Bolt loosening positioning method and system based on environmental vibration and multi-scale pnn

Abstract

This invention discloses a bolt loosening location method and system based on environmental vibration and multi-scale PNN. The location method includes the following steps: S1: Establishing a refined finite element simulation model of the transmission tower and dividing it into multi-scale hierarchical substructures; S2: Simulating bolt loosening conditions in the finite element simulation model and extracting operational modal analysis results; S3: Calculating the frequency change ratio and modal compliance residual matrix norm, and constructing a multi-source damage-sensitive feature vector after weighted fusion; S4: Constructing and training a probabilistic neural network with an adaptive smoothing factor adjustment mechanism; S5: Acquiring on-site environmental vibration data and identifying modal parameters; S6: Extracting and calculating the measured data of the tower using the random subspace identification method, and inputting it into the trained probabilistic neural network to obtain the location results of the main bolt loosening area. This method achieves accurate bolt loosening location through scientific substructure division and multi-source fusion strategies, providing a reference for the safe operation and maintenance of tower bolts.

Description

Technical Field

[0001] This invention relates to the field of structural health monitoring and damage identification technology for transmission line towers, and in particular to a bolt loosening location method and system based on environmental vibration and multi-scale PNN. Background Technology

[0002] Bolted connections, as the core weakest link in the structure of transmission line towers, bear the functions of connecting components and transmitting forces. Their health directly determines the stability and safety of the tower structure. Transmission towers operate for extended periods in complex outdoor environments, enduring multiple forces such as wind, rain, strong wind vibrations, diurnal temperature variations, atmospheric corrosion, and line loads. This makes them highly susceptible to damage such as loosening, fatigue, and even detachment of bolts. If the damaged locations are not located promptly and accurately, uneven stress distribution and localized stress concentration will gradually occur in the tower components. In severe cases, this can lead to tower tilting and collapse, causing transmission line interruptions and resulting in significant economic losses and safety hazards to the power system.

[0003] Currently, the main methods for detecting loose bolts on power transmission towers are as follows:

[0004] (1) Manual inspection method. This method relies on maintenance personnel to climb the tower or use binoculars to visually inspect whether the bolts are loose. This method is labor-intensive, inefficient, and limited by factors such as terrain and weather, resulting in a high rate of missed inspections.

[0005] (2) Image recognition-based detection methods. For example, using a drone equipped with a camera to take images of tower nodes and using image processing technology to identify loose bolts. However, this type of method can only detect bolts that are visible in appearance, and it is difficult to effectively cover nodes that are high up or obstructed, and it is easily affected by factors such as lighting and angle.

[0006] (3) Detection method based on contact sensors. Such as installing strain gauges or piezoelectric sensors on bolts to directly measure changes in bolt preload. This type of method requires installing sensors on each bolt, which is costly to implement in large-scale iron towers, and the sensors themselves have long-term reliability and maintenance issues.

[0007] (4) Detection methods based on active excitation and vibration signal analysis. These methods require active excitation or rely on specific vibration events. Under natural environmental excitation, the signal-to-noise ratio is low, and the stability and noise resistance of feature extraction need to be improved.

[0008] In summary, existing technologies have the following shortcomings: ① Some methods require installing sensors on each bolt or performing contact-based detection, resulting in high implementation costs and maintenance difficulties; ② Some methods rely on manual inspection or image recognition, leading to limited coverage and a high rate of missed detections; ③ Some vibration analysis-based methods require active excitation or can only determine whether a bolt is loose but cannot accurately locate the loose area. Therefore, there is an urgent need to develop an online diagnostic method that can accurately locate the loose bolt area under non-power-off, non-contact conditions. Summary of the Invention

[0009] To overcome the above problems, the purpose of this invention is to provide a bolt loosening location method and system based on environmental vibration and multi-scale PNN. The location method achieves accurate bolt loosening location through scientific substructure division, reasonable parameter setting, finite element working condition simulation and probabilistic neural network training and testing, providing technical reference for the safe operation and maintenance of tower bolts.

[0010] The technical solution adopted in this invention is:

[0011] A bolt loosening location method based on environmental vibration and multi-scale PNN includes the following steps:

[0012] S1: Establish a refined finite element simulation model of the power transmission tower and divide it into multi-scale hierarchical substructures.

[0013] S2: In the finite element simulation model generated in S1, the bolt loosening condition is simulated, and the operational modal analysis results of the simulation model under environmental excitation are extracted.

[0014] S3: Based on the results of the operational modal analysis, calculate the frequency change ratio between the first and fourth orders and the norm of the modal compliance residual matrix, and construct a multi-source damage sensitive feature vector after weighted fusion.

[0015] S4: Construct and train a probabilistic neural network with an adaptive smoothing factor adjustment mechanism.

[0016] S5: On-site environmental vibration data acquisition and modal parameter identification.

[0017] S6: Extract the first four natural frequencies and corresponding mode shapes of the tower from the data collected in S5 using the random subspace recognition method. Calculate the measured frequency change ratio and the norm of the measured modal compliance residual matrix. After normalization, input the results into the probabilistic neural network trained in step S4 to obtain the location results of the secondary substructure area where the main bolts are loose.

[0018] As a further description of the present invention, the principle of multi-scale hierarchical substructure partitioning in S1 is as follows:

[0019] Based on the stress hierarchy of the tower structure and the distribution characteristics of the main bolts, the main material of the tower is divided into multi-scale hierarchical substructures, forming N primary substructures. Each primary substructure is further divided into M secondary substructures, and a unique pattern category label is assigned to each secondary substructure.

[0020] The specific process of multi-scale hierarchical substructure partitioning in S1 is as follows:

[0021] S11: Based on the stress characteristics of the tower legs, tower body, and crossarms, the main materials are divided into 7 to 15 primary substructures, each of which has a high degree of axial symmetry.

[0022] S12: For each primary substructure, it is divided into 3 to 5 secondary substructures along the main material direction according to the segment length between two flange connection nodes. Each secondary substructure contains all main material angle steel components and their connecting bolt groups in the same height plane.

[0023] As a further description of the present invention, the construction formula of the multi-source damage-sensitive feature vector D in S3 is as follows:

[0024] .

[0025] in, , and These represent the first and fourth natural frequencies, respectively, with superscripts. Indicates the state of damage, superscript It indicates that the item is in good condition.

[0026] , representing the good state compliance matrix With the compliance matrix of the damage state The Frobenius norm of the difference, the compliance matrix is ​​constructed from the measured or simulated first four frequencies and mass-normalized mode shapes.

[0027] and For the weighting coefficients, satisfying ,and The value range is 0.4 to 0.6.

[0028] As a further description of the present invention, the construction and training process of the probabilistic neural network in S4 is as follows:

[0029] S41: The multi-source damage-sensitive feature vectors obtained in S3 are normalized and divided into a training set and a test set according to a preset ratio, with the training set accounting for 70%~80% and the test set accounting for 20%~30%.

[0030] S42: Construct a probabilistic neural network, which includes an input layer, a radial base layer, a summation layer and an output layer. The input layer receives the multi-source damage-sensitive feature vector, and the output layer outputs the posterior probability of loosening of each secondary substructure. The secondary substructure corresponding to the largest posterior probability is taken as the preliminary localization result of the loosening region.

[0031] S43: Substitute the training set into the probabilistic neural network for training, use the adaptive smoothing factor adjustment mechanism to determine the smoothing factor for each output category, and after training, input the test set into the trained network for prediction verification.

[0032] S44: The training termination condition is that the region localization accuracy on the test set is not less than 95%. If it is not met, the smoothing factor parameter is adjusted or training samples are added and retraining is performed until the termination condition is met.

[0033] As a further description of the present invention, the adaptive smoothing factor adjustment mechanism in S4 is as follows:

[0034] .

[0035] in, For the first Smoothing factors corresponding to each output category.

[0036] It is the basic smoothing factor, with a value range of 0.1 to 0.3.

[0037] This is an adjustment coefficient, with a value ranging from 0.5 to 1.5.

[0038] For the training set The standard deviation of the feature vectors of each category of samples.

[0039] For the training set The mean of the feature vectors of samples in each category.

[0040] As a further description of the present invention, in S2, the stiffness of the bolt joint connection is reduced to simulate different bolt loosening conditions. The specific simulation method is as follows:

[0041] In the finite element model, the bolted connection is equivalent to a spring-damped element with stiffness in six directions. By randomly selecting values ​​within the range of 10% to 90% of the original design value to reduce the normal contact stiffness, different degrees of bolt loosening are simulated.

[0042] At least 50 sets of simulation samples were generated for each secondary substructure, including single bolt loosening, multi-bolt combination loosening, and different degrees of loosening.

[0043] As a further description of the present invention, the on-site environmental vibration data acquisition method in S5 is to install environmental vibration sensors on the transmission tower to be tested and collect the acceleration time history response data of the tower under wind load or ground pulsation environmental excitation.

[0044] As a further description of the present invention, the environmental vibration sensor is a low-frequency piezoelectric accelerometer or MEMS accelerometer, which is installed at the main material nodes of each primary substructure of the tower, with a measured data acquisition time of not less than 5 minutes and a sampling frequency of not less than 200Hz.

[0045] As a further description of the present invention, the operational modal analysis results in S3 include the first four natural frequency values ​​and the mode displacements at key nodes of the tower body.

[0046] A system for locating loose bolt areas on the main components of a power transmission tower, used to implement the aforementioned bolt loosening location method based on environmental vibration and multi-scale PNN, the system comprising:

[0047] The data acquisition and transmission module includes environmental vibration sensors, a data acquisition instrument, and a wireless transmission module deployed at the nodes of the main structure of the tower, used to acquire the acceleration response under environmental excitation in real time.

[0048] The simulation sample generation module includes a finite element model unit for the iron tower, a simulation unit for bolt loosening conditions, and a feature extraction unit. It is used to generate a multi-source damage-sensitive feature vector sample library corresponding to the loosening of different substructures.

[0049] The model training and recognition module includes a normalization processing unit, an adaptive smoothing factor probabilistic neural network training unit, and a measured data feature extraction unit. It is used to train the loosened region recognition model and classify and recognize the measured features, and output the location labels of the loosened secondary substructures.

[0050] The beneficial effects of this invention are:

[0051] This invention presents a bolt loosening location method based on environmental vibration and multi-scale PNN. It utilizes the acceleration response data of transmission towers under natural environmental excitations such as wind loads and ground pulsations to identify modal parameters. This method eliminates the need for manual tower climbing, active vibration equipment, and bolt-by-bolt contact sensor installation. Data acquisition can be completed by deploying only a small number of acceleration sensors at key nodes of the tower's main structure, achieving non-contact, uninterrupted online monitoring. Compared to methods based on impact elastic waves (requiring electromagnetic excitation devices), this invention significantly reduces hardware deployment costs and is more suitable for large-scale application. Compared to methods based on fractional Fourier transforms, this invention directly utilizes environmental excitation responses, without relying on specific vibration events, thus offering broader applicability.

[0052] This invention presents a bolt loosening location method based on environmental vibration and multi-scale Probabilistic Neural Network (PNN). It employs a multi-scale hierarchical partitioning strategy of "first-level substructure – second-level substructure." First, based on the stress characteristics of the tower legs, tower body, and crossarms, the main structure is divided into 7-15 first-level substructures. Each first-level substructure is further subdivided into 3-5 second-level substructures along the main structure direction according to segment length. This partitioning method considers both the stress hierarchy characteristics of the tower structure and the distribution pattern of bolt connection nodes, ensuring that each second-level substructure has relatively independent and identifiable modal response characteristics. The multi-scale partitioning strategy of this invention provides clearer hierarchy and finer classification granularity, effectively balancing positioning accuracy with the classification complexity of the PNN network (probabilistic neural network).

[0053] This invention presents a bolt loosening location method based on environmental vibration and multi-scale PNN. The constructed feature vector integrates two complementary damage-sensitive indices: first, the ratio of first-order to fourth-order frequency changes, reflecting the differentiated impact of overall stiffness changes on lower-order frequencies; second, the Frobenius norm of the modal compliance matrix residuals, reflecting the local variation characteristics of the structural compliance distribution. These two indices characterize the structural stiffness degradation caused by bolt loosening from different dimensions. Weighted fusion effectively reduces the risk of individual indices being affected by environmental noise, improving identification robustness. The multi-source fusion strategy of this method exhibits higher identification stability under complex environmental conditions. Attached Figure Description

[0054] Figure 1 This is a flowchart of the bolt loosening location method based on environmental vibration and multi-scale PNN proposed in this invention.

[0055] Figure 2 This is a schematic diagram of the substructure division of the iron tower in Embodiment 3 of the bolt loosening location method and system based on environmental vibration and multi-scale PNN proposed in this invention. Detailed Implementation

[0056] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0057] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0058] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0059] This invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of this invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not adhering to the usual scale. Furthermore, the schematic diagrams are merely examples and should not be construed as limiting the scope of protection of this invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.

[0060] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used solely for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0061] Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" in this invention should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; similarly, they can refer to mechanical connections, electrical connections, or direct connections, or indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0062] like Figures 1-2 As shown, it illustrates a specific embodiment of the present invention:

[0063] Example 1:

[0064] A bolt loosening location method based on environmental vibration and multi-scale PNN includes the following steps:

[0065] S1: Establish a refined finite element simulation model of the power transmission tower and divide it into multi-scale hierarchical substructures.

[0066] Specifically, the principle of multi-scale hierarchical substructure partitioning in S1 is as follows:

[0067] Based on the stress hierarchy of the tower structure and the distribution characteristics of the main bolts, the main material of the tower is divided into multi-scale hierarchical substructures, forming N primary substructures. Each primary substructure is further divided into M secondary substructures, and a unique pattern category label is assigned to each secondary substructure.

[0068] The specific process of multi-scale hierarchical substructure partitioning in S1 is as follows:

[0069] S11: Based on the stress characteristics of the tower legs, tower body, and crossarms, the main materials are divided into 7 to 15 primary substructures, each of which has a high degree of axial symmetry.

[0070] S12: For each primary substructure, it is divided into 3 to 5 secondary substructures along the main material direction according to the segment length between two flange connection nodes. Each secondary substructure contains all main material angle steel components and their connecting bolt groups in the same height plane.

[0071] S2: In the finite element simulation model generated in S1, the bolt loosening condition is simulated, and the operational modal analysis results of the simulation model under environmental excitation are extracted.

[0072] Specifically, in S2, the stiffness of the bolted joint connection is reduced to simulate different bolt loosening conditions. The specific simulation method is as follows:

[0073] In the finite element model, the bolted connection is equivalent to a spring-damped element with stiffness in six directions. By randomly selecting values ​​within the range of 10% to 90% of the original design value to reduce the normal contact stiffness, different degrees of bolt loosening are simulated.

[0074] At least 50 sets of simulation samples were generated for each secondary substructure, including single bolt loosening, multi-bolt combination loosening, and different degrees of loosening.

[0075] S3: Based on the results of the operational modal analysis, calculate the frequency change ratio between the first and fourth orders and the norm of the modal compliance residual matrix, and construct a multi-source damage sensitive feature vector after weighted fusion.

[0076] Specifically, the construction formula for the multi-source damage-sensitive feature vector D in S3 is as follows:

[0077] .

[0078] in, , and These represent the first and fourth natural frequencies, respectively, with superscripts. Indicates the state of damage, superscript It indicates that the item is in good condition.

[0079] , representing the integrity state compliance matrix With the damage state compliance matrix The Frobenius norm of the difference, the compliance matrix is ​​constructed from the measured or simulated first four frequencies and mass-normalized mode shapes.

[0080] and For the weighting coefficients, satisfying ,and The value range is 0.4 to 0.6.

[0081] Specifically, the operational modal analysis results in S3 include the first four natural frequency values ​​and the mode displacements at key nodes of the tower.

[0082] S4: Construct and train a probabilistic neural network with an adaptive smoothing factor adjustment mechanism.

[0083] Specifically, the construction and training process of the probabilistic neural network in S4 is as follows:

[0084] S41: The multi-source damage-sensitive feature vectors obtained in S3 are normalized and divided into a training set and a test set according to a preset ratio, with the training set accounting for 70%~80% and the test set accounting for 20%~30%.

[0085] S42: Construct a probabilistic neural network, which includes an input layer, a radial base layer, a summation layer and an output layer. The input layer receives the multi-source damage-sensitive feature vector, and the output layer outputs the posterior probability of loosening of each secondary substructure. The secondary substructure corresponding to the largest posterior probability is taken as the preliminary localization result of the loosening region.

[0086] S43: Substitute the training set into the probabilistic neural network for training, use the adaptive smoothing factor adjustment mechanism to determine the smoothing factor for each output category, and after training, input the test set into the trained network for prediction verification.

[0087] S44: The training termination condition is that the region localization accuracy on the test set is not less than 95%. If it is not met, the smoothing factor parameter is adjusted or training samples are added and retraining is performed until the termination condition is met.

[0088] Specifically, the adaptive smoothing factor adjustment mechanism in S4 is as follows:

[0089] .

[0090] in, For the first Smoothing factors corresponding to each output category.

[0091] It is the basic smoothing factor, with a value range of 0.1 to 0.3.

[0092] This is an adjustment coefficient, with a value ranging from 0.5 to 1.5.

[0093] For the training set The standard deviation of the feature vectors of each category of samples.

[0094] For the training set The mean of the feature vectors of each category of samples.

[0095] S5: On-site environmental vibration data acquisition and modal parameter identification.

[0096] Specifically, the method for collecting on-site environmental vibration data in S5 is to install environmental vibration sensors on the transmission tower to be tested and collect the acceleration time history response data of the tower under wind load or ground pulsation environmental excitation.

[0097] Specifically, the environmental vibration sensor is a low-frequency piezoelectric accelerometer or MEMS accelerometer, which is installed at the main material nodes of each primary substructure of the tower. The actual data acquisition time is not less than 5 minutes, and the sampling frequency is not less than 200Hz.

[0098] S6: Extract the first four natural frequencies and corresponding mode shapes of the tower from the data collected in S5 using the random subspace recognition method. Calculate the measured frequency change ratio and the norm of the measured modal compliance residual matrix. After normalization, input the results into the probabilistic neural network trained in step S4 to obtain the location results of the secondary substructure area where the main bolts are loose.

[0099] In this embodiment, the positioning method utilizes the acceleration response data of the transmission tower under natural environmental excitations such as wind loads and ground pulsations to identify modal parameters. It eliminates the need for manual tower climbing, active vibration equipment, and bolt-by-bolt installation of contact sensors. Data acquisition can be completed by deploying only a small number of acceleration sensors at key nodes of the main tower structure, achieving non-contact, uninterrupted online monitoring. Furthermore, this positioning method employs a multi-scale hierarchical division strategy of "first-level substructure – second-level substructure." Based on the stress characteristics of the tower legs, tower body, and crossarms, the main structure is divided into 7-15 first-level substructures. Each first-level substructure is further subdivided into 3-5 second-level substructures along the main structure direction according to segment length. This division method considers both the stress hierarchy characteristics of the tower structure and the distribution patterns of bolt connection nodes, ensuring that each second-level substructure has relatively independent and identifiable modal response characteristics. This multi-scale division strategy provides clearer hierarchy and finer classification granularity, effectively balancing positioning accuracy with the classification complexity of the PNN network (probabilistic neural network). Finally, the feature vector constructed by this localization method integrates two complementary damage-sensitive indices: first, the ratio of first-order to fourth-order frequency changes, reflecting the differentiated impact of overall stiffness changes on lower-order frequencies; and second, the Frobenius norm of the modal compliance matrix residuals, reflecting the local variation characteristics of the structural compliance distribution. These two indices characterize the structural stiffness degradation caused by bolt loosening from different dimensions. Weighted fusion effectively reduces the risk of single indices being affected by environmental noise, improving identification robustness. The multi-source fusion strategy of this method exhibits higher identification stability under complex environmental conditions.

[0100] Example 2:

[0101] A system for locating loose bolt areas on the main components of a power transmission tower, used to implement the aforementioned bolt loosening location method based on environmental vibration and multi-scale PNN, the system comprising:

[0102] The data acquisition and transmission module includes environmental vibration sensors, a data acquisition instrument, and a wireless transmission module deployed at the nodes of the main structure of the tower, used to acquire the acceleration response under environmental excitation in real time.

[0103] The simulation sample generation module includes a finite element model unit for the iron tower, a simulation unit for bolt loosening conditions, and a feature extraction unit. It is used to generate a multi-source damage-sensitive feature vector sample library corresponding to the loosening of different substructures.

[0104] The model training and recognition module includes a normalization processing unit, an adaptive smoothing factor probabilistic neural network training unit, and a measured data feature extraction unit. It is used to train the loosened region recognition model and classify and recognize the measured features, and output the location labels of the loosened secondary substructures.

[0105] In this embodiment, a straight-line tower in a 220kV transmission line is taken as the target object. The tower is 45 meters high and its main material is Q345 angle steel, which is bolted together to form the main body of the tower. To achieve rapid and accurate positioning of the loose bolt areas of the main material of the tower, the following system is built and the corresponding method is implemented.

[0106] The data acquisition and transmission module includes sensor deployment and data acquisition and transmission.

[0107] The sensors are deployed at key nodes on the main structural members of the transmission tower (selecting four main transverse diaphragms at heights of 5m, 15m, 25m, and 35m above the ground where they intersect with the main structural members). ICP-type triaxial accelerometers are used (sensitivity 100mV / g, range ±50g). One sensor is installed at each node, for a total of 16 sensors, located at key nodes on the main structural members of the four tower faces (A, B, C, and D). The sensors are securely attached to the surface of the main structural members using magnetic bases or adhesive bonding. The X-axis points outward from the tower body, the Y-axis runs tangentially along the tower body, and the Z-axis points vertically upward.

[0108] The data acquisition and transmission process utilizes an NI CompactDAQ 9185 chassis, paired with an NI9234 four-channel dynamic signal acquisition module. The sampling frequency is set to 2000 Hz, and the anti-aliasing filter cutoff frequency is 800 Hz. Each sensor acquires the acceleration response under environmental excitations (wind vibration, ground pulsation, etc.), with each sampling segment lasting 10 minutes, and six sets of data are continuously collected. The wireless transmission module employs a 4G DTU (Data Transmission Unit) to upload the collected data to the central server in real time. The data is packaged as a compressed binary file, accompanied by a timestamp and node number.

[0109] The simulation sample generation module includes the establishment of finite element models of iron towers, simulation of bolt loosening conditions, and extraction of multi-scale damage-sensitive features.

[0110] The finite element model was established by creating a detailed finite element model of the target tower in ANSYS. Main members, diagonal members, and cross members were all simulated using beam elements, while bolted connections were simulated using a combination of rigid beam elements and spring elements to simulate connection stiffness. The model consisted of 12,845 nodes and 38,212 elements. Material parameters: elastic modulus 206 GPa, Poisson's ratio 0.3, density 7850 kg / m³. The tower was divided into three levels of substructures: the first level consisted of four tower faces (A / B / C / D); the second level consisted of each tower face divided into upper, middle, and lower sections along the height direction (e.g., upper A, middle A, lower A), totaling 12 second-level substructures; the third level consisted of individual bolts or bolt groups within specific tower sections.

[0111] The simulation process for bolt loosening conditions is as follows: the degree of loosening is divided into mild loosening (30% torque loss), moderate loosening (60% torque loss), and severe loosening (90% torque loss) according to the percentage of torque attenuation. In the model, the corresponding spring unit stiffness is reduced to 70%, 40%, and 10% of the original stiffness to simulate this. Loosening conditions are sequentially set for each secondary substructure: each time, only the main connecting bolts within one secondary substructure are allowed to loosen to a certain degree, while the remaining substructures remain intact. Each degree of loosening is repeated 3 times with random phase environmental excitation (wind spectrum excitation applied to the tower base and tower body, with an average wind speed of 4~12 m / s). A total of 12 (secondary substructures) × 3 (degrees of loosening) × 3 (excitation repetitions) = 108 conditions are generated, plus one condition of all intact conditions, for a total of 109 simulation conditions.

[0112] The multi-scale damage-sensitive feature extraction involves extracting the acceleration response of each sensor node under simulated conditions, followed by the following multi-scale feature extraction (each measurement point is considered a sample channel):

[0113] Time-domain characteristics: root mean square, peak value, peak-to-peak value, kurtosis factor, and waveform factor of the acceleration signal.

[0114] Frequency domain characteristics: Perform a fast Fourier transform on the signal and take the amplitude ratio at the first 5 natural frequencies within the 0~200 Hz frequency band (amplitude ratio between loose and intact conditions).

[0115] Time-frequency domain features: Wavelet packets are used for 3-level decomposition to extract the energy proportion of 8 frequency band nodes.

[0116] Transitivity characteristics: Using the tower leg reference node as input and other nodes as output, calculate the difference integral of the frequency response function amplitude from 1 to 100 Hz.

[0117] The model training and recognition module includes normalization processing, adaptive smoothing factor probabilistic neural network (PNN) training, and feature extraction and recognition from measured data. The normalization processing normalizes all feature vectors in the sample database, calculating the mean and standard deviation of each feature dimension to ensure each feature has zero mean and unit variance. The normalization parameters are also saved for use with the measured data. The adaptive smoothing factor probabilistic neural network (PNN) training constructs a multi-class PNN classifier with 304 input layer nodes, 109 pattern layer nodes (equal to the number of training samples), 12 summation layer nodes (corresponding to 12 secondary substructures), and the output layer consisting of competing neurons outputting the class. An adaptive smoothing factor strategy is used for training. After training, the model converges, achieving a 98.2% accuracy rate in reclassifying training samples.

[0118] In this embodiment, on an actual iron tower, the deployed data acquisition and transmission module is used to obtain the acceleration response under environmental vibration (collecting for 10 minutes as well, with a sampling frequency of 2000 Hz). The original signal is preprocessed by removing the trend term and band-pass filtering (0.5 Hz - 150 Hz). According to the same feature extraction process as the simulation samples, a 304-dimensional feature vector is extracted from each measured sensor data, and standardized using the training normalization parameters. Then, this feature vector is input into the trained PNN classifier. The model outputs a 12-dimensional posterior probability vector, and the secondary substructure label corresponding to the maximum value is taken as the loosening area positioning result.

[0119] In this embodiment, the system sends the recognition result to the maintenance personnel's handheld terminal or the remote monitoring platform through the wireless transmission module, and marks the positions of the iron tower and the loosened secondary substructure on the geographic information system map. In this embodiment, the on-site measured data of the same iron tower (it is known that the actual torque of the bolts in section A in the middle has decreased by 55%) is selected for testing. The system outputs the label "A middle", with a probability of 0.83, and the positioning error does not exceed the range of one secondary substructure (about a 5-meter height interval), verifying the effectiveness of the method. At the same time, each recognition result can be sent back to the simulation sample generation module for incremental learning to update the PNN model (adaptive adjustment of the smoothing factor and addition of new samples), improving the accuracy during long-term operation.

[0120] Embodiment 3:

[0121] In this embodiment, a certain cat-head transmission iron tower is taken as the object. It is 36 m high, with a root opening of 8 m. The main members are made of Q345B steel, the cross braces and diagonal members are made of Q235B steel, and the main member bolts are M24 high-strength bolts.

[0122] S1: Establish a refined finite element model of the iron tower and divide it into multi-scale hierarchical substructures.

[0123] A finite element simulation model is established through ANSY simulation software. The core stressed members such as the main members, cross braces, and diagonal members of the tower body are retained, and the non-core stressed members such as the small brackets and grounding wires attached to the iron tower are ignored. While simplifying the model, the simulation accuracy is ensured; the bolt connections are simulated using beam elements, the main members, cross braces, and diagonal members are simulated using shell elements, and quadrilateral meshes are used for mesh division.

[0124] In this embodiment, combined with the iron tower structure level, component connection, and bolt density, the main members are divided into 9 first-level substructures in the height direction, as Figure 2As shown. Each primary substructure is axisymmetric. For each primary substructure, it is divided into four secondary substructures along the main material direction according to the segment length between two flange connection nodes. Each secondary substructure contains four main material angle steel members and their connecting bolt groups in the same height plane. There are a total of 9 × 4 = 36 secondary substructures, and the main materials in each secondary substructure are numbered (numbered 1 to 36).

[0125] S2: Simulate bolt loosening conditions and generate training samples.

[0126] In the finite element simulation model, the bolt loosening condition is simulated by reducing the normal contact stiffness of the bolted connection spring elements. The stiffness reduction range is set to 10%~90% of the original design value, with random values ​​selected to simulate different degrees of loosening. The following conditions are simulated for each of the 36 secondary substructures:

[0127] (a) A single bolt is loose.

[0128] (b) Two to three bolts in the same substructure are loose.

[0129] Each secondary substructure generates no fewer than 50 sets of simulation samples, for a total of approximately 2000 sets of samples.

[0130] S3: Calculate the frequency change ratio and modal compliance residual norm to construct a multi-source damage sensitive feature vector.

[0131] Each sample group under environmental excitation under operational modal analysis, and the first four natural frequencies were extracted. And the vibrational displacement of key nodes in the tower body.

[0132] Calculate the frequency change ratio and modal compliance residual norm According to the formula Construct a multi-source damage-sensitive feature vector, where , All feature vectors are subjected to Min-Max linear normalization and uniformly mapped to... The output label uses 9-dimensional one-hot encoding, with the position corresponding to the loose substructure being 1 and the rest being 0.

[0133] S4: Construct and train an adaptive smoothing factor probabilistic neural network (PNN).

[0134] Construct a four-layer probabilistic neural network, with the number of neurons in the input layer corresponding to... and Two features: the number of radial base layer neurons equals the number of training samples, and the number of neurons in the summation layer is 36 (corresponding to the number of secondary substructures). The output layer uses a competitive mechanism to output the category corresponding to the maximum posterior probability.

[0135] The smoothing factor employs an adaptive adjustment mechanism: ,in Take 0.2, Take 1.0.

[0136] The training and test sets were randomly divided in a 75%:25% ratio. After training, the accuracy of area localization was verified on the test set, reaching 96.3%, meeting the preset 95% threshold requirement.

[0137] S5: On-site environmental vibration data acquisition and modal parameter identification.

[0138] Low-frequency piezoelectric accelerometers (range ±2g, frequency response range 0.1~100Hz) were installed at the main material nodes of each primary substructure of the transmission tower to be tested, for a total of 12 measuring points. Acceleration time history response data of the tower under wind load were collected, with a sampling frequency of 512Hz and a sampling duration of 10 minutes.

[0139] The random subspace identification method was used to identify modal parameters from the collected data, extracting the first four natural frequencies and corresponding mode shapes of the tower. The measured frequency variation ratio and modal compliance residual norm were calculated and normalized to form a measured multi-source damage sensitive feature vector.

[0140] S6: PNN online recognition outputs the location results of loose areas.

[0141] The measured feature vectors are input into the trained probabilistic neural network, which outputs the posterior probability values ​​for each of the 36 secondary substructures. The secondary substructure with the highest posterior probability is taken as the location result of the loose bolt area. In this embodiment, the network identifies the 15th secondary substructure (located in the middle of the tower) as a loose area with a posterior probability of 0.94. Based on this, maintenance personnel manually re-inspect the four main bolts in this area, confirming that two bolts are loose, and complete the maintenance and tightening work.

[0142] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

[0143] Many other changes and modifications can be made without departing from the concept and scope of this invention. It should be understood that this invention is not limited to the specific embodiments, and the scope of this invention is defined by the appended claims.

Claims

1. A bolt loosening location method based on environmental vibration and multi-scale PNN, characterized in that, Includes the following steps: S1: Establish a refined finite element simulation model of the power transmission tower and divide it into multi-scale hierarchical substructures; S2: In the finite element simulation model generated in S1, the bolt loosening condition is simulated, and the operational modal analysis results of the simulation model under environmental excitation are extracted; S3: Based on the results of the operational modal analysis, calculate the frequency change ratio between the first and fourth orders and the norm of the modal compliance residual matrix, and construct a multi-source damage sensitive feature vector after weighted fusion; S4: Construct and train a probabilistic neural network with an adaptive smoothing factor adjustment mechanism; S5: On-site environmental vibration data acquisition and modal parameter identification; S6: Extract the first four natural frequencies and corresponding mode shapes of the tower from the data collected in S5 using the random subspace recognition method. Calculate the measured frequency change ratio and the norm of the measured modal compliance residual matrix. After normalization, input the results into the probabilistic neural network trained in step S4 to obtain the location results of the secondary substructure area where the main bolts are loose.

2. The bolt loosening location method based on environmental vibration and multi-scale PNN according to claim 1, characterized in that, The principle of multi-scale hierarchical substructure partitioning in S1 is as follows: Based on the stress hierarchy of the tower structure and the distribution characteristics of the main bolts, the main material of the tower is divided into multi-scale hierarchical substructures, forming N primary substructures. Each primary substructure is further divided into M secondary substructures, and a unique mode category label is assigned to each secondary substructure. The specific process of multi-scale hierarchical substructure partitioning in S1 is as follows: S11: Based on the stress characteristics of the tower legs, tower body, and crossarms, the main materials are divided into 7 to 15 primary substructures, each of which has a high degree of axial symmetry. S12: For each primary substructure, it is divided into 3 to 5 secondary substructures along the main material direction according to the segment length between two flange connection nodes. Each secondary substructure contains all main material angle steel components and their connecting bolt groups in the same height plane.

3. The bolt loosening location method based on environmental vibration and multi-scale PNN according to claim 1, characterized in that, The formula for constructing the multi-source damage-sensitive feature vector D in S3 is as follows: ； in, , and These represent the first and fourth natural frequencies, respectively, with superscripts. Indicates the state of damage, superscript Indicates a good condition; , representing the good state compliance matrix With the damage state compliance matrix The Frobenius norm of the difference, the compliance matrix is ​​constructed from the first four measured or simulated frequencies and mass-normalized mode shapes, and For the weighting coefficients, satisfying ,and The value range is 0.4 to 0.

6.

4. The bolt loosening location method based on environmental vibration and multi-scale PNN according to claim 1, characterized in that, The construction and training process of the probabilistic neural network in S4 is as follows: S41: The multi-source damage-sensitive feature vectors obtained in S3 are normalized and divided into a training set and a test set according to a preset ratio, with the training set accounting for 70%~80% and the test set accounting for 20%~30%. S42: Construct a probabilistic neural network, which includes an input layer, a radial base layer, a summation layer and an output layer. The input layer receives the multi-source damage-sensitive feature vector, and the output layer outputs the posterior probability of loosening of each secondary substructure. The secondary substructure corresponding to the largest posterior probability is taken as the preliminary localization result of the loosening region. S43: Substitute the training set into the probabilistic neural network for training, use the adaptive smoothing factor adjustment mechanism to determine the smoothing factor for each output category, and after training, input the test set into the trained network for prediction verification. S44: The training termination condition is that the region localization accuracy on the test set is not less than 95%. If it is not met, the smoothing factor parameter is adjusted or training samples are added and retraining is performed until the termination condition is met.

5. The bolt loosening location method based on environmental vibration and multi-scale PNN according to claim 1, characterized in that, The adaptive smoothing factor adjustment mechanism in S4 is as follows: ； in, For the first Smoothing factor corresponding to each output category; The basic smoothing factor has a value range of 0.1 to 0.

3. This is an adjustment coefficient, with a value ranging from 0.5 to 1.

5. For the training set The standard deviation of the feature vectors of each category of samples; For the training set The mean of the feature vectors of each category of samples.

6. The bolt loosening location method based on environmental vibration and multi-scale PNN according to claim 1, characterized in that, In S2, the stiffness of the bolted joint connection is reduced to simulate different bolt loosening conditions. The specific simulation method is as follows: In the finite element model, the bolted connection is equivalent to a spring-damped element with stiffness in six directions. By randomly selecting values ​​within the range of 10% to 90% of the original design value to reduce the normal contact stiffness, different degrees of bolt loosening are simulated. At least 50 sets of simulation samples were generated for each secondary substructure, including single bolt loosening, multi-bolt combination loosening, and different degrees of loosening.

7. The bolt loosening location method based on environmental vibration and multi-scale PNN according to claim 1, characterized in that, The method for acquiring on-site environmental vibration data in S5 is to install environmental vibration sensors on the transmission tower to be tested and collect the acceleration time history response data of the tower under wind load or ground pulsation environmental excitation.

8. The bolt loosening location method based on environmental vibration and multi-scale PNN according to claim 7, characterized in that, The environmental vibration sensor is a low-frequency piezoelectric accelerometer or MEMS accelerometer, which is installed at the main material nodes of each primary substructure of the tower. The actual data acquisition time is not less than 5 minutes and the sampling frequency is not less than 200Hz.

9. The bolt loosening location method based on environmental vibration and multi-scale PNN according to claim 1, characterized in that, The operational modal analysis results in S3 include the first four natural frequencies and the mode displacements at key nodes of the tower.

10. A positioning system for loose bolt areas on the main components of a power transmission tower, characterized in that, For implementing the bolt loosening location method based on environmental vibration and multi-scale PNN as described in any one of claims 1-9, the system comprises: The data acquisition and transmission module includes an environmental vibration sensor, a data acquisition instrument, and a wireless transmission module deployed at the nodes of the main material of the tower, which are used to acquire the acceleration response under environmental excitation in real time. The simulation sample generation module includes a finite element model unit for iron towers, a simulation unit for bolt loosening conditions, and a feature extraction unit, which is used to generate a multi-source damage-sensitive feature vector sample library corresponding to the loosening of different substructures. The model training and recognition module includes a normalization processing unit, an adaptive smoothing factor probabilistic neural network training unit, and a measured data feature extraction unit. It is used to train the loosened region recognition model and classify and recognize the measured features, and output the location labels of the loosened secondary substructures.

Images