors installed on or near the surface of the antenna structure under test; the damage information at the previous moment is determined by the particle filter algorithm, specifying details such as the location, type, and extent of structural damage.

[0155] The probabilistic observation model predicts the activation probability of each fundamental mode shape function of the antenna structure under test under current operating conditions based on the current input ambient temperature data and damage information. Specifically, the probabilistic observation model utilizes pre-learned responses of temperature and damage state to modal parameters to predict the likelihood of activation of the fundamental mode shape functions based on the current temperature conditions and damage state. This likelihood is typically expressed as a probability, forming a priori weight vector. Each element in this vector corresponds to the activation probability of each fundamental mode shape function in the mode shape dictionary, with values ​​ranging from 0 to 1. A higher probability indicates that the corresponding fundamental mode shape function is more likely to be activated under the current operating conditions.

[0156] Subsequently, the prior weight vector obtained above is used to weight each basic mode function in the mode shape dictionary to form a weighted mode shape dictionary. Specifically, the activation probability value corresponding to each basic mode function is used as a weighting factor and multiplied by the column vector corresponding to that basic mode function. Through this operation, basic mode functions with higher activation probabilities will play a more important role in subsequent sparse reconstruction calculations, while basic mode functions with lower activation probabilities are relatively suppressed, ensuring that the reconstruction process more closely reflects the actual vibration characteristics of the structure under the current operating conditions.

[0157] Furthermore, using the weighted mode shape dictionary and the vibration response data collected by the optimized sparse sensor array as observations, a weighted L1 norm minimization algorithm is employed to solve the sparse reconstruction problem, thereby determining the mode shape coefficients used for full-field mode shape data reconstruction. In specific implementation, the weighted L1 norm minimization algorithm is achieved through the following optimization problem:

[0158] First, define the objective function as follows:

[0159]

[0160] in, This is the vector of mode shape coefficients to be solved, whose elements correspond to the basic mode shape functions in the mode shape dictionary; The vibration response measurement vector acquired by the sensor; This is the measurement matrix corresponding to the sensor positions in the mode shape dictionary; It is a diagonal matrix, and its diagonal elements are the prior weight vectors obtained in the above steps; This is the preset error tolerance value.

[0161] In practice, convex optimization tools such as CVX, SPGL1, or YALL1 can be used to solve the above sparse optimization problem quickly and efficiently, thereby obtaining the optimal mode shape coefficient vector.

[0162] Finally, by combining the determined mode shape coefficient vectors and the original mode shape dictionary, the accurate full-field mode shape data of the antenna structure under test under the current operating conditions are reconstructed. Specifically, this can be achieved using the following matrix operations:

[0163]

[0164] in, To reconstruct the full-field mode shape data vector, This is the matrix form of the mode shape dictionary.

[0165] In this way, the influence of ambient temperature and structural damage state on the vibration characteristics of the antenna under test is comprehensively considered, and efficient and accurate sparse reconstruction of measurement data is achieved. This significantly improves the quality and accuracy of the full-field vibration mode data, thereby further improving the reliability and accuracy of subsequent structural damage diagnosis and more effectively meeting the actual needs of complex engineering environments.

[0166] As an optional implementation, the acquisition of vibration response data using a sparse sensor array arranged on the surface of the antenna structure under test includes:

[0167] Based on the finite element model, multiple potential sensor location points are defined, and an initial measurement matrix is ​​constructed using the response values ​​of the basic mode functions in the mode dictionary at all potential sensor location points.

[0168] Based on the initial measurement matrix and the mode shape dictionary, a cross-coherence coefficient is calculated to characterize the sparse reconstruction accuracy; the cross-coherence coefficient is used to measure the degree of linear correlation between the responses of each potential sensor location point in the initial measurement matrix;

[0169] An iterative optimization algorithm is used to screen potential sensor locations with the goal of minimizing the mutual coherence coefficient, in order to determine an optimal combination of sensor locations that makes the mutual coherence coefficient reach a preset convergence condition.

[0170] Multiple sensors are arranged on the surface of the antenna structure under test according to the optimal sensor position combination to form the sparse sensor array.

[0171] In practical implementation, to ensure the acquisition of high-quality vibration response data using a limited number of sensors for accurate sparse reconstruction, the arrangement process of the sparse sensor array includes the following steps:

[0172] First, based on the finite element model of the antenna structure under test constructed in step S101, multiple potential sensor location points are defined on the surface of the structure. Specifically, these potential location points are usually nodes on the finite element model mesh or specific discrete sampling locations. The number of location points is large to ensure that the possible area of ​​the structure's global vibration response is covered.

[0173] Subsequently, using the mode shape dictionary constructed in step S101, the response values ​​of each basic mode shape function in the dictionary at all the aforementioned potential sensor locations are calculated to construct an initial measurement matrix. Specifically, the response of each basic mode shape function at each potential sensor location is formed into a column vector, and these column vectors constitute the initial measurement matrix to characterize the vibration response features at different locations.

[0174] Next, based on the initial measurement matrix and mode shape dictionary constructed above, the mutual coherence coefficient, used to characterize the sparse reconstruction accuracy, is calculated. Specifically, the mutual coherence coefficient is defined as the maximum absolute inner product between the columns of the initial measurement matrix, reflecting the degree of linear correlation between the responses of different potential sensor locations. The calculation formula is as follows:

[0175]

[0176] in, The coefficient represents the mutual coherence coefficient, which measures the maximum linear correlation between the response signals of different potential sensor locations in the measurement matrix. The smaller the value, the lower the correlation of the signals, and the more conducive it is to high-precision sparse reconstruction.

[0177] , This represents the column vectors in the initial measurement matrix, where each column vector represents the response data at a specific potential sensor location; specifically, the column vectors... The vector corresponding to the response of the i-th potential sensor location point under all mode shape functions in the mode shape dictionary;

[0178] Represents column vectors and The inner product between the two points, the absolute value of which reflects the similarity between the response data of location point i and location point j. The larger the inner product value, the more similar the responses of the two location points are, and the higher the correlation.

[0179] , Representing column vectors respectively , of The norm, or Euclidean norm, is used to normalize the inner product to ensure that the calculation of the coherence coefficient is not affected by the magnitude of the vibration response amplitude.

[0180] This represents the combination of taking all distinct positions, i.e., any two distinct positions i and j that satisfy... The maximum value among the normalized inner product values ​​of all these combinations is selected as the final calculation result of the mutual coherence coefficient.

[0181] To determine the optimal sensor locations, an iterative optimization algorithm is employed, with the goal of minimizing the cross-coherence coefficient, to iteratively screen and optimize the potential sensor locations. Specific implementation methods may include greedy algorithms such as forward selection and greedy pruning, evolutionary algorithms such as genetic algorithms, and other optimization strategies.

[0182] For example, when the forward selection algorithm is used as the iterative optimization algorithm, the implementation process is as follows:

[0183] Initialize an empty set of optimal sensor locations;

[0184] In each iteration, a location point is selected from the remaining potential location points one by one, and the coherence coefficient of the measurement matrix obtained after adding the location point to the optimal location set is calculated.

[0185] Select potential location points that maximize the reduction in the mutual coherence coefficient and add them to the optimal location set;

[0186] Repeat the above steps until the coherence coefficients reach the preset convergence conditions, such as the coefficient value being lower than the set threshold, or it being unable to be further reduced, or the preset limit on the number of sensors is reached.

[0187] Once the optimization algorithm reaches the aforementioned preset convergence condition, it will eventually obtain an optimal combination of sensor positions that minimizes the mutual coherence coefficient.

[0188] Finally, based on the optimal sensor placement combination determined above, multiple sensors are physically arranged at corresponding positions on the surface of the antenna structure under test, thus forming the final sparse sensor array. In practical implementation, measuring devices such as accelerometers, strain sensors, or laser vibration sensors can be selected and precisely installed at the corresponding positions to achieve high-quality acquisition of vibration response data of the antenna structure under test under actual operating conditions.

[0189] This effectively optimizes the sensor placement, significantly reduces the number of sensors, and ensures the accuracy and robustness of sparse reconstruction of vibration data, providing efficient and reliable data support for structural damage diagnosis methods.

[0190] As an optional implementation, the pre-construction of the probabilistic observation model based on the Gaussian process regression method includes:

[0191] The response relationship is decomposed into a global smoothing trend term characterizing the influence of ambient temperature and a local nonlinear effect term characterizing the influence of structural damage.

[0192] For the global smoothing trend term, a first Gaussian kernel function is constructed, and the influence weight of the ambient temperature input dimension on the trend term is determined using a first automatic association and discrimination mechanism;

[0193] For the local nonlinear effect term, a second Gaussian kernel function is constructed, and a second automatic association discrimination mechanism is used to determine the influence weight of the structural damage state input dimension on the effect term; wherein, the first Gaussian kernel function and the second Gaussian kernel function are of different types;

[0194] The first Gaussian kernel function and the second Gaussian kernel function are summed to form a composite Gaussian kernel function that can decouple different physical influences, and the probabilistic observation model is constructed based on the composite Gaussian kernel function.

[0195] In practice, the probabilistic observation model is pre-constructed using the Gaussian Process Regression (GPR) method to describe the response relationship between the vibration modal parameters of the antenna structure under test and the ambient temperature and structural damage state. Specific implementation steps include:

[0196] First, the vibration modal parameter response relationship of the antenna structure under test is decomposed into two independent and complementary components: a global smoothing trend term, which mainly reflects the overall linear or slowly varying nonlinear influence of ambient temperature on the structural modal parameters; and a local nonlinear effect term, which mainly characterizes the complex local nonlinear disturbance effect caused by structural damage.

[0197] Specifically, the global smoothing trend term can be assumed to be a slow, smooth global change trend of modal parameters under changes in ambient temperature. To effectively characterize this trend, a first Gaussian kernel function with strong smoothing properties can be selected, such as the common squared exponential (SE) kernel function or radial basis function (RBF) kernel function, which are continuous and infinitely differentiable. To determine the degree of influence of each input dimension of ambient temperature on the global trend, this embodiment further employs a first automatic relevance determination (ARD) mechanism. Specifically, this mechanism sets independent feature scale parameters or weight parameters for each input dimension of the kernel function, and automatically adjusts and learns these parameters during training. This allows for the automatic identification and determination of the significant contribution of each dimension of ambient temperature variables to the global trend term, retaining input dimensions with significant contributions and suppressing or removing redundant or less contributing input dimensions, thereby improving the model's prediction accuracy and generalization ability.

[0198] Automatic association discrimination is a technique in Bayesian machine learning. Its core idea is to set independent hyperparameters for each input dimension of the model and automatically learn these parameters during model training. The magnitude of the parameter values ​​directly reflects the importance or correlation of the corresponding input dimension, thereby achieving automatic identification and suppression of redundant features, and realizing the purpose of feature selection and improving the model's generalization ability.

[0199] The first / second automatic association discrimination mechanism is a specific implementation of the above-mentioned ARD prior art applied in this application.

[0200] The first automatic association discrimination mechanism specifically refers to using ARD technology for the first Gaussian kernel function to determine the influence weight of each input dimension of ambient temperature on the global trend term; the second automatic association discrimination mechanism specifically refers to using ARD technology for the second Gaussian kernel function to determine the significant contribution of each input dimension of structural damage state to the local nonlinear effect term.

[0201] For local nonlinear effects, they typically manifest as complex local nonlinear responses caused by structural damage, exhibiting distinct local characteristics in their spatial distribution. To effectively capture these characteristics, this embodiment constructs a second Gaussian kernel function, different from the first Gaussian kernel function. For example, the second Gaussian kernel function can be selected from kernel functions with local properties and spatial variability, such as the Matrn kernel function or the compactly supported kernel function, to effectively reflect the spatial local characteristics of the damage effect on the structural modal parameters.

[0202] Similarly, for this local nonlinear effect term, a second automatic association discrimination mechanism is also employed to determine the significant contribution of each input dimension of the structural damage state to the local nonlinear effect. Through automatic adjustment and optimization during the training process, it is possible to accurately determine and identify which damage state input dimensions have a higher influence weight on the local nonlinear effect, thereby achieving effective screening of input dimensions and accurate extraction of damage-sensitive features.

[0203] Furthermore, to better utilize the two Gaussian kernel functions with different characteristics to jointly model the complex response behavior of vibration modal parameters, this embodiment linearly sums and combines the first and second Gaussian kernel functions to form a composite Gaussian kernel. Specifically, the two Gaussian kernel functions are expressed as a weighted sum:

[0204]

[0205] in:

[0206] For input variables based on ambient temperature The global smoothing trend term kernel function;

[0207] Input variables based on structural damage state The kernel function of the local nonlinear effect term;

[0208] This represents the overall vector of input variables, which includes all input dimensions for ambient temperature and structural damage state.

[0209] By using the aforementioned composite Gaussian kernel function, the probabilistic observation model can significantly improve the ability to model complex response relationships of modal parameters. In particular, it can effectively decouple and identify the respective contributions of environmental temperature and structural damage to modal parameters, thereby enhancing the robustness and accuracy of structural damage diagnosis.

[0210] Finally, using the composite Gaussian kernel function developed above, a Gaussian process regression model is trained on a large amount of structural modal parameter sample data under various temperature and damage conditions. This estimates the hyperparameters of the Gaussian process, including the weight parameters of each input dimension and the noise parameters of the Gaussian process regression. The trained probabilistic observation model can then be used for high-precision prediction of modal parameters in the damage diagnosis method for the antenna under test structure. This provides a reliable likelihood function calculation basis for the particle filter algorithm, significantly improving the overall performance of the damage diagnosis method for the antenna under test structure.

[0211] As an optional implementation, determining the damage information of the antenna structure under test based on the updated particle weights includes:

[0212] A density clustering algorithm is used to cluster the particle set updated according to the particle weight in the damage information state space to determine at least one high probability density particle cluster.

[0213] Calculate the statistical center of the particle state for each of the at least one high probability density particle clusters to obtain at least one candidate damage information;

[0214] The at least one candidate damage information and the current ambient temperature data are input into the probabilistic observation model to calculate and obtain the prediction confidence level corresponding to each candidate damage information; the prediction confidence level is used to characterize the reliability of the candidate damage information under the current ambient temperature conditions.

[0215] The candidate damage information with the highest prediction confidence is selected from the at least one candidate damage information and determined as the final damage information of the antenna structure under test.

[0216] When determining the damage information of the antenna structure under test based on the particle weights updated by the particle filtering algorithm, this application first performs density clustering analysis of the particle set to further accurately identify the most likely damage state from a large number of damage information hypotheses.

[0217] In the specific implementation, all particles after weight update in step S103 are first used as input data. These particles explicitly represent the structural damage information hypothesis in the form of their state vectors, including information such as damage location, damage degree, and damage type. Since the particle set contains a large number of particles, density clustering algorithm is used to automatically cluster the state space in order to efficiently identify the most representative and probable damage states.

[0218] Specifically, density-based spatial clustering algorithms can be selected as density clustering methods, and their main implementation process includes:

[0219] First, a distance measurement criterion is defined between particles in the damage state space. In this embodiment, the particle state vector can be represented as a point in a multi-dimensional space, where each dimension represents different damage information attributes, such as position coordinates, percentage reduction in stiffness, or modal strain energy variation. Euclidean distance or weighted Euclidean distance is used as the distance measurement standard between particles to reflect the degree of difference between damage states.

[0220] Next, we define two key parameters required for the DBSCAN algorithm: the distance threshold (eps) and the minimum particle count threshold (MinPts). Specifically, the distance threshold defines the neighborhood radius of a particle cluster. When the number of particles within this threshold radius exceeds the preset minimum particle count threshold, the particle is considered a density center or core particle; otherwise, it is a boundary particle or noise particle. These two parameters can be determined through preliminary simulation analysis, empirical value selection, or based on the density distribution characteristics of the particle ensemble's state space, to effectively identify particle cluster regions that may correspond to the actual damage state.

[0221] During clustering, the algorithm arbitrarily selects a particle from the particle set and determines whether the number of particles in its neighborhood meets the density requirement. If the requirement is met, the particle is marked as the core particle of a new cluster, and the cluster is recursively expanded from other particles in its neighborhood, continuously incorporating other particles that meet the density condition, thus forming a complete particle cluster. If the requirement is not met, the particle is temporarily marked as noise or a boundary particle, and the above process is repeated for other unmarked particles until all particles have been processed. Through this process, one or more high-probability-density particle clusters can be automatically identified from the particle set, significantly improving the clarity of damage state information.

[0222] After cluster analysis, for each identified high-probability-density particle cluster, the state statistical center of the particles within the cluster is further determined to form at least one candidate damage information state. Specifically, for each particle cluster, the state vectors of each particle within the cluster are weighted using particle weights as weighting coefficients to calculate a weighted average, thus obtaining the state statistical center of the particle cluster. This state statistical center represents the typical state of the particle cluster in the damage state space, clearly characterizing the possible range of values ​​for the damage location, type, and degree pointed to by the particle cluster, thereby constituting a candidate damage state.

[0223] After obtaining at least one candidate damage information state, these candidate damage information states, along with the currently synchronously acquired ambient temperature data, are further input into the pre-constructed probabilistic observation model in step S101 to calculate the prediction confidence level corresponding to each candidate damage information state. In specific implementation, the probabilistic observation model will utilize the response relationship between the ambient temperature and damage state input variables to the modal parameter output variables obtained through training to calculate the predicted modal parameters under each candidate damage information state, such as modal frequencies and mode shapes.

[0224] Subsequently, the modal parameters predicted by the probabilistic observation model are compared one by one with the actual measured modal parameters obtained from the antenna structure under test in step S102, and the difference index between the two is calculated, such as frequency error or mode guarantee criterion (MAC) value. Further, the above difference index is transformed into a numerical prediction confidence level using a Gaussian likelihood function or other suitable probability density function to quantitatively characterize the matching degree and accuracy of each candidate damage information under the current ambient temperature conditions.

[0225] Finally, from the multiple candidate damage information, the candidate damage state with the highest prediction confidence is selected as the final damage information of the antenna structure under test. Specifically, the damage information with the highest prediction confidence is the damage location and degree that best matches the actual vibration data and ambient temperature conditions under the current operating conditions and is most likely to be accurate.

[0226] Through the meticulous implementation of multiple steps, including density clustering analysis, state statistical center calculation, probability model prediction, and prediction confidence assessment, this implementation method significantly improves the accuracy and stability of damage diagnosis results, making the damage diagnosis of the antenna structure under test more accurate and reliable in practical engineering applications.

[0227] Based on the same inventive concept, this application also provides a base station antenna reliability diagnosis system corresponding to a base station antenna reliability diagnosis method. Since the principle of the system in this application is similar to the base station antenna reliability diagnosis method described above, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be described again.

[0228] Reference Figure 4 The diagram shown is a schematic of a base station antenna reliability diagnostic system provided in this application. The system includes:

[0229] The acquisition module 10 is used to acquire the finite element model of the antenna structure under test, perform modal analysis on the finite element model, generate multiple basic mode shape functions, and construct a mode shape dictionary based on the multiple basic mode shape functions; it also pre-constructs a probabilistic observation model based on the Gaussian process regression method, which characterizes the response relationship between the vibration modal parameters of the antenna structure under test and the corresponding ambient temperature and structural damage state; and it collects vibration response data using a sparse sensor array arranged on the surface of the antenna structure under test, and simultaneously collects the corresponding ambient temperature data.

[0230] The first processing module 20 is used to process the vibration response data collected by the sparse sensor array based on the mode shape dictionary and using a sparse reconstruction algorithm to generate full-field mode shape data of the antenna structure under test; and to perform mode recognition operation based on the full-field mode shape data to extract the current measured mode parameters of the antenna structure under test.

[0231] The second processing module 30 is used to initialize multiple particles, each particle's state corresponding to a damage information hypothesis; based on the particle filtering algorithm, the current measured modal parameters, the ambient temperature data, and the damage information hypothesis of each particle are input into the probabilistic observation model to calculate the posterior probability value of each particle, and the weight of each particle is updated based on the posterior probability value.

[0232] The diagnostic module 40 is used to determine the damage information of the antenna structure under test based on the updated particle weights; the damage information includes damage location information and damage degree information.

[0233] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

Claims

1. A method for diagnosing the reliability of a base station antenna, characterized in that, include: Initial testing is performed on the antenna under test to determine its baseline performance state before environmental stress is applied. The baseline performance state includes appearance, mechanical properties, and electrical properties. The antenna under test is placed under preset environmental test conditions, and environmental stress is applied for a preset duration. After the environmental stress is applied, the antenna under test is subjected to a final test to obtain the final performance status of the antenna under test. The final performance state is compared with the baseline performance state, and the reliability information of the antenna is obtained according to a preset criterion. Also includes: A finite element model of the antenna structure under test is obtained, and modal analysis is performed on the finite element model to generate multiple basic mode shape functions. A mode shape dictionary is constructed based on the multiple basic mode shape functions. A probabilistic observation model is constructed in advance based on the Gaussian process regression method. The probabilistic observation model characterizes the response relationship between the vibration modal parameters of the antenna structure under test and the corresponding ambient temperature and structural damage state. Vibration response data is collected using a sparse sensor array arranged on the surface of the antenna structure under test, and corresponding ambient temperature data is collected simultaneously. Based on the mode shape dictionary, a sparse reconstruction algorithm is used to process the vibration response data collected by the sparse sensor array to generate full-field mode shape data of the antenna structure under test; based on the full-field mode shape data, a modal identification operation is performed to extract the current measured modal parameters of the antenna structure under test. Multiple particles are initialized, and the state of each particle corresponds to a damage information hypothesis. Based on the particle filtering algorithm, the current measured modal parameters, the ambient temperature data, and the damage information hypothesis of each particle are input into the probabilistic observation model to calculate the posterior probability value of each particle, and the weight of each particle is updated based on the posterior probability value. The damage information of the antenna structure under test is determined based on the updated particle weights; the damage information includes damage location information and damage degree information.

2. The base station antenna reliability diagnosis method according to claim 1, characterized in that, The environmental stress is random vibration stress, and the environmental stress applied for a preset duration includes: Random vibrations following a predetermined acceleration spectral density curve are sequentially applied to the antenna under test along multiple preset axial directions.

3. The base station antenna reliability diagnosis method according to claim 1, characterized in that, The step of determining the damage information of the antenna structure under test based on the updated particle weights includes: The effective sample size of the plurality of particles is calculated based on the updated particle weights; the effective sample size is determined based on the weight values ​​of the plurality of particles. The effective sample size is compared with a preset particle set degradation threshold to determine whether the particle set has degraded. In response to the effective sample size being less than the particle set degradation threshold, a regularized resampling operation is performed on the plurality of particles; The damage information of the antenna structure under test is determined based on the particle state distribution of the descendant particle set after regularization resampling.

4. The base station antenna reliability diagnosis method according to claim 3, characterized in that, The regularized resampling includes: Based on the updated particle weights, a descendant particle set is generated by extracting particles with replacement from the plurality of particles. Calculate the particle state covariance of the descendant particle set, and adaptively determine the bandwidth of the Gaussian kernel function based on the particle state covariance; The descendant particle set is smoothed based on the Gaussian kernel function after the bandwidth is determined, so as to apply a small perturbation to the particle state in the descendant particle set and obtain a descendant particle set with enhanced diversity.

5. The base station antenna reliability diagnosis method according to claim 4, characterized in that, Smoothing the descendant particle set based on the Gaussian kernel function with the determined bandwidth includes: For each descendant particle in the descendant particle set, the corresponding damage location is extracted from the damage information of the descendant particle. Based on the damage location, local physical property parameters of the region corresponding to the damage location are extracted from the finite element model; the local physical property parameters specifically include the local stiffness matrix or modal strain energy density of the region. An anisotropic Gaussian perturbation kernel function is constructed for each descendant particle based on the local physical property parameters; specifically, the covariance matrix of the anisotropic Gaussian perturbation kernel function is calculated and generated based on the local physical property parameters. The state of the descendant particles is subjected to a small perturbation using the anisotropic Gaussian perturbation kernel function to generate a diverse set of descendant particles that matches the physical characteristics of the antenna structure under test.

6. The base station antenna reliability diagnosis method according to claim 1, characterized in that, Generating the full-field mode shape data of the antenna structure under test includes: Based on the ambient temperature data and the damage information determined by particle filtering at the previous moment, the ambient temperature data and the damage information are input into the probabilistic observation model to generate a prior weight vector; wherein, each element of the prior weight vector represents the activation probability of the corresponding basic mode function in the mode dictionary under the current working condition. The prior weight vector is used to weight each basic mode function in the mode shape dictionary to form a weighted mode shape dictionary; Based on the weighted mode shape dictionary, the weighted L1 norm minimization algorithm is used to solve the vibration response data collected by the sparse sensor array to determine the mode shape coefficients used to reconstruct the full-field mode shape data; Based on the mode shape coefficients and the weighted mode shape dictionary, the full-field mode shape data of the antenna structure under test are reconstructed.

7. The base station antenna reliability diagnosis method according to claim 1, characterized in that, The acquisition of vibration response data using a sparse sensor array arranged on the surface of the antenna structure under test includes: Based on the finite element model, multiple potential sensor location points are defined, and an initial measurement matrix is ​​constructed using the response values ​​of the basic mode functions in the mode dictionary at all potential sensor location points. Based on the initial measurement matrix and the mode shape dictionary, a cross-coherence coefficient is calculated to characterize the sparse reconstruction accuracy; the cross-coherence coefficient is used to measure the degree of linear correlation between the responses of each potential sensor location point in the initial measurement matrix; An iterative optimization algorithm is used to screen potential sensor locations with the goal of minimizing the mutual coherence coefficient, in order to determine an optimal combination of sensor locations that makes the mutual coherence coefficient reach a preset convergence condition. Multiple sensors are arranged on the surface of the antenna structure under test according to the optimal sensor position combination to form the sparse sensor array.

8. The base station antenna reliability diagnosis method according to claim 1, characterized in that, The pre-construction of the probabilistic observation model based on the Gaussian process regression method includes: The response relationship is decomposed into a global smoothing trend term characterizing the influence of ambient temperature and a local nonlinear effect term characterizing the influence of structural damage. For the global smoothing trend term, a first Gaussian kernel function is constructed, and the influence weight of the ambient temperature input dimension on the trend term is determined using a first automatic association and discrimination mechanism; For the local nonlinear effect term, a second Gaussian kernel function is constructed, and a second automatic association discrimination mechanism is used to determine the influence weight of the structural damage state input dimension on the effect term; wherein, the first Gaussian kernel function and the second Gaussian kernel function are of different types; The first Gaussian kernel function and the second Gaussian kernel function are summed to form a composite Gaussian kernel function that can decouple different physical influences, and the probabilistic observation model is constructed based on the composite Gaussian kernel function.

Images